app/vmselect/promql: add {topk|bottomk}_{min|max|avg|median} aggregate functions for returning the exact k time series on the given time range

The full list of functions added:
- `topk_min(k, q)` - returns top K time series with the max minimums on the given time range
- `topk_max(k, q)` - returns top K time series with the max maximums on the given time range
- `topk_avg(k, q)` - returns top K time series with the max averages on the given time range
- `topk_median(k, q)` - returns top K time series with the max medians on the given time range
- `bottomk_min(k, q)` - returns bottom K time series with the min minimums on the given time range
- `bottomk_max(k, q)` - returns bottom K time series with the min maximums on the given time range
- `bottomk_avg(k, q)` - returns bottom K time series with the min averages on the given time range
- `bottomk_median(k, q)` - returns bottom K time series with the min medians on the given time range

.github/workflows/github-pages.yml

@@ -0,0 +1,30 @@
+name: github-pages
+on:
+  push:
+    paths:
+      - 'docs/*.md'
+      - 'README.md'
+    branches:
+      - master
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@master
+      - name: publish
+        shell: bash
+        env:
+          TOKEN: ${{secrets.CI_TOKEN}}
+        run: |
+          git clone https://vika:${TOKEN}@github.com/VictoriaMetrics/VictoriaMetrics.github.io.git gpages
+          cp docs/*.md gpages
+          cp README.md gpages
+          cd gpages
+          git config --local user.email "info@victoriametrics.com"
+          git config --local user.name "Vika"
+          git add "*.md"
+          git commit -m "update github pages"
+          remote_repo="https://vika:${TOKEN}@github.com/VictoriaMetrics/VictoriaMetrics.github.io.git"
+          git push "${remote_repo}"
+          cd ..
+          rm -rf gpages

.github/workflows/main.yml

@@ -1,6 +1,13 @@
 name: main
 on:
-  - push
+  push:
+    paths-ignore:
+      - 'docs/**'
+      - '**.md'
+  pull_request:
+    paths-ignore:
+      - 'docs/**'
+      - '**.md'
 jobs:
   build:
     name: Build
@@ -9,7 +16,7 @@ jobs:
       - name: Setup Go
         uses: actions/setup-go@v1
         with:
-          go-version: 1.12
+          go-version: 1.13
         id: go
       - name: Code checkout
         uses: actions/checkout@v1
@@ -23,19 +30,21 @@ jobs:
         env:
           GO111MODULE: on
         run: |
-            export PATH=$PATH:$(go env GOPATH)/bin # temporary fix. See https://github.com/actions/setup-go/issues/14
-            make check-all
-            git diff --exit-code
-            make test-full
-            make test-pure
-            make victoria-metrics
-            make victoria-metrics-pure
-            make victoria-metrics-arm
-            make victoria-metrics-arm64
-            GOOS=freebsd go build -mod=vendor ./app/victoria-metrics
-            GOOS=darwin go build -mod=vendor ./app/victoria-metrics
+          export PATH=$PATH:$(go env GOPATH)/bin # temporary fix. See https://github.com/actions/setup-go/issues/14
+          make check-all
+          git diff --exit-code
+          make test-full
+          make test-pure
+          make test-full-386
+          make victoria-metrics
+          make victoria-metrics-pure
+          make victoria-metrics-arm
+          make victoria-metrics-arm64
+          make vmutils
+          GOOS=freebsd go build -mod=vendor ./app/victoria-metrics
+          GOOS=darwin go build -mod=vendor ./app/victoria-metrics
       - name: Publish coverage
-        uses: codecov/codecov-action@v1.0.0
+        uses: codecov/codecov-action@v1.0.4
         with:
           token: ${{secrets.CODECOV_TOKEN}}
           file: ./coverage.txt

.github/workflows/wiki.yml

@@ -0,0 +1,29 @@
+name: wiki
+on:
+  push:
+    paths:
+      - 'docs/*.md'
+    branches:
+      - master
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@master
+      - name: publish
+        shell: bash
+        env:
+          TOKEN: ${{secrets.CI_TOKEN}}
+        run: |
+          cd docs
+          git clone https://vika:${TOKEN}@github.com/VictoriaMetrics/VictoriaMetrics.wiki.git wiki
+          find ./ -name '*.md' -exec cp -prv '{}' 'wiki' ';'
+          cd wiki
+          git config --local user.email "info@victoriametrics.com"
+          git config --local user.name "Vika"
+          git add "*.md"
+          git commit -m "update wiki pages"
+          remote_repo="https://vika:${TOKEN}@github.com/VictoriaMetrics/VictoriaMetrics.wiki.git"
+          git push "${remote_repo}"
+          cd ..
+          rm -rf wiki

.gitignore

@@ -1,3 +1,4 @@
+/tmp
 /tags
 /pkg
 *.pprof

Makefile

@@ -19,12 +19,36 @@ include deployment/*/Makefile
 clean:
 	rm -rf bin/*
 
-publish: publish-victoria-metrics
+publish: \
+	publish-victoria-metrics \
+	publish-vmbackup \
+	publish-vmrestore
 
-package: package-victoria-metrics
+package: \
+	package-victoria-metrics \
+	package-vmbackup \
+	package-vmrestore
 
-release: victoria-metrics-prod
-	cd bin && tar czf victoria-metrics-$(PKG_TAG).tar.gz victoria-metrics-prod
+vmutils: \
+	vmbackup \
+	vmrestore
+
+release: \
+	release-victoria-metrics \
+	release-vmutils
+
+release-victoria-metrics: victoria-metrics-prod
+	cd bin && tar czf victoria-metrics-$(PKG_TAG).tar.gz victoria-metrics-prod && \
+		sha256sum victoria-metrics-$(PKG_TAG).tar.gz > victoria-metrics-$(PKG_TAG)_checksums.txt
+
+release-vmutils: \
+	vmbackup-prod \
+	vmrestore-prod
+	cd bin && tar czf vmutils-$(PKG_TAG).tar.gz vmbackup-prod vmrestore-prod && \
+		sha256sum vmutils-$(PKG_TAG).tar.gz > vmutils-$(PKG_TAG)_checksums.txt
+
+pprof-cpu:
+	go tool pprof -trim_path=github.com/VictoriaMetrics/VictoriaMetrics@ $(PPROF_FILE)
 
 fmt:
 	GO111MODULE=on gofmt -l -w -s ./lib
@@ -39,13 +63,15 @@ lint: install-golint
 	golint app/...
 
 install-golint:
-	which golint || GO111MODULE=off go get -u github.com/golang/lint/golint
+	which golint || GO111MODULE=off go get -u golang.org/x/lint/golint
 
 errcheck: install-errcheck
 	errcheck -exclude=errcheck_excludes.txt ./lib/...
 	errcheck -exclude=errcheck_excludes.txt ./app/vminsert/...
 	errcheck -exclude=errcheck_excludes.txt ./app/vmselect/...
 	errcheck -exclude=errcheck_excludes.txt ./app/vmstorage/...
+	errcheck -exclude=errcheck_excludes.txt ./app/vmbackup/...
+	errcheck -exclude=errcheck_excludes.txt ./app/vmrestore/...
 
 install-errcheck:
 	which errcheck || GO111MODULE=off go get -u github.com/kisielk/errcheck
@@ -61,6 +87,9 @@ test-pure:
 test-full:
 	GO111MODULE=on go test -tags=integration -mod=vendor -coverprofile=coverage.txt -covermode=atomic ./lib/... ./app/...
 
+test-full-386:
+	GO111MODULE=on GOARCH=386 go test -tags=integration -mod=vendor -coverprofile=coverage.txt -covermode=atomic ./lib/... ./app/...
+
 benchmark:
 	GO111MODULE=on go test -mod=vendor -bench=. ./lib/...
 	GO111MODULE=on go test -mod=vendor -bench=. ./app/...
@@ -89,7 +118,7 @@ install-qtc:
 
 
 golangci-lint: install-golangci-lint
-	golangci-lint run --exclude '(SA4003|SA1019):' -D errcheck
+	golangci-lint run --exclude '(SA4003|SA1019):' -D errcheck -D structcheck
 
 install-golangci-lint:
 	which golangci-lint || GO111MODULE=off go get -u github.com/golangci/golangci-lint/cmd/golangci-lint

README.md

@@ -26,6 +26,7 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
   and [selects](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4).
   [Outperforms InfluxDB and TimescaleDB by up to 20x](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
 * [Uses 10x less RAM than InfluxDB](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) when working with millions of unique time series (aka high cardinality).
+* Optimized for time series with high churn rate. Think about [prometheus-operator](https://github.com/coreos/prometheus-operator) metrics from frequent deployments in Kubernetes.
 * High data compression, so [up to 70x more data points](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4)
   may be crammed into limited storage comparing to TimescaleDB.
 * Optimized for storage with high-latency IO and low IOPS (HDD and network storage in AWS, Google Cloud, Microsoft Azure, etc). See [graphs from these benchmarks](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b).
@@ -33,11 +34,13 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
   See [vertical scalability benchmarks](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
   and [comparing Thanos to VictoriaMetrics cluster](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683).
 * Easy operation:
-  * VictoriaMetrics consists of a single executable without external dependencies.
+  * VictoriaMetrics consists of a single [small executable](https://medium.com/@valyala/stripping-dependency-bloat-in-victoriametrics-docker-image-983fb5912b0d) without external dependencies.
   * All the configuration is done via explicit command-line flags with reasonable defaults.
   * All the data is stored in a single directory pointed by `-storageDataPath` flag.
-  * Easy backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
-* Storage is protected from corruption on unclean shutdown (i.e. hardware reset or `kill -9`) thanks to [the storage architecture](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
+  * Easy and fast backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
+  to S3 or GCS with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md) / [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md).
+  See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+* Storage is protected from corruption on unclean shutdown (i.e. OOM, hardware reset or `kill -9`) thanks to [the storage architecture](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
 * Supports metrics' ingestion and [backfilling](#backfilling) via the following protocols:
   * [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
   * [InfluxDB line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/)
@@ -45,7 +48,7 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
     if `-graphiteListenAddr` is set.
   * [OpenTSDB put message](http://opentsdb.net/docs/build/html/api_telnet/put.html) if `-opentsdbListenAddr` is set.
   * [HTTP OpenTSDB /api/put requests](http://opentsdb.net/docs/build/html/api_http/put.html) if `-opentsdbHTTPListenAddr` is set.
-* Ideally works with big amounts of time series data from Kubernetes, IoT sensors, connected cars, industrial telemetry and various Enterprise workloads.
+* Ideally works with big amounts of time series data from Kubernetes, IoT sensors, connected cars, industrial telemetry, financial data and various Enterprise workloads.
 * Has open source [cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
 
 
@@ -89,6 +92,7 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
   - [Troubleshooting](#troubleshooting)
   - [Backfilling](#backfilling)
   - [Profiling](#profiling)
+- [Integrations](#integrations)
 - [Roadmap](#roadmap)
 - [Contacts](#contacts)
 - [Community and contributions](#community-and-contributions)
@@ -107,8 +111,8 @@ or [docker image](https://hub.docker.com/r/victoriametrics/victoria-metrics/) wi
 
 The following command-line flags are used the most:
 
-* `-storageDataPath` - path to data directory. VictoriaMetrics stores all the data in this directory.
-* `-retentionPeriod` - retention period in months for the data. Older data is automatically deleted.
+* `-storageDataPath` - path to data directory. VictoriaMetrics stores all the data in this directory. Default path is `victoria-metrics-data` in current working directory.
+* `-retentionPeriod` - retention period in months for the data. Older data is automatically deleted. Default period is 1 month.
 * `-httpListenAddr` - TCP address to listen to for http requests. By default, it listens port `8428` on all the network interfaces.
 * `-graphiteListenAddr` - TCP and UDP address to listen to for Graphite data. By default, it is disabled.
 * `-opentsdbListenAddr` - TCP and UDP address to listen to for OpenTSDB data over telnet protocol. By default, it is disabled.
@@ -156,7 +160,7 @@ The label name may be arbitrary - `datacenter` is just an example. The label val
 across Prometheus instances, so those time series may be filtered and grouped by this label.
 
 
-It is recommended upgrading Prometheus to [v2.10.0](https://github.com/prometheus/prometheus/releases) or newer,
+It is recommended upgrading Prometheus to [v2.12.0](https://github.com/prometheus/prometheus/releases) or newer,
 since the previous versions may have issues with `remote_write`.
 
 
@@ -171,7 +175,7 @@ http://<victoriametrics-addr>:8428
 Substitute `<victoriametrics-addr>` with the hostname or IP address of VictoriaMetrics.
 
 Then build graphs with the created datasource using [Prometheus query language](https://prometheus.io/docs/prometheus/latest/querying/basics/).
-VictoriaMetrics supports native PromQL and [extends it with useful features](ExtendedPromQL).
+VictoriaMetrics supports native PromQL and [extends it with useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
 
 
 ### How to upgrade VictoriaMetrics?
@@ -219,7 +223,7 @@ VictoriaMetrics maps Influx data using the following rules:
   unless `db` tag exists in the Influx line.
 * Field names are mapped to time series names prefixed with `{measurement}{separator}` value,
   where `{separator}` equals to `_` by default. It can be changed with `-influxMeasurementFieldSeparator` command-line flag.
-  See also `-influxSkipSingleField` command-line flag.
+  See also `-influxSkipSingleField` command-line flag. If `{measurement}` is empty, then time series names correspond to field names.
 * Field values are mapped to time series values.
 * Tags are mapped to Prometheus labels as-is.
 
@@ -247,14 +251,14 @@ An arbitrary number of lines delimited by '\n' may be sent in a single request.
 After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
 
 ```
-curl -G 'http://localhost:8428/api/v1/export' -d 'match={__name__!=""}'
+curl -G 'http://localhost:8428/api/v1/export' -d 'match={__name__=~"measurement_.*"}'
 ```
 
 The `/api/v1/export` endpoint should return the following response:
 
 ```
-{"metric":{"__name__":"measurement.field1","tag1":"value1","tag2":"value2"},"values":[123],"timestamps":[1560272508147]}
-{"metric":{"__name__":"measurement.field2","tag1":"value1","tag2":"value2"},"values":[1.23],"timestamps":[1560272508147]}
+{"metric":{"__name__":"measurement_field1","tag1":"value1","tag2":"value2"},"values":[123],"timestamps":[1560272508147]}
+{"metric":{"__name__":"measurement_field2","tag1":"value1","tag2":"value2"},"values":[1.23],"timestamps":[1560272508147]}
 ```
 
 Note that Influx line protocol expects [timestamps in *nanoseconds* by default](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/#timestamp),
@@ -285,7 +289,7 @@ An arbitrary number of lines delimited by `\n` may be sent in one go.
 After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
 
 ```
-curl -G 'http://localhost:8428/api/v1/export' -d 'match={__name__!=""}'
+curl -G 'http://localhost:8428/api/v1/export' -d 'match=foo.bar.baz'
 ```
 
 The `/api/v1/export` endpoint should return the following response:
@@ -330,7 +334,7 @@ An arbitrary number of lines delimited by `\n` may be sent in one go.
 After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
 
 ```
-curl -G 'http://localhost:8428/api/v1/export' -d 'match={__name__!=""}'
+curl -G 'http://localhost:8428/api/v1/export' -d 'match=foo.bar.baz'
 ```
 
 The `/api/v1/export` endpoint should return the following response:
@@ -460,8 +464,8 @@ The page will return the following JSON response:
 ```
 
 Snapshots are created under `<-storageDataPath>/snapshots` directory, where `<-storageDataPath>`
-is the command-line flag value. Snapshots can be archived to backup storage via `cp -L`, `rsync -L`, `scp -r`
-or any similar tool that follows symlinks during copying.
+is the command-line flag value. Snapshots can be archived to backup storage at any time
+with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md).
 
 The `http://<victoriametrics-addr>:8428/snapshot/list` page contains the list of available snapshots.
 
@@ -472,9 +476,9 @@ Navigate to `http://<victoriametrics-addr>:8428/snapshot/delete_all` in order to
 
 Steps for restoring from a snapshot:
 1. Stop VictoriaMetrics with `kill -INT`.
-2. Remove the entire contents of the directory pointed by `-storageDataPath` command-line flag.
-3. Copy snapshot contents to the directory pointed by `-storageDataPath`.
-4. Start VictoriaMetrics.
+2. Restore snapshot contents from backup with [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md)
+   to the directory pointed by `-storageDataPath`.
+3. Start VictoriaMetrics.
 
 
 ### How to delete time series?
@@ -511,7 +515,7 @@ at `http://<victoriametrics-addr>:8428/federate?match[]=<timeseries_selector_for
 
 Optional `start` and `end` args may be added to the request in order to scrape the last point for each selected time series on the `[start ... end]` interval.
 `start` and `end` may contain either unix timestamp in seconds or [RFC3339](https://www.ietf.org/rfc/rfc3339.txt) values. By default, the last point
-on the interval `[now - max_lookback ... now]` is scraped for each time series. The default value for `max_lookback` is `5m` (5 minutes), but can be overridden.
+on the interval `[now - max_lookback ... now]` is scraped for each time series. The default value for `max_lookback` is `5m` (5 minutes), but it can be overridden.
 For instance, `/federate?match[]=up&max_lookback=1h` would return last points on the `[now - 1h ... now]` interval. This may be useful for time series federation
 with scrape intervals exceeding `5m`.
 
@@ -527,7 +531,7 @@ A rough estimation of the required resources for ingestion path:
   VictoriaMetrics stores various caches in RAM. Memory size for these caches may be limited by `-memory.allowedPercent` flag.
 
 * CPU cores: a CPU core per 300K inserted data points per second. So, ~4 CPU cores are required for processing
-  the insert stream of 1M data points per second. The ingestion rate may be lower for high cardinality data.
+  the insert stream of 1M data points per second. The ingestion rate may be lower for high cardinality data or for time series with high number of labels.
   See [this article](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) for details.
   If you see lower numbers per CPU core, then it is likely active time series info doesn't fit caches,
   so you need more RAM for lowering CPU usage.
@@ -652,6 +656,14 @@ For example, substitute `-graphiteListenAddr=:2003` with `-graphiteListenAddr=<i
 * There is no need in Operating System tuning since VictoriaMetrics is optimized for default OS settings.
   The only option is increasing the limit on [the number of open files in the OS](https://medium.com/@muhammadtriwibowo/set-permanently-ulimit-n-open-files-in-ubuntu-4d61064429a),
   so Prometheus instances could establish more connections to VictoriaMetrics.
+* The recommended filesystem is `ext4`, the recommended persistent storage is [persistent HDD-based disk on GCP](https://cloud.google.com/compute/docs/disks/#pdspecs),
+  since it is protected from hardware failures via internal replication and it can be [resized on the fly](https://cloud.google.com/compute/docs/disks/add-persistent-disk#resize_pd).
+  If you plan storing more than 1TB of data on `ext4` partition or plan extending it to more than 16TB,
+  then the following options are recommended to pass to `mkfs.ext4`:
+
+```
+mkfs.ext4 ... -O 64bit,huge_file,extent -T huge
+```
 
 
 ### Monitoring
@@ -664,10 +676,8 @@ The most interesting metrics are:
 
 * `vm_cache_entries{type="storage/hour_metric_ids"}` - the number of time series with new data points during the last hour
   aka active time series.
-* `vm_rows{type="indexdb"}` - the number of rows in inverted index. Each label in each unique time series adds a single
-  row into the inverted index. An approximate number of time series in the database may be calculated as
-  `vm_rows{type="indexdb"} / (avg_labels_per_series + 1)`, where `avg_labels_per_series` is the average number of labels
-  per each time series.
+* `rate(vm_new_timeseries_created_total[5m])` - time series churn rate.
+* `vm_rows{type="indexdb"}` - the number of rows in inverted index. High value for this number usually mean high churn rate for time series.
 * Sum of `vm_rows{type="storage/big"}` and `vm_rows{type="storage/small"}` - total number of `(timestamp, value)` data points
   in the database.
 * Sum of all the `vm_cache_size_bytes` metrics - the total size of all the caches in the database.
@@ -678,6 +688,9 @@ The most interesting metrics are:
 
 ### Troubleshooting
 
+* It is recommended to use default command-line flag values (i.e. don't set them explicitly) until the need
+  in tweaking these flag values arises.
+
 * If VictoriaMetrics works slowly and eats more than a CPU core per 100K ingested data points per second,
   then it is likely you have too many active time series for the current amount of RAM.
   It is recommended increasing the amount of RAM on the node with VictoriaMetrics in order to improve
@@ -723,6 +736,15 @@ The command for collecting CPU profile waits for 30 seconds before returning.
 The collected profiles may be analyzed with [go tool pprof](https://github.com/google/pprof).
 
 
+## Integrations
+
+* [netdata](https://github.com/netdata/netdata) can push data into VictoriaMetrics via `Prometheus remote_write API`.
+  See [these docs](https://github.com/netdata/netdata#integrations).
+* [go-graphite/carbonapi](https://github.com/go-graphite/carbonapi) can use VictoriaMetrics as time series backend.
+  See [this example](/blob/master/cmd/carbonapi/carbonapi.example.prometheus.yaml).
+* [Ansible role for installing VictoriaMetrics](https://github.com/dreamteam-gg/ansible-victoriametrics-role).
+
+
 ## Roadmap
 
 - [ ] Replication [#118](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/118)
@@ -745,8 +767,8 @@ Contact us with any questions regarding VictoriaMetrics at [info@victoriametrics
 Feel free asking any questions regarding VictoriaMetrics:
 
 - [slack](http://slack.victoriametrics.com/)
-- [telergam-en](https://t.me/VictoriaMetrics_en)
-- [telergam-ru](https://t.me/VictoriaMetrics_ru1)
+- [telegram-en](https://t.me/VictoriaMetrics_en)
+- [telegram-ru](https://t.me/VictoriaMetrics_ru1)
 - [google groups](https://groups.google.com/forum/#!forum/victorametrics-users)
 
 

app/victoria-metrics/Makefile

@@ -32,6 +32,18 @@ victoria-metrics-arm64:
 victoria-metrics-arm64-prod:
 	APP_NAME=victoria-metrics APP_SUFFIX='-arm64' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=arm64' $(MAKE) app-via-docker
 
+victoria-metrics-ppc64le:
+	CGO_ENABLED=0 GOOS=linux GOARCH=ppc64le GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/victoria-metrics-ppc64le ./app/victoria-metrics
+
+victoria-metrics-ppc64le-prod:
+	APP_NAME=victoria-metrics APP_SUFFIX='-ppc64le' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=ppc64le' $(MAKE) app-via-docker
+
+victoria-metrics-386:
+	CGO_ENABLED=0 GOOS=linux GOARCH=386 GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/victoria-metrics-386 ./app/victoria-metrics
+
+victoria-metrics-386-prod:
+	APP_NAME=victoria-metrics APP_SUFFIX='-386' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=386' $(MAKE) app-via-docker
+
 victoria-metrics-pure:
 	APP_NAME=victoria-metrics $(MAKE) app-local-pure
 

app/victoria-metrics/deployment/Dockerfile

@@ -1,5 +1,5 @@
 FROM scratch
-COPY --from=local/certs:1.0.2 /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
+COPY --from=local/certs:1.0.3 /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
 COPY bin/victoria-metrics-prod .
 EXPOSE 8428
 ENTRYPOINT ["/victoria-metrics-prod"]

app/victoria-metrics/main_test.go

@@ -7,6 +7,7 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
 	"io/ioutil"
 	"log"
 	"net"
@@ -18,6 +19,7 @@ import (
 	"testing"
 	"time"
 
+	testutil "github.com/VictoriaMetrics/VictoriaMetrics/app/victoria-metrics/test"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vminsert"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmstorage"
@@ -27,18 +29,21 @@ import (
 )
 
 const (
-	testFixturesDir        = "testdata"
-	testStorageSuffix      = "vm-test-storage"
-	testHTTPListenAddr     = ":7654"
-	testStatsDListenAddr   = ":2003"
-	testOpenTSDBListenAddr = ":4242"
-	testLogLevel           = "INFO"
+	testFixturesDir            = "testdata"
+	testStorageSuffix          = "vm-test-storage"
+	testHTTPListenAddr         = ":7654"
+	testStatsDListenAddr       = ":2003"
+	testOpenTSDBListenAddr     = ":4242"
+	testOpenTSDBHTTPListenAddr = ":4243"
+	testLogLevel               = "INFO"
 )
 
 const (
-	testReadHTTPPath   = "http://127.0.0.1" + testHTTPListenAddr
-	testWriteHTTPPath  = "http://127.0.0.1" + testHTTPListenAddr + "/write"
-	testHealthHTTPPath = "http://127.0.0.1" + testHTTPListenAddr + "/health"
+	testReadHTTPPath          = "http://127.0.0.1" + testHTTPListenAddr
+	testWriteHTTPPath         = "http://127.0.0.1" + testHTTPListenAddr + "/write"
+	testOpenTSDBWriteHTTPPath = "http://127.0.0.1" + testOpenTSDBHTTPListenAddr + "/api/put"
+	testPromWriteHTTPPath     = "http://127.0.0.1" + testHTTPListenAddr + "/api/v1/write"
+	testHealthHTTPPath        = "http://127.0.0.1" + testHTTPListenAddr + "/health"
 )
 
 const (
@@ -51,18 +56,69 @@ var (
 )
 
 type test struct {
-	Name   string `json:"name"`
-	Data   string `json:"data"`
-	Query  string `json:"query"`
-	Result []Row  `json:"result"`
+	Name             string     `json:"name"`
+	Data             []string   `json:"data"`
+	Query            []string   `json:"query"`
+	ResultMetrics    []Metric   `json:"result_metrics"`
+	ResultSeries     Series     `json:"result_series"`
+	ResultQuery      Query      `json:"result_query"`
+	ResultQueryRange QueryRange `json:"result_query_range"`
+	Issue            string     `json:"issue"`
 }
 
-type Row struct {
+type Metric struct {
 	Metric     map[string]string `json:"metric"`
 	Values     []float64         `json:"values"`
 	Timestamps []int64           `json:"timestamps"`
 }
 
+func (r *Metric) UnmarshalJSON(b []byte) error {
+	type plain Metric
+	return json.Unmarshal(testutil.PopulateTimeTpl(b, insertionTime), (*plain)(r))
+}
+
+type Series struct {
+	Status string              `json:"status"`
+	Data   []map[string]string `json:"data"`
+}
+type Query struct {
+	Status string    `json:"status"`
+	Data   QueryData `json:"data"`
+}
+type QueryData struct {
+	ResultType string            `json:"resultType"`
+	Result     []QueryDataResult `json:"result"`
+}
+
+type QueryDataResult struct {
+	Metric map[string]string `json:"metric"`
+	Value  []interface{}     `json:"value"`
+}
+
+func (r *QueryDataResult) UnmarshalJSON(b []byte) error {
+	type plain QueryDataResult
+	return json.Unmarshal(testutil.PopulateTimeTpl(b, insertionTime), (*plain)(r))
+}
+
+type QueryRange struct {
+	Status string         `json:"status"`
+	Data   QueryRangeData `json:"data"`
+}
+type QueryRangeData struct {
+	ResultType string                 `json:"resultType"`
+	Result     []QueryRangeDataResult `json:"result"`
+}
+
+type QueryRangeDataResult struct {
+	Metric map[string]string `json:"metric"`
+	Values [][]interface{}   `json:"values"`
+}
+
+func (r *QueryRangeDataResult) UnmarshalJSON(b []byte) error {
+	type plain QueryRangeDataResult
+	return json.Unmarshal(testutil.PopulateTimeTpl(b, insertionTime), (*plain)(r))
+}
+
 func TestMain(m *testing.M) {
 	setUp()
 	code := m.Run()
@@ -102,6 +158,7 @@ func processFlags() {
 		{flag: "graphiteListenAddr", value: testStatsDListenAddr},
 		{flag: "opentsdbListenAddr", value: testOpenTSDBListenAddr},
 		{flag: "loggerLevel", value: testLogLevel},
+		{flag: "opentsdbHTTPListenAddr", value: testOpenTSDBHTTPListenAddr},
 	} {
 		// panics if flag doesn't exist
 		if err := flag.Lookup(fv.flag).Value.Set(fv.value); err != nil {
@@ -123,67 +180,124 @@ func waitFor(timeout time.Duration, f func() bool) error {
 
 func tearDown() {
 	if err := httpserver.Stop(*httpListenAddr); err != nil {
-		log.Fatalf("cannot stop the webservice: %s", err)
+		log.Printf("cannot stop the webservice: %s", err)
 	}
 	vminsert.Stop()
 	vmstorage.Stop()
 	vmselect.Stop()
 	fs.MustRemoveAll(storagePath)
-	fs.MustStopDirRemover()
 }
 
 func TestWriteRead(t *testing.T) {
 	t.Run("write", testWrite)
 	time.Sleep(1 * time.Second)
 	vmstorage.Stop()
-
 	// open storage after stop in write
 	vmstorage.InitWithoutMetrics()
 	t.Run("read", testRead)
 }
 
 func testWrite(t *testing.T) {
+	t.Run("prometheus", func(t *testing.T) {
+		for _, test := range readIn("prometheus", t, insertionTime) {
+			s := newSuite(t)
+			r := testutil.WriteRequest{}
+			s.noError(json.Unmarshal([]byte(strings.Join(test.Data, "\n")), &r.Timeseries))
+			data, err := testutil.Compress(r)
+			s.greaterThan(len(r.Timeseries), 0)
+			if err != nil {
+				t.Errorf("error compressing %v %s", r, err)
+				t.Fail()
+			}
+			httpWrite(t, testPromWriteHTTPPath, bytes.NewBuffer(data))
+		}
+	})
+
 	t.Run("influxdb", func(t *testing.T) {
-		for _, test := range readIn("influxdb", t, fmt.Sprintf("%d", insertionTime.UnixNano())) {
+		for _, x := range readIn("influxdb", t, insertionTime) {
+			test := x
 			t.Run(test.Name, func(t *testing.T) {
 				t.Parallel()
-				httpWrite(t, testWriteHTTPPath, test.Data)
+				httpWrite(t, testWriteHTTPPath, bytes.NewBufferString(strings.Join(test.Data, "\n")))
 			})
 		}
 	})
 	t.Run("graphite", func(t *testing.T) {
-		for _, test := range readIn("graphite", t, fmt.Sprintf("%d", insertionTime.Unix())) {
+		for _, x := range readIn("graphite", t, insertionTime) {
+			test := x
 			t.Run(test.Name, func(t *testing.T) {
 				t.Parallel()
-				tcpWrite(t, "127.0.0.1"+testStatsDListenAddr, test.Data)
+				tcpWrite(t, "127.0.0.1"+testStatsDListenAddr, strings.Join(test.Data, "\n"))
 			})
 		}
 	})
 	t.Run("opentsdb", func(t *testing.T) {
-		for _, test := range readIn("opentsdb", t, fmt.Sprintf("%d", insertionTime.Unix())) {
+		for _, x := range readIn("opentsdb", t, insertionTime) {
+			test := x
 			t.Run(test.Name, func(t *testing.T) {
 				t.Parallel()
-				tcpWrite(t, "127.0.0.1"+testOpenTSDBListenAddr, test.Data)
+				tcpWrite(t, "127.0.0.1"+testOpenTSDBListenAddr, strings.Join(test.Data, "\n"))
+			})
+		}
+	})
+	t.Run("opentsdbhttp", func(t *testing.T) {
+		for _, x := range readIn("opentsdbhttp", t, insertionTime) {
+			test := x
+			t.Run(test.Name, func(t *testing.T) {
+				t.Parallel()
+				logger.Infof("writing %s", test.Data)
+				httpWrite(t, testOpenTSDBWriteHTTPPath, bytes.NewBufferString(strings.Join(test.Data, "\n")))
 			})
 		}
 	})
 }
 
 func testRead(t *testing.T) {
-	for _, engine := range []string{"graphite", "opentsdb", "influxdb"} {
+	for _, engine := range []string{"prometheus", "graphite", "opentsdb", "influxdb", "opentsdbhttp"} {
 		t.Run(engine, func(t *testing.T) {
-			for _, test := range readIn(engine, t, fmt.Sprintf("%d", insertionTime.UnixNano())) {
-				test := test
+			for _, x := range readIn(engine, t, insertionTime) {
+				test := x
 				t.Run(test.Name, func(t *testing.T) {
 					t.Parallel()
-					rowContains(t, httpRead(t, testReadHTTPPath, test.Query), test.Result)
+					for _, q := range test.Query {
+						q = testutil.PopulateTimeTplString(q, insertionTime)
+						if test.Issue != "" {
+							test.Issue = "Regression in " + test.Issue
+						}
+						switch true {
+						case strings.HasPrefix(q, "/api/v1/export"):
+							if err := checkMetricsResult(httpReadMetrics(t, testReadHTTPPath, q), test.ResultMetrics); err != nil {
+								t.Fatalf("Export. %s fails with error %s.%s", q, err, test.Issue)
+							}
+						case strings.HasPrefix(q, "/api/v1/series"):
+							s := Series{}
+							httpReadStruct(t, testReadHTTPPath, q, &s)
+							if err := checkSeriesResult(s, test.ResultSeries); err != nil {
+								t.Fatalf("Series. %s fails with error %s.%s", q, err, test.Issue)
+							}
+						case strings.HasPrefix(q, "/api/v1/query_range"):
+							queryResult := QueryRange{}
+							httpReadStruct(t, testReadHTTPPath, q, &queryResult)
+							if err := checkQueryRangeResult(queryResult, test.ResultQueryRange); err != nil {
+								t.Fatalf("Query Range. %s fails with error %s.%s", q, err, test.Issue)
+							}
+						case strings.HasPrefix(q, "/api/v1/query"):
+							queryResult := Query{}
+							httpReadStruct(t, testReadHTTPPath, q, &queryResult)
+							if err := checkQueryResult(queryResult, test.ResultQuery); err != nil {
+								t.Fatalf("Query. %s fails with error %s.%s", q, err, test.Issue)
+							}
+						default:
+							t.Fatalf("unsupported read query %s", q)
+						}
+					}
 				})
 			}
 		})
 	}
 }
 
-func readIn(readFor string, t *testing.T, timeStr string) []test {
+func readIn(readFor string, t *testing.T, insertTime time.Time) []test {
 	t.Helper()
 	s := newSuite(t)
 	var tt []test
@@ -195,7 +309,9 @@ func readIn(readFor string, t *testing.T, timeStr string) []test {
 		s.noError(err)
 		item := test{}
 		s.noError(json.Unmarshal(b, &item))
-		item.Data = strings.Replace(item.Data, "{TIME}", timeStr, 1)
+		for i := range item.Data {
+			item.Data[i] = testutil.PopulateTimeTplString(item.Data[i], insertTime)
+		}
 		tt = append(tt, item)
 		return nil
 	}))
@@ -205,10 +321,10 @@ func readIn(readFor string, t *testing.T, timeStr string) []test {
 	return tt
 }
 
-func httpWrite(t *testing.T, address string, data string) {
+func httpWrite(t *testing.T, address string, r io.Reader) {
 	t.Helper()
 	s := newSuite(t)
-	resp, err := http.Post(address, "", bytes.NewBufferString(data))
+	resp, err := http.Post(address, "", r)
 	s.noError(err)
 	s.noError(resp.Body.Close())
 	s.equalInt(resp.StatusCode, 204)
@@ -225,35 +341,122 @@ func tcpWrite(t *testing.T, address string, data string) {
 	s.equalInt(n, len(data))
 }
 
-func httpRead(t *testing.T, address, query string) []Row {
+func httpReadMetrics(t *testing.T, address, query string) []Metric {
 	t.Helper()
 	s := newSuite(t)
 	resp, err := http.Get(address + query)
 	s.noError(err)
 	defer resp.Body.Close()
 	s.equalInt(resp.StatusCode, 200)
-	var rows []Row
+	var rows []Metric
 	for dec := json.NewDecoder(resp.Body); dec.More(); {
-		var row Row
+		var row Metric
 		s.noError(dec.Decode(&row))
 		rows = append(rows, row)
 	}
 	return rows
 }
-
-func rowContains(t *testing.T, rows, contains []Row) {
+func httpReadStruct(t *testing.T, address, query string, dst interface{}) {
 	t.Helper()
-	for _, r := range rows {
-		contains = removeIfFound(r, contains)
-	}
-	if len(contains) > 0 {
-		t.Fatalf("result rows %+v not found in %+v", contains, rows)
-	}
+	s := newSuite(t)
+	resp, err := http.Get(address + query)
+	s.noError(err)
+	defer resp.Body.Close()
+	s.equalInt(resp.StatusCode, 200)
+	s.noError(json.NewDecoder(resp.Body).Decode(dst))
 }
 
-func removeIfFound(r Row, contains []Row) []Row {
+func checkMetricsResult(got, want []Metric) error {
+	for _, r := range append([]Metric(nil), got...) {
+		want = removeIfFoundMetrics(r, want)
+	}
+	if len(want) > 0 {
+		return fmt.Errorf("exptected metrics %+v not found in %+v", want, got)
+	}
+	return nil
+}
+
+func removeIfFoundMetrics(r Metric, contains []Metric) []Metric {
+	for i, item := range contains {
+		if reflect.DeepEqual(r.Metric, item.Metric) && reflect.DeepEqual(r.Values, item.Values) &&
+			reflect.DeepEqual(r.Timestamps, item.Timestamps) {
+			contains[i] = contains[len(contains)-1]
+			return contains[:len(contains)-1]
+		}
+	}
+	return contains
+}
+
+func checkSeriesResult(got, want Series) error {
+	if got.Status != want.Status {
+		return fmt.Errorf("status mismatch %q - %q", want.Status, got.Status)
+	}
+	wantData := append([]map[string]string(nil), want.Data...)
+	for _, r := range got.Data {
+		wantData = removeIfFoundSeries(r, wantData)
+	}
+	if len(wantData) > 0 {
+		return fmt.Errorf("expected seria(s) %+v not found in %+v", wantData, got.Data)
+	}
+	return nil
+}
+
+func removeIfFoundSeries(r map[string]string, contains []map[string]string) []map[string]string {
+	for i, item := range contains {
+		if reflect.DeepEqual(r, item) {
+			contains[i] = contains[len(contains)-1]
+			return contains[:len(contains)-1]
+		}
+	}
+	return contains
+}
+
+func checkQueryResult(got, want Query) error {
+	if got.Status != want.Status {
+		return fmt.Errorf("status mismatch %q - %q", want.Status, got.Status)
+	}
+	if got.Data.ResultType != want.Data.ResultType {
+		return fmt.Errorf("result type mismatch %q - %q", want.Data.ResultType, got.Data.ResultType)
+	}
+	wantData := append([]QueryDataResult(nil), want.Data.Result...)
+	for _, r := range got.Data.Result {
+		wantData = removeIfFoundQueryData(r, wantData)
+	}
+	if len(wantData) > 0 {
+		return fmt.Errorf("expected query result %+v not found in %+v", wantData, got.Data.Result)
+	}
+	return nil
+}
+
+func removeIfFoundQueryData(r QueryDataResult, contains []QueryDataResult) []QueryDataResult {
+	for i, item := range contains {
+		if reflect.DeepEqual(r.Metric, item.Metric) && reflect.DeepEqual(r.Value[0], item.Value[0]) && reflect.DeepEqual(r.Value[1], item.Value[1]) {
+			contains[i] = contains[len(contains)-1]
+			return contains[:len(contains)-1]
+		}
+	}
+	return contains
+}
+
+func checkQueryRangeResult(got, want QueryRange) error {
+	if got.Status != want.Status {
+		return fmt.Errorf("status mismatch %q - %q", want.Status, got.Status)
+	}
+	if got.Data.ResultType != want.Data.ResultType {
+		return fmt.Errorf("result type mismatch %q - %q", want.Data.ResultType, got.Data.ResultType)
+	}
+	wantData := append([]QueryRangeDataResult(nil), want.Data.Result...)
+	for _, r := range got.Data.Result {
+		wantData = removeIfFoundQueryRangeData(r, wantData)
+	}
+	if len(wantData) > 0 {
+		return fmt.Errorf("expected query range result %+v not found in %+v", wantData, got.Data.Result)
+	}
+	return nil
+}
+
+func removeIfFoundQueryRangeData(r QueryRangeDataResult, contains []QueryRangeDataResult) []QueryRangeDataResult {
 	for i, item := range contains {
-		// todo check time
 		if reflect.DeepEqual(r.Metric, item.Metric) && reflect.DeepEqual(r.Values, item.Values) {
 			contains[i] = contains[len(contains)-1]
 			return contains[:len(contains)-1]
@@ -281,3 +484,11 @@ func (s *suite) equalInt(a, b int) {
 		s.t.FailNow()
 	}
 }
+
+func (s *suite) greaterThan(a, b int) {
+	s.t.Helper()
+	if a <= b {
+		s.t.Errorf("%d less or equal then %d", a, b)
+		s.t.FailNow()
+	}
+}

app/victoria-metrics/test/parser.go

@@ -0,0 +1,52 @@
+package test
+
+import (
+	"fmt"
+	"log"
+	"regexp"
+	"strings"
+	"time"
+)
+
+var (
+	parseTimeExpRegex = regexp.MustCompile(`"?{TIME[^}]*}"?`)
+	extractRegex      = regexp.MustCompile(`"?{([^}]*)}"?`)
+)
+
+// PopulateTimeTplString substitutes {TIME_*} with t in s and returns the result.
+func PopulateTimeTplString(s string, t time.Time) string {
+	return string(PopulateTimeTpl([]byte(s), t))
+}
+
+// PopulateTimeTpl substitutes {TIME_*} with tGlobal in b and returns the result.
+func PopulateTimeTpl(b []byte, tGlobal time.Time) []byte {
+	return parseTimeExpRegex.ReplaceAllFunc(b, func(repl []byte) []byte {
+		t := tGlobal
+		repl = extractRegex.FindSubmatch(repl)[1]
+		parts := strings.SplitN(string(repl), "-", 2)
+		if len(parts) == 2 {
+			duration, err := time.ParseDuration(strings.TrimSpace(parts[1]))
+			if err != nil {
+				log.Fatalf("error %s parsing duration %s in %s", err, parts[1], repl)
+			}
+			t = t.Add(-duration)
+		}
+		switch strings.TrimSpace(parts[0]) {
+		case `TIME_S`:
+			return []byte(fmt.Sprintf("%d", t.Unix()))
+		case `TIME_MSZ`:
+			return []byte(fmt.Sprintf("%d", t.Unix()*1e3))
+		case `TIME_MS`:
+			return []byte(fmt.Sprintf("%d", timeToMillis(t)))
+		case `TIME_NS`:
+			return []byte(fmt.Sprintf("%d", t.UnixNano()))
+		default:
+			log.Fatalf("unknown time pattern %s in %s", parts[0], repl)
+		}
+		return repl
+	})
+}
+
+func timeToMillis(t time.Time) int64 {
+	return t.UnixNano() / 1e6
+}

app/victoria-metrics/test/parser_test.go

@@ -0,0 +1,24 @@
+package test
+
+import (
+	"testing"
+	"time"
+)
+
+func TestPopulateTimeTplString(t *testing.T) {
+	now, err := time.Parse(time.RFC3339, "2006-01-02T15:04:05Z")
+	if err != nil {
+		t.Fatalf("unexpected error when parsing time: %s", err)
+	}
+	f := func(s, resultExpected string) {
+		t.Helper()
+		result := PopulateTimeTplString(s, now)
+		if result != resultExpected {
+			t.Fatalf("unexpected result; got %q; want %q", result, resultExpected)
+		}
+	}
+	f("", "")
+	f("{TIME_S}", "1136214245")
+	f("now: {TIME_S}, past 30s: {TIME_MS-30s}, now: {TIME_S}", "now: 1136214245, past 30s: 1136214215000, now: 1136214245")
+	f("now: {TIME_MS}, past 30m: {TIME_MSZ-30m}, past 2h: {TIME_NS-2h}", "now: 1136214245000, past 30m: 1136212445000, past 2h: 1136207045000000000")
+}

app/victoria-metrics/test/prom_types.go

@@ -0,0 +1,338 @@
+// +build integration
+
+// Source https://github.com/prometheus/prometheus/blob/master/prompb/remote.pb.go . Code is copy pasted and cleaned up
+package test
+
+import (
+	"encoding/binary"
+	"math"
+	"math/bits"
+)
+
+type WriteRequest struct {
+	Timeseries []TimeSeries `protobuf:"bytes,1,rep,name=timeseries,proto3" json:"timeseries"`
+}
+
+func (m *WriteRequest) Size() (n int) {
+	if m == nil {
+		return 0
+	}
+	var l int
+	_ = l
+	if len(m.Timeseries) > 0 {
+		for _, e := range m.Timeseries {
+			l = e.Size()
+			n += 1 + l + sovRemote(uint64(l))
+		}
+	}
+	return n
+}
+func sovRemote(x uint64) (n int) {
+	return (bits.Len64(x|1) + 6) / 7
+}
+
+func (m *WriteRequest) Marshal() (dAtA []byte, err error) {
+	size := m.Size()
+	dAtA = make([]byte, size)
+	n, err := m.MarshalToSizedBuffer(dAtA[:size])
+	if err != nil {
+		return nil, err
+	}
+	return dAtA[:n], nil
+}
+
+func (m *WriteRequest) MarshalTo(dAtA []byte) (int, error) {
+	size := m.Size()
+	return m.MarshalToSizedBuffer(dAtA[:size])
+}
+
+func (m *WriteRequest) MarshalToSizedBuffer(dAtA []byte) (int, error) {
+	i := len(dAtA)
+	if len(m.Timeseries) > 0 {
+		for iNdEx := len(m.Timeseries) - 1; iNdEx >= 0; iNdEx-- {
+			{
+				size, err := m.Timeseries[iNdEx].MarshalToSizedBuffer(dAtA[:i])
+				if err != nil {
+					return 0, err
+				}
+				i -= size
+				i = encodeVarintRemote(dAtA, i, uint64(size))
+			}
+			i--
+			dAtA[i] = 0xa
+		}
+	}
+	return len(dAtA) - i, nil
+}
+
+func encodeVarintRemote(dAtA []byte, offset int, v uint64) int {
+	offset -= sovRemote(v)
+	base := offset
+	for v >= 1<<7 {
+		dAtA[offset] = uint8(v&0x7f | 0x80)
+		v >>= 7
+		offset++
+	}
+	dAtA[offset] = uint8(v)
+	return base
+}
+
+type Sample struct {
+	Value     float64 `protobuf:"fixed64,1,opt,name=value,proto3" json:"value,omitempty"`
+	Timestamp int64   `protobuf:"varint,2,opt,name=timestamp,proto3" json:"timestamp,omitempty"`
+}
+
+func (m *Sample) Reset() { *m = Sample{} }
+
+// TimeSeries represents samples and labels for a single time series.
+type TimeSeries struct {
+	Labels  []Label  `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels"`
+	Samples []Sample `protobuf:"bytes,2,rep,name=samples,proto3" json:"samples"`
+}
+
+func (m *TimeSeries) Reset() { *m = TimeSeries{} }
+
+type Label struct {
+	Name  string `protobuf:"bytes,1,opt,name=name,proto3" json:"name,omitempty"`
+	Value string `protobuf:"bytes,2,opt,name=value,proto3" json:"value,omitempty"`
+}
+
+func (m *Label) Reset() { *m = Label{} }
+
+type Labels struct {
+	Labels []Label `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels"`
+}
+
+func (m *Labels) Reset() { *m = Labels{} }
+
+func (m *Sample) Marshal() (dAtA []byte, err error) {
+	size := m.Size()
+	dAtA = make([]byte, size)
+	n, err := m.MarshalToSizedBuffer(dAtA[:size])
+	if err != nil {
+		return nil, err
+	}
+	return dAtA[:n], nil
+}
+
+func (m *Sample) MarshalTo(dAtA []byte) (int, error) {
+	size := m.Size()
+	return m.MarshalToSizedBuffer(dAtA[:size])
+}
+
+func (m *Sample) MarshalToSizedBuffer(dAtA []byte) (int, error) {
+	i := len(dAtA)
+	if m.Timestamp != 0 {
+		i = encodeVarintTypes(dAtA, i, uint64(m.Timestamp))
+		i--
+		dAtA[i] = 0x10
+	}
+	if m.Value != 0 {
+		i -= 8
+		binary.LittleEndian.PutUint64(dAtA[i:], uint64(math.Float64bits(float64(m.Value))))
+		i--
+		dAtA[i] = 0x9
+	}
+	return len(dAtA) - i, nil
+}
+
+func (m *TimeSeries) Marshal() (dAtA []byte, err error) {
+	size := m.Size()
+	dAtA = make([]byte, size)
+	n, err := m.MarshalToSizedBuffer(dAtA[:size])
+	if err != nil {
+		return nil, err
+	}
+	return dAtA[:n], nil
+}
+
+func (m *TimeSeries) MarshalTo(dAtA []byte) (int, error) {
+	size := m.Size()
+	return m.MarshalToSizedBuffer(dAtA[:size])
+}
+
+func (m *TimeSeries) MarshalToSizedBuffer(dAtA []byte) (int, error) {
+	i := len(dAtA)
+	if len(m.Samples) > 0 {
+		for iNdEx := len(m.Samples) - 1; iNdEx >= 0; iNdEx-- {
+			{
+				size, err := m.Samples[iNdEx].MarshalToSizedBuffer(dAtA[:i])
+				if err != nil {
+					return 0, err
+				}
+				i -= size
+				i = encodeVarintTypes(dAtA, i, uint64(size))
+			}
+			i--
+			dAtA[i] = 0x12
+		}
+	}
+	if len(m.Labels) > 0 {
+		for iNdEx := len(m.Labels) - 1; iNdEx >= 0; iNdEx-- {
+			{
+				size, err := m.Labels[iNdEx].MarshalToSizedBuffer(dAtA[:i])
+				if err != nil {
+					return 0, err
+				}
+				i -= size
+				i = encodeVarintTypes(dAtA, i, uint64(size))
+			}
+			i--
+			dAtA[i] = 0xa
+		}
+	}
+	return len(dAtA) - i, nil
+}
+
+func (m *Label) Marshal() (dAtA []byte, err error) {
+	size := m.Size()
+	dAtA = make([]byte, size)
+	n, err := m.MarshalToSizedBuffer(dAtA[:size])
+	if err != nil {
+		return nil, err
+	}
+	return dAtA[:n], nil
+}
+
+func (m *Label) MarshalTo(dAtA []byte) (int, error) {
+	size := m.Size()
+	return m.MarshalToSizedBuffer(dAtA[:size])
+}
+
+func (m *Label) MarshalToSizedBuffer(dAtA []byte) (int, error) {
+	i := len(dAtA)
+	_ = i
+	var l int
+	_ = l
+	if len(m.Value) > 0 {
+		i -= len(m.Value)
+		copy(dAtA[i:], m.Value)
+		i = encodeVarintTypes(dAtA, i, uint64(len(m.Value)))
+		i--
+		dAtA[i] = 0x12
+	}
+	if len(m.Name) > 0 {
+		i -= len(m.Name)
+		copy(dAtA[i:], m.Name)
+		i = encodeVarintTypes(dAtA, i, uint64(len(m.Name)))
+		i--
+		dAtA[i] = 0xa
+	}
+	return len(dAtA) - i, nil
+}
+
+func (m *Labels) Marshal() (dAtA []byte, err error) {
+	size := m.Size()
+	dAtA = make([]byte, size)
+	n, err := m.MarshalToSizedBuffer(dAtA[:size])
+	if err != nil {
+		return nil, err
+	}
+	return dAtA[:n], nil
+}
+
+func (m *Labels) MarshalTo(dAtA []byte) (int, error) {
+	size := m.Size()
+	return m.MarshalToSizedBuffer(dAtA[:size])
+}
+
+func (m *Labels) MarshalToSizedBuffer(dAtA []byte) (int, error) {
+	i := len(dAtA)
+	if len(m.Labels) > 0 {
+		for iNdEx := len(m.Labels) - 1; iNdEx >= 0; iNdEx-- {
+			{
+				size, err := m.Labels[iNdEx].MarshalToSizedBuffer(dAtA[:i])
+				if err != nil {
+					return 0, err
+				}
+				i -= size
+				i = encodeVarintTypes(dAtA, i, uint64(size))
+			}
+			i--
+			dAtA[i] = 0xa
+		}
+	}
+	return len(dAtA) - i, nil
+}
+
+func encodeVarintTypes(dAtA []byte, offset int, v uint64) int {
+	offset -= sovTypes(v)
+	base := offset
+	for v >= 1<<7 {
+		dAtA[offset] = uint8(v&0x7f | 0x80)
+		v >>= 7
+		offset++
+	}
+	dAtA[offset] = uint8(v)
+	return base
+}
+
+func (m *Sample) Size() (n int) {
+	if m == nil {
+		return 0
+	}
+	if m.Value != 0 {
+		n += 9
+	}
+	if m.Timestamp != 0 {
+		n += 1 + sovTypes(uint64(m.Timestamp))
+	}
+	return n
+}
+
+func (m *TimeSeries) Size() (n int) {
+	if m == nil {
+		return 0
+	}
+	var l int
+	_ = l
+	if len(m.Labels) > 0 {
+		for _, e := range m.Labels {
+			l = e.Size()
+			n += 1 + l + sovTypes(uint64(l))
+		}
+	}
+	if len(m.Samples) > 0 {
+		for _, e := range m.Samples {
+			l = e.Size()
+			n += 1 + l + sovTypes(uint64(l))
+		}
+	}
+	return n
+}
+
+func (m *Label) Size() (n int) {
+	if m == nil {
+		return 0
+	}
+	var l int
+	_ = l
+	l = len(m.Name)
+	if l > 0 {
+		n += 1 + l + sovTypes(uint64(l))
+	}
+	l = len(m.Value)
+	if l > 0 {
+		n += 1 + l + sovTypes(uint64(l))
+	}
+	return n
+}
+
+func (m *Labels) Size() (n int) {
+	if m == nil {
+		return 0
+	}
+	var l int
+	_ = l
+	if len(m.Labels) > 0 {
+		for _, e := range m.Labels {
+			l = e.Size()
+			n += 1 + l + sovTypes(uint64(l))
+		}
+	}
+	return n
+}
+
+func sovTypes(x uint64) (n int) {
+	return (bits.Len64(x|1) + 6) / 7
+}

app/victoria-metrics/test/prom_writter.go

@@ -0,0 +1,13 @@
+// +build integration
+
+package test
+
+import "github.com/golang/snappy"
+
+func Compress(wr WriteRequest) ([]byte, error) {
+	data, err := wr.Marshal()
+	if err != nil {
+		return nil, err
+	}
+	return snappy.Encode(nil, data), nil
+}

app/victoria-metrics/testdata/graphite/basic.json

@@ -1,8 +1,8 @@
 {
   "name": "basic_insertion",
-  "data": "graphite.foo.bar.baz;tag1=value1;tag2=value2 123 {TIME}",
-  "query": "/api/v1/export?match={__name__!=\"\"}",
-  "result": [
-    {"metric":{"__name__":"graphite.foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123]}
+  "data": ["graphite.foo.bar.baz;tag1=value1;tag2=value2 123 {TIME_S}"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"graphite.foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123], "timestamps": ["{TIME_MSZ}"]}
   ]
 }

app/victoria-metrics/testdata/graphite/comparison-not-inf-not-nan.json

@@ -0,0 +1,16 @@
+{
+  "name": "comparison-not-inf-not-nan",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/150",
+  "data": [
+    "not_nan_not_inf;item=x 1 {TIME_S-1m}",
+    "not_nan_not_inf;item=x 1 {TIME_S-2m}",
+    "not_nan_not_inf;item=y 3 {TIME_S-1m}",
+    "not_nan_not_inf;item=y 1 {TIME_S-2m}"],
+  "query": ["/api/v1/query_range?query=1/(not_nan_not_inf-1)!=inf!=nan&start={TIME_S-3m}&end={TIME_S}&step=60"],
+  "result_query_range": {
+    "status":"success",
+    "data":{"resultType":"matrix",
+      "result":[
+	      {"metric":{"item":"y"},"values":[["{TIME_S-1m}","0.5"],["{TIME_S}","0.5"]]}
+      ]}}
+}

app/victoria-metrics/testdata/graphite/max_lookback_set.json

@@ -0,0 +1,24 @@
+{
+  "name": "max_lookback_set",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/209",
+  "data": [
+    "max_lookback_set 1 {TIME_S-30s}",
+    "max_lookback_set 2 {TIME_S-60s}",
+    "max_lookback_set 3 {TIME_S-120s}",
+    "max_lookback_set 4 {TIME_S-150s}"
+  ],
+  "query": ["/api/v1/query_range?query=max_lookback_set&start={TIME_S-150s}&end={TIME_S}&step=10s&max_lookback=1s"],
+  "result_query_range": {
+    "status":"success",
+    "data":{"resultType":"matrix",
+      "result":[{"metric":{"__name__":"max_lookback_set"},"values":[
+	      ["{TIME_S-150s}","4"],
+	      ["{TIME_S-140s}","4"],
+	      ["{TIME_S-120s}","3"],
+	      ["{TIME_S-110s}","3"],
+	      ["{TIME_S-60s}","2"],
+	      ["{TIME_S-50s}","2"],
+	      ["{TIME_S-30s}","1"],
+	      ["{TIME_S-20s}","1"]
+      ]}]}}
+}

app/victoria-metrics/testdata/graphite/max_lookback_unset.json

@@ -0,0 +1,32 @@
+{
+  "name": "max_lookback_unset",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/209",
+  "data": [
+    "max_lookback_unset 1 {TIME_S-30s}",
+    "max_lookback_unset 2 {TIME_S-60s}",
+    "max_lookback_unset 3 {TIME_S-120s}",
+    "max_lookback_unset 4 {TIME_S-150s}"
+  ],
+  "query": ["/api/v1/query_range?query=max_lookback_unset&start={TIME_S-150s}&end={TIME_S}&step=10s"],
+  "result_query_range": {
+    "status":"success",
+    "data":{"resultType":"matrix",
+      "result":[{"metric":{"__name__":"max_lookback_unset"},"values":[
+	      ["{TIME_S-150s}","4"],
+	      ["{TIME_S-140s}","4"],
+	      ["{TIME_S-130s}","4"],
+	      ["{TIME_S-120s}","3"],
+	      ["{TIME_S-110s}","3"],
+	      ["{TIME_S-100s}","3"],
+	      ["{TIME_S-90s}","3"],
+	      ["{TIME_S-80s}","3"],
+	      ["{TIME_S-70s}","3"],
+	      ["{TIME_S-60s}","2"],
+	      ["{TIME_S-50s}","2"],
+	      ["{TIME_S-40s}","2"],
+	      ["{TIME_S-30s}","1"],
+	      ["{TIME_S-20s}","1"],
+	      ["{TIME_S-10s}","1"],
+	      ["{TIME_S}","1"]
+      ]}]}}
+}

app/victoria-metrics/testdata/graphite/not-nan-as-missing-data.json

@@ -0,0 +1,18 @@
+{
+  "name": "not-nan-as-missing-data",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/153",
+  "data": [
+    "not_nan_as_missing_data;item=x 2 {TIME_S-2m}",
+    "not_nan_as_missing_data;item=x 1 {TIME_S-1m}",
+    "not_nan_as_missing_data;item=y 4 {TIME_S-2m}",
+    "not_nan_as_missing_data;item=y 3 {TIME_S-1m}"
+  ],
+  "query": ["/api/v1/query_range?query=not_nan_as_missing_data>1&start={TIME_S-2m}&end={TIME_S}&step=60"],
+  "result_query_range": {
+    "status":"success",
+    "data":{"resultType":"matrix",
+      "result":[
+	      {"metric":{"__name__":"not_nan_as_missing_data","item":"x"},"values":[["{TIME_S-2m}","2"]]},
+	      {"metric":{"__name__":"not_nan_as_missing_data","item":"y"},"values":[["{TIME_S-2m}","4"],["{TIME_S-1m}","3"],["{TIME_S}","3"]]}
+      ]}}
+}

app/victoria-metrics/testdata/graphite/subquery-aggregation.json

@@ -0,0 +1,14 @@
+{
+  "name": "subquery-aggregation",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/184",
+  "data": [
+    "forms_daily_count;item=x 1 {TIME_S-1m}",
+    "forms_daily_count;item=x 2 {TIME_S-2m}",
+    "forms_daily_count;item=y 3 {TIME_S-1m}",
+    "forms_daily_count;item=y 4 {TIME_S-2m}"],
+  "query": ["/api/v1/query?query=min%20by%20(item)%20(min_over_time(forms_daily_count[10m:1m]))&time={TIME_S-1m}"],
+  "result_query": {
+    "status":"success",
+    "data":{"resultType":"vector","result":[{"metric":{"item":"x"},"value":["{TIME_S-1m}","1"]},{"metric":{"item":"y"},"value":["{TIME_S-1m}","3"]}]}
+  }
+}

app/victoria-metrics/testdata/influxdb/basic.json

@@ -1,9 +1,9 @@
 {
   "name": "basic_insertion",
-  "data": "measurement,tag1=value1,tag2=value2 field1=1.23,field2=123",
-  "query": "/api/v1/export?match={__name__!=\"\"}",
-  "result": [
-    {"metric":{"__name__":"measurement_field2","tag1":"value1","tag2":"value2"},"values":[123]},
-    {"metric":{"__name__":"measurement_field1","tag1":"value1","tag2":"value2"},"values":[1.23]}
+  "data": ["measurement,tag1=value1,tag2=value2 field1=1.23,field2=123 {TIME_NS}"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"measurement_field2","tag1":"value1","tag2":"value2"},"values":[123], "timestamps": ["{TIME_MS}"]},
+    {"metric":{"__name__":"measurement_field1","tag1":"value1","tag2":"value2"},"values":[1.23], "timestamps": ["{TIME_MS}"]}
   ]
 }

app/victoria-metrics/testdata/opentsdb/basic.json

@@ -1,8 +1,8 @@
 {
   "name": "basic_insertion",
-  "data": "put openstdb.foo.bar.baz {TIME} 123 tag1=value1 tag2=value2",
-  "query": "/api/v1/export?match={__name__!=\"\"}",
-  "result": [
-    {"metric":{"__name__":"openstdb.foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123]}
+  "data": ["put openstdb.foo.bar.baz {TIME_S} 123 tag1=value1 tag2=value2"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"openstdb.foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123], "timestamps": ["{TIME_MSZ}"]}
   ]
 }

app/victoria-metrics/testdata/opentsdbhttp/basic.json

@@ -0,0 +1,8 @@
+{
+  "name": "basic_insertion",
+  "data": ["{\"metric\": \"opentsdbhttp.foo\", \"value\": 1001, \"timestamp\": {TIME_S}, \"tags\": {\"bar\":\"baz\", \"x\": \"y\"}}"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"opentsdbhttp.foo","bar":"baz","x":"y"},"values":[1001], "timestamps": ["{TIME_MSZ}"]}
+  ]
+}

app/victoria-metrics/testdata/opentsdbhttp/multi_line.json

@@ -0,0 +1,9 @@
+{
+  "name": "multiline",
+  "data": ["[{\"metric\": \"opentsdbhttp.multiline1\", \"value\": 1001, \"timestamp\": \"{TIME_S}\", \"tags\": {\"bar\":\"baz\", \"x\": \"y\"}}, {\"metric\": \"opentsdbhttp.multiline2\", \"value\": 1002, \"timestamp\": {TIME_S}}]"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"opentsdbhttp.multiline1","bar":"baz","x":"y"},"values":[1001], "timestamps": ["{TIME_MSZ}"]},
+    {"metric":{"__name__":"opentsdbhttp.multiline2"},"values":[1002], "timestamps": ["{TIME_MSZ}"]}
+  ]
+}

app/victoria-metrics/testdata/prometheus/basic.json

@@ -0,0 +1,8 @@
+{
+  "name": "basic_insertion",
+  "data": ["[{\"labels\":[{\"name\":\"__name__\",\"value\":\"prometheus.bar\"},{\"name\":\"baz\",\"value\":\"qux\"}],\"samples\":[{\"value\":100000,\"timestamp\":\"{TIME_MS}\"}]}]"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"prometheus.bar","baz":"qux"},"values":[100000], "timestamps": ["{TIME_MS}"]}
+  ]
+}

app/victoria-metrics/testdata/prometheus/case-sensitive-regex.json

@@ -0,0 +1,10 @@
+{
+  "name": "case-sensitive-regex",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/161",
+  "data": ["[{\"labels\":[{\"name\":\"__name__\",\"value\":\"prometheus.sensitiveRegex\"},{\"name\":\"label\",\"value\":\"sensitiveRegex\"}],\"samples\":[{\"value\":2,\"timestamp\":\"{TIME_MS}\"}]},{\"labels\":[{\"name\":\"__name__\",\"value\":\"prometheus.sensitiveRegex\"},{\"name\":\"label\",\"value\":\"SensitiveRegex\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]}]"],
+  "query": ["/api/v1/export?match={label=~'(?i)sensitiveregex'}"],
+  "result_metrics": [
+    {"metric":{"__name__":"prometheus.sensitiveRegex","label":"sensitiveRegex"},"values":[2], "timestamps": ["{TIME_MS}"]},
+    {"metric":{"__name__":"prometheus.sensitiveRegex","label":"SensitiveRegex"},"values":[1], "timestamps": ["{TIME_MS}"]}
+  ]
+}

app/victoria-metrics/testdata/prometheus/duplicate-label.json

@@ -0,0 +1,9 @@
+{
+  "name": "duplicate_label",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/172",
+  "data": ["[{\"labels\":[{\"name\":\"__name__\",\"value\":\"prometheus.duplicate_label\"},{\"name\":\"duplicate\",\"value\":\"label\"},{\"name\":\"duplicate\",\"value\":\"label\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]}]"],
+  "query": ["/api/v1/export?match={__name__!=''}"],
+  "result_metrics": [
+    {"metric":{"__name__":"prometheus.duplicate_label","duplicate":"label"},"values":[1], "timestamps": ["{TIME_MS}"]}
+  ]
+}

app/victoria-metrics/testdata/prometheus/match-series.json

@@ -0,0 +1,15 @@
+{
+  "name": "match_series",
+  "issue": "https://github.com/VictoriaMetrics/VictoriaMetrics/issues/155",
+  "data": ["[{\"labels\":[{\"name\":\"__name__\",\"value\":\"MatchSeries\"},{\"name\":\"db\",\"value\":\"TenMinute\"},{\"name\":\"TurbineType\",\"value\":\"V112\"},{\"name\":\"Park\",\"value\":\"1\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]},{\"labels\":[{\"name\":\"__name__\",\"value\":\"MatchSeries\"},{\"name\":\"db\",\"value\":\"TenMinute\"},{\"name\":\"TurbineType\",\"value\":\"V112\"},{\"name\":\"Park\",\"value\":\"2\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]},{\"labels\":[{\"name\":\"__name__\",\"value\":\"MatchSeries\"},{\"name\":\"db\",\"value\":\"TenMinute\"},{\"name\":\"TurbineType\",\"value\":\"V112\"},{\"name\":\"Park\",\"value\":\"3\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]},{\"labels\":[{\"name\":\"__name__\",\"value\":\"MatchSeries\"},{\"name\":\"db\",\"value\":\"TenMinute\"},{\"name\":\"TurbineType\",\"value\":\"V112\"},{\"name\":\"Park\",\"value\":\"4\"}],\"samples\":[{\"value\":1,\"timestamp\":\"{TIME_MS}\"}]}]"],
+  "query": ["/api/v1/series?match[]={__name__='MatchSeries'}", "/api/v1/series?match[]={__name__=~'MatchSeries.*'}"],
+  "result_series": {
+    "status": "success",
+    "data": [
+      {"__name__":"MatchSeries","db":"TenMinute","Park":"1","TurbineType":"V112"},
+      {"__name__":"MatchSeries","db":"TenMinute","Park":"2","TurbineType":"V112"},
+      {"__name__":"MatchSeries","db":"TenMinute","Park":"3","TurbineType":"V112"},
+      {"__name__":"MatchSeries","db":"TenMinute","Park":"4","TurbineType":"V112"}
+    ]
+  }
+}

app/vmbackup/Makefile

@@ -0,0 +1,37 @@
+# All these commands must run from repository root.
+
+vmbackup:
+	APP_NAME=vmbackup $(MAKE) app-local
+
+vmbackup-prod:
+	APP_NAME=vmbackup $(MAKE) app-via-docker
+
+package-vmbackup:
+	APP_NAME=vmbackup $(MAKE) package-via-docker
+
+publish-vmbackup:
+	APP_NAME=vmbackup $(MAKE) publish-via-docker
+
+vmbackup-arm:
+	CGO_ENABLED=0 GOOS=linux GOARCH=arm GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmbackup-arm ./app/vmbackup
+
+vmbackup-arm-prod:
+	APP_NAME=vmbackup APP_SUFFIX='-arm' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=arm' $(MAKE) app-via-docker
+
+vmbackup-arm64:
+	CGO_ENABLED=0 GOOS=linux GOARCH=arm64 GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmbackup-arm64 ./app/vmbackup
+
+vmbackup-arm64-prod:
+	APP_NAME=vmbackup APP_SUFFIX='-arm64' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=arm64' $(MAKE) app-via-docker
+
+vmbackup-386:
+	CGO_ENABLED=0 GOOS=linux GOARCH=386 GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmbackup-386 ./app/vmbackup
+
+vmbackup-386-prod:
+	APP_NAME=vmbackup APP_SUFFIX='-386' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=386' $(MAKE) app-via-docker
+
+vmbackup-pure:
+	APP_NAME=vmbackup $(MAKE) app-local-pure
+
+vmbackup-pure-prod:
+	APP_NAME=vmbackup APP_SUFFIX='-pure' DOCKER_OPTS='--env CGO_ENABLED=0' $(MAKE) app-via-docker

app/vmbackup/README.md

@@ -0,0 +1,181 @@
+## vmbackup
+
+`vmbackup` creates VictoriaMetrics data backups from [instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+
+Supported storage systems for backups:
+
+* [GCS](https://cloud.google.com/storage/). Example: `gcs://<bucket>/<path/to/backup>`
+* [S3](https://aws.amazon.com/s3/). Example: `s3://<bucket>/<path/to/backup>`
+* Any S3-compatible storage such as [MinIO](https://github.com/minio/minio). See `-customS3Endpoint` command-line flag.
+* Local filesystem. Example: `fs://</absolute/path/to/backup>`
+
+Incremental backups and full backups are supported. Incremental backups are created automatically if the destination path already contains data from the previous backup.
+Full backups can be sped up with `-origin` pointing to already existing backup on the same remote storage. In this case `vmbackup` makes server-side copy for the shared
+data between the existing backup and new backup. This saves time and costs on data transfer.
+
+Backup process can be interrupted at any time. It is automatically resumed from the interruption point when restarting `vmbackup` with the same args.
+
+Backed up data can be restored with [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md).
+
+See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+
+
+### Use cases
+
+#### Regular backups
+
+Regular backup can be performed with the following command:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/new/backup>
+```
+
+* `</path/to/victoria-metrics-data>` - path to VictoriaMetrics data pointed by `-storageDataPath` command-line flag in single-node VictoriaMetrics or in cluster `vmstorage`.
+  There is no need to stop VictoriaMetrics for creating backups, since they are performed from immutable [instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+* `<local-snapshot>` is the snapshot to backup. See [how to create instant snapshots](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+* `<bucket>` is already existing name for [GCS bucket](https://cloud.google.com/storage/docs/creating-buckets).
+* `<path/to/new/backup>` is the destination path where new backup will be placed.
+
+
+#### Regular backups with server-side copy from existing backup
+
+If the destination GCS bucket already contains the previous backup at `-origin` path, then new backup can be sped up
+with the following command:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/new/backup> -origin=gcs://<bucket>/<path/to/existing/backup>
+```
+
+This saves time and network bandwidth costs by performing server-side copy for the shared data from the `-origin` to `-dst`.
+
+
+#### Incremental backups
+
+Incremental backups are performed if `-dst` points to already existing backup. In this case only new data is uploaded to remote storage.
+This saves time and network bandwidth costs when working with big backups:
+
+```
+vmbackup -storageDataPath=</path/to/victoria-metrics-data> -snapshotName=<local-snapshot> -dst=gcs://<bucket>/<path/to/existing/backup>
+```
+
+
+#### Smart backups
+
+Smart backups mean storing full daily backups into `YYYYMMDD` folders and creating incremental hourly backup into `latest` folder:
+
+* Run the following command every hour:
+
+```
+vmbackup -snapshotName=<latest-snapshot> -dst=gcs://<bucket>/latest
+```
+
+Where `<latest-snapshot>` is the latest [snapshot](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots).
+The command will upload only changed data to `gcs://<bucket>/latest`.
+
+* Run the following command once a day:
+
+```
+vmbackup -snapshotName=<daily-snapshot> -dst=gcs://<bucket>/<YYYYMMDD> -origin=gcs://<bucket>/latest
+```
+
+Where `<daily-snapshot>` is the snapshot for the last day `<YYYYMMDD>`.
+
+
+This apporach saves network bandwidth costs on hourly backups (since they are incremental) and allows recovering data from either the last hour (`latest` backup)
+or from any day (`YYYYMMDD` backups). Note that hourly backup shouldn't run when creating daily backup.
+
+Do not forget removing old snapshots and backups when they are no longer needed for saving storage costs.
+
+
+### How does it work?
+
+The backup algorithm is the following:
+
+1. Collect information about files in the `-snapshotName`, in the `-dst` and in the `-origin`.
+2. Determine files in `-dst`, which are missing in `-snapshotName`, and delete them. These are usually small files, which are already merged into bigger files in the snapshot.
+3. Determine files from `-snapshotName`, which are missing in `-dst`. These are usually small new files and bigger merged files.
+4. Determine files from step 3, which exist in the `-origin`, and perform server-side copy of these files from `-origin` to `-dst`.
+   This are usually the biggest and the oldest files, which are shared between backups.
+5. Upload the remaining files from setp 3 from `-snapshotName` to `-dst`.
+
+The algorithm splits source files into 100MB chunks in the backup. Each chunk is stored as a separate file in the backup.
+Such splitting minimizes the amounts of data to re-transfer after temporary errors.
+
+`vmbackup` relies on [instant snapshot](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282) properties:
+
+- All the files in the snapshot are immutable.
+- Old files are periodically merged into new files.
+- Smaller files have higher probability to be merged.
+- Consecutive snapshots share many identical files.
+
+These properties allow performing fast and cheap incremental backups and server-side copying from `-origin` paths.
+See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+`vmbackup` can work improperly or slowly when these properties are violated.
+
+
+### Troubleshooting
+
+* If the backup is slow, then try setting higher value for `-concurrency` flag. This will increase the number of concurrent workers that upload data to backup storage.
+* If `vmbackup` eats all the network bandwidth, then set `-maxBytesPerSecond` to the desired value.
+* If `vmbackup` has been interrupted due to temporary error, then just restart it with the same args. It will resume the backup process.
+
+
+### Advanced usage
+
+Run `vmbackup -help` in order to see all the available options:
+
+```
+  -concurrency int
+    	The number of concurrent workers. Higher concurrency may reduce backup duration (default 10)
+  -configFilePath string
+    	Path to file with S3 configs. Configs are loaded from default location if not set.
+    	See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -configProfile string
+    	Profile name for S3 configs (default "default")
+  -credsFilePath string
+    	Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set.
+    	See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -customS3Endpoint string
+    	Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set
+  -dst string
+    	Where to put the backup on the remote storage. Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir
+    	-dst can point to the previous backup. In this case incremental backup is performed, i.e. only changed data is uploaded
+  -loggerLevel string
+    	Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC (default "INFO")
+  -maxBytesPerSecond int
+    	The maximum upload speed. There is no limit if it is set to 0
+  -memory.allowedPercent float
+    	Allowed percent of system memory VictoriaMetrics caches may occupy (default 60)
+  -origin string
+    	Optional origin directory on the remote storage with old backup for server-side copying when performing full backup. This speeds up full backups
+  -snapshotName string
+    	Name for the snapshot to backup. See https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots
+  -storageDataPath string
+    	Path to VictoriaMetrics data. Must match -storageDataPath from VictoriaMetrics or vmstorage (default "victoria-metrics-data")
+  -version
+    	Show VictoriaMetrics version
+```
+
+
+### How to build from sources
+
+It is recommended using [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) - see `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make vmbackup` from the root folder of the repository.
+   It builds `vmbackup` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmbackup-prod` from the root folder of the repository.
+   It builds `vmbackup-prod` binary and puts it into the `bin` folder.
+
+#### Building docker images
+
+Run `make package-vmbackup`. It builds `victoriametrics/vmbackup:<PKG_TAG>` docker image locally.
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package-vmbackup`.

app/vmbackup/deployment/Dockerfile

@@ -0,0 +1,5 @@
+FROM scratch
+COPY --from=local/certs:1.0.3 /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
+COPY bin/vmbackup-prod .
+EXPOSE 8428
+ENTRYPOINT ["/vmbackup-prod"]

app/vmbackup/main.go

@@ -0,0 +1,114 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/actions"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fslocal"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/buildinfo"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+var (
+	storageDataPath = flag.String("storageDataPath", "victoria-metrics-data", "Path to VictoriaMetrics data. Must match -storageDataPath from VictoriaMetrics or vmstorage")
+	snapshotName    = flag.String("snapshotName", "", "Name for the snapshot to backup. See https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-work-with-snapshots")
+	dst             = flag.String("dst", "", "Where to put the backup on the remote storage. "+
+		"Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir\n"+
+		"-dst can point to the previous backup. In this case incremental backup is performed, i.e. only changed data is uploaded")
+	origin            = flag.String("origin", "", "Optional origin directory on the remote storage with old backup for server-side copying when performing full backup. This speeds up full backups")
+	concurrency       = flag.Int("concurrency", 10, "The number of concurrent workers. Higher concurrency may reduce backup duration")
+	maxBytesPerSecond = flag.Int("maxBytesPerSecond", 0, "The maximum upload speed. There is no limit if it is set to 0")
+)
+
+func main() {
+	flag.Usage = usage
+	flag.Parse()
+	buildinfo.Init()
+
+	srcFS, err := newSrcFS()
+	if err != nil {
+		logger.Fatalf("%s", err)
+	}
+	dstFS, err := newDstFS()
+	if err != nil {
+		logger.Fatalf("%s", err)
+	}
+	originFS, err := newOriginFS()
+	if err != nil {
+		logger.Fatalf("%s", err)
+	}
+	a := &actions.Backup{
+		Concurrency: *concurrency,
+		Src:         srcFS,
+		Dst:         dstFS,
+		Origin:      originFS,
+	}
+	if err := a.Run(); err != nil {
+		logger.Fatalf("cannot create backup: %s", err)
+	}
+}
+
+func usage() {
+	const s = `
+vmbackup performs backups for VictoriaMetrics data from instant snapshots to gcs, s3
+or local filesystem. Backed up data can be restored with vmrestore.
+
+See the docs at https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md .
+`
+
+	f := flag.CommandLine.Output()
+	fmt.Fprintf(f, "%s\n", s)
+	flag.PrintDefaults()
+}
+
+func newSrcFS() (*fslocal.FS, error) {
+	if len(*snapshotName) == 0 {
+		return nil, fmt.Errorf("`-snapshotName` cannot be empty")
+	}
+	snapshotPath := *storageDataPath + "/snapshots/" + *snapshotName
+
+	// Verify the snapshot exists.
+	f, err := os.Open(snapshotPath)
+	if err != nil {
+		return nil, fmt.Errorf("cannot open snapshot at %q: %s", snapshotPath, err)
+	}
+	fi, err := f.Stat()
+	_ = f.Close()
+	if err != nil {
+		return nil, fmt.Errorf("cannot stat %q: %s", snapshotPath, err)
+	}
+	if !fi.IsDir() {
+		return nil, fmt.Errorf("snapshot %q must be a directory", snapshotPath)
+	}
+
+	fs := &fslocal.FS{
+		Dir:               snapshotPath,
+		MaxBytesPerSecond: *maxBytesPerSecond,
+	}
+	if err := fs.Init(); err != nil {
+		return nil, fmt.Errorf("cannot initialize fs: %s", err)
+	}
+	return fs, nil
+}
+
+func newDstFS() (common.RemoteFS, error) {
+	fs, err := actions.NewRemoteFS(*dst)
+	if err != nil {
+		return nil, fmt.Errorf("cannot parse `-dst`=%q: %s", *dst, err)
+	}
+	return fs, nil
+}
+
+func newOriginFS() (common.RemoteFS, error) {
+	if len(*origin) == 0 {
+		return nil, nil
+	}
+	fs, err := actions.NewRemoteFS(*origin)
+	if err != nil {
+		return nil, fmt.Errorf("cannot parse `-origin`=%q: %s", *origin, err)
+	}
+	return fs, nil
+}

app/vminsert/graphite/parser_test.go

@@ -85,6 +85,15 @@ func TestRowsUnmarshalSuccess(t *testing.T) {
 		}},
 	})
 
+	// Timestamp bigger than 1<<31
+	f("aaa 1123 429496729600", &Rows{
+		Rows: []Row{{
+			Metric:    "aaa",
+			Value:     1123,
+			Timestamp: 429496729600,
+		}},
+	})
+
 	// Tags
 	f("foo;bar=baz 1 2", &Rows{
 		Rows: []Row{{

app/vminsert/influx/parser.go

@@ -85,9 +85,7 @@ func (r *Row) unmarshal(s string, tagsPool []Tag, fieldsPool []Field, noEscapeCh
 		measurementTags = measurementTags[:n]
 	}
 	r.Measurement = unescapeTagValue(measurementTags, noEscapeChars)
-	if len(r.Measurement) == 0 {
-		return tagsPool, fieldsPool, fmt.Errorf("measurement cannot be empty. measurementTags=%q", s)
-	}
+	// Allow empty r.Measurement. In this case metric name is constructed directly from field keys.
 
 	// Parse fields
 	fieldsStart := len(fieldsPool)

app/vminsert/influx/parser_test.go

@@ -86,9 +86,6 @@ func TestRowsUnmarshalFailure(t *testing.T) {
 		}
 	}
 
-	// Missing measurement
-	f(",foo=bar baz=123")
-
 	// No fields
 	f("foo")
 	f("foo,bar=baz 1234")
@@ -147,6 +144,30 @@ func TestRowsUnmarshalSuccess(t *testing.T) {
 	f("#foobar baz", &Rows{})
 	f("#foobar baz\n#sss", &Rows{})
 
+	// Missing measurement
+	f(" baz=123", &Rows{
+		Rows: []Row{{
+			Measurement: "",
+			Fields: []Field{{
+				Key:   "baz",
+				Value: 123,
+			}},
+		}},
+	})
+	f(",foo=bar baz=123", &Rows{
+		Rows: []Row{{
+			Measurement: "",
+			Tags: []Tag{{
+				Key:   "foo",
+				Value: "bar",
+			}},
+			Fields: []Field{{
+				Key:   "baz",
+				Value: 123,
+			}},
+		}},
+	})
+
 	// Minimal line without tags and timestamp
 	f("foo bar=123", &Rows{
 		Rows: []Row{{

app/vminsert/influx/request_handler.go

@@ -104,7 +104,7 @@ func (ctx *pushCtx) InsertRows(db string) error {
 		ctx.metricNameBuf = storage.MarshalMetricNameRaw(ctx.metricNameBuf[:0], ic.Labels)
 		ctx.metricGroupBuf = append(ctx.metricGroupBuf[:0], r.Measurement...)
 		skipFieldKey := len(r.Fields) == 1 && *skipSingleField
-		if !skipFieldKey {
+		if len(ctx.metricGroupBuf) > 0 && !skipFieldKey {
 			ctx.metricGroupBuf = append(ctx.metricGroupBuf, *measurementFieldSeparator...)
 		}
 		metricGroupPrefixLen := len(ctx.metricGroupBuf)

app/vminsert/opentsdbhttp/server.go

@@ -38,7 +38,7 @@ func Serve(addr string, maxReqSize int64) {
 			return
 		}
 		if err != nil {
-			logger.Fatalf("FATAL: error serving HTTP OpenTSDB: %s", err)
+			logger.Fatalf("error serving HTTP OpenTSDB: %s", err)
 		}
 	}()
 }
@@ -65,6 +65,6 @@ func Stop() {
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
 	if err := httpServer.Shutdown(ctx); err != nil {
-		logger.Fatalf("FATAL: cannot close HTTP OpenTSDB server: %s", err)
+		logger.Fatalf("cannot close HTTP OpenTSDB server: %s", err)
 	}
 }

app/vmrestore/Makefile

@@ -0,0 +1,37 @@
+# All these commands must run from repository root.
+
+vmrestore:
+	APP_NAME=vmrestore $(MAKE) app-local
+
+vmrestore-prod:
+	APP_NAME=vmrestore $(MAKE) app-via-docker
+
+package-vmrestore:
+	APP_NAME=vmrestore $(MAKE) package-via-docker
+
+publish-vmrestore:
+	APP_NAME=vmrestore $(MAKE) publish-via-docker
+
+vmrestore-arm:
+	CGO_ENABLED=0 GOOS=linux GOARCH=arm GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmrestore-arm ./app/vmrestore
+
+vmrestore-arm-prod:
+	APP_NAME=vmrestore APP_SUFFIX='-arm' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=arm' $(MAKE) app-via-docker
+
+vmrestore-arm64:
+	CGO_ENABLED=0 GOOS=linux GOARCH=arm64 GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmrestore-arm64 ./app/vmrestore
+
+vmrestore-arm64-prod:
+	APP_NAME=vmrestore APP_SUFFIX='-arm64' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=arm64' $(MAKE) app-via-docker
+
+vmrestore-386:
+	CGO_ENABLED=0 GOOS=linux GOARCH=386 GO111MODULE=on go build -mod=vendor -ldflags "$(GO_BUILDINFO)" -o bin/vmrestore-386 ./app/vmrestore
+
+vmrestore-386-prod:
+	APP_NAME=vmrestore APP_SUFFIX='-386' DOCKER_OPTS='--env CGO_ENABLED=0 --env GOARCH=386' $(MAKE) app-via-docker
+
+vmrestore-pure:
+	APP_NAME=vmrestore $(MAKE) app-local-pure
+
+vmrestore-pure-prod:
+	APP_NAME=vmrestore APP_SUFFIX='-pure' DOCKER_OPTS='--env CGO_ENABLED=0' $(MAKE) app-via-docker

app/vmrestore/README.md

@@ -0,0 +1,86 @@
+## vmrestore
+
+`vmrestore` restores data from backups created by [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md).
+VictoriaMetrics `v1.29.0` and newer versions must be used for working with the restored data.
+
+Restore process can be interrupted at any time. It is automatically resumed from the inerruption point
+when restarting `vmrestore` with the same args.
+
+
+### Usage
+
+VictoriaMetrics must be stopped during the restore process.
+
+```
+vmrestore -src=gcs://<bucket>/<path/to/backup> -storageDataPath=<local/path/to/restore>
+
+```
+
+* `<bucket>` is [GCS bucket](https://cloud.google.com/storage/docs/creating-buckets) name.
+* `<path/to/backup>` is the path to backup made with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md) on GCS bucket.
+* `<local/path/to/restore>` is the path to folder where data will be restored. This folder must be passed
+  to VictoriaMetrics in `-storageDataPath` command-line flag after the restore process is complete.
+
+The original `-storageDataPath` directory may contain old files. They will be susbstituted by the files from backup.
+
+
+### Troubleshooting
+
+* If `vmrestore` eats all the network bandwidth, then set `-maxBytesPerSecond` to the desired value.
+* If `vmrestore` has been interrupted due to temporary error, then just restart it with the same args. It will resume the restore process.
+
+
+### Advanced usage
+
+Run `vmrestore -help` in order to see all the available options:
+
+```
+  -concurrency int
+    	The number of concurrent workers. Higher concurrency may reduce restore duration (default 10)
+  -configFilePath string
+    	Path to file with S3 configs. Configs are loaded from default location if not set.
+    	See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -configProfile string
+    	Profile name for S3 configs (default "default")
+  -credsFilePath string
+    	Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set.
+    	See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html
+  -customS3Endpoint string
+    	Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set
+  -loggerLevel string
+    	Minimum level of errors to log. Possible values: INFO, ERROR, FATAL, PANIC (default "INFO")
+  -maxBytesPerSecond int
+    	The maximum download speed. There is no limit if it is set to 0
+  -memory.allowedPercent float
+    	Allowed percent of system memory VictoriaMetrics caches may occupy (default 60)
+  -src string
+    	Source path with backup on the remote storage. Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir
+  -storageDataPath string
+    	Destination path where backup must be restored. VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case only missing data is downloaded from backup (default "victoria-metrics-data")
+  -version
+    	Show VictoriaMetrics version
+```
+
+
+### How to build from sources
+
+It is recommended using [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) - see `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make vmrestore` from the root folder of the repository.
+   It builds `vmrestore` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmrestore-prod` from the root folder of the repository.
+   It builds `vmrestore-prod` binary and puts it into the `bin` folder.
+
+#### Building docker images
+
+Run `make package-vmrestore`. It builds `victoriametrics/vmrestore:<PKG_TAG>` docker image locally.
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package-vmrestore`.

app/vmrestore/deployment/Dockerfile

@@ -0,0 +1,5 @@
+FROM scratch
+COPY --from=local/certs:1.0.3 /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
+COPY bin/vmrestore-prod .
+EXPOSE 8428
+ENTRYPOINT ["/vmrestore-prod"]

app/vmrestore/main.go

@@ -0,0 +1,78 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/actions"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fslocal"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/buildinfo"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+var (
+	src = flag.String("src", "", "Source path with backup on the remote storage. "+
+		"Example: gcs://bucket/path/to/backup/dir, s3://bucket/path/to/backup/dir or fs:///path/to/local/backup/dir")
+	storageDataPath = flag.String("storageDataPath", "victoria-metrics-data", "Destination path where backup must be restored. "+
+		"VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case only missing data is downloaded from backup")
+	concurrency       = flag.Int("concurrency", 10, "The number of concurrent workers. Higher concurrency may reduce restore duration")
+	maxBytesPerSecond = flag.Int("maxBytesPerSecond", 0, "The maximum download speed. There is no limit if it is set to 0")
+)
+
+func main() {
+	flag.Usage = usage
+	flag.Parse()
+	buildinfo.Init()
+
+	srcFS, err := newSrcFS()
+	if err != nil {
+		logger.Fatalf("%s", err)
+	}
+	dstFS, err := newDstFS()
+	if err != nil {
+		logger.Fatalf("%s", err)
+	}
+	a := &actions.Restore{
+		Concurrency: *concurrency,
+		Src:         srcFS,
+		Dst:         dstFS,
+	}
+	if err := a.Run(); err != nil {
+		logger.Fatalf("cannot restore from backup: %s", err)
+	}
+}
+
+func usage() {
+	const s = `
+vmrestore restores VictoriaMetrics data from backups made by vmbackup.
+
+See the docs at https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md .
+`
+
+	f := flag.CommandLine.Output()
+	fmt.Fprintf(f, "%s\n", s)
+	flag.PrintDefaults()
+}
+
+func newDstFS() (*fslocal.FS, error) {
+	if len(*storageDataPath) == 0 {
+		return nil, fmt.Errorf("`-storageDataPath` cannot be empty")
+	}
+	fs := &fslocal.FS{
+		Dir:               *storageDataPath,
+		MaxBytesPerSecond: *maxBytesPerSecond,
+	}
+	if err := fs.Init(); err != nil {
+		return nil, fmt.Errorf("cannot initialize local fs: %s", err)
+	}
+	return fs, nil
+}
+
+func newSrcFS() (common.RemoteFS, error) {
+	fs, err := actions.NewRemoteFS(*src)
+	if err != nil {
+		return nil, fmt.Errorf("cannot parse `-src`=%q: %s", *src, err)
+	}
+	return fs, nil
+}

app/vmselect/main.go

@@ -167,6 +167,18 @@ func RequestHandler(w http.ResponseWriter, r *http.Request) bool {
 			return true
 		}
 		return true
+	case "/api/v1/rules":
+		// Return dumb placeholder
+		rulesRequests.Inc()
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintf(w, "%s", `{"status":"success","data":{"groups":[]}}`)
+		return true
+	case "/api/v1/alerts":
+		// Return dumb placehloder
+		alertsRequests.Inc()
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintf(w, "%s", `{"status":"success","data":{"alerts":[]}}`)
+		return true
 	case "/api/v1/admin/tsdb/delete_series":
 		deleteRequests.Inc()
 		authKey := r.FormValue("authKey")
@@ -228,4 +240,7 @@ var (
 
 	federateRequests = metrics.NewCounter(`vm_http_requests_total{path="/federate"}`)
 	federateErrors   = metrics.NewCounter(`vm_http_request_errors_total{path="/federate"}`)
+
+	rulesRequests  = metrics.NewCounter(`vm_http_requests_total{path="/api/v1/rules"}`)
+	alertsRequests = metrics.NewCounter(`vm_http_requests_total{path="/api/v1/alerts"}`)
 )

app/vmselect/netstorage/fadvise_darwin.go

@@ -4,6 +4,6 @@ import (
 	"os"
 )
 
-func mustFadviseRandomRead(f *os.File) {
+func mustFadviseSequentialRead(f *os.File) {
 	// Do nothing :)
 }

app/vmselect/netstorage/fadvise_freebsd.go

@@ -7,9 +7,9 @@ import (
 	"golang.org/x/sys/unix"
 )
 
-func mustFadviseRandomRead(f *os.File) {
+func mustFadviseSequentialRead(f *os.File) {
 	fd := int(f.Fd())
-	if err := unix.Fadvise(int(fd), 0, 0, unix.FADV_RANDOM|unix.FADV_WILLNEED); err != nil {
-		logger.Panicf("FATAL: error returned from unix.Fadvise(RANDOM|WILLNEED): %s", err)
+	if err := unix.Fadvise(int(fd), 0, 0, unix.FADV_SEQUENTIAL|unix.FADV_WILLNEED); err != nil {
+		logger.Panicf("FATAL: error returned from unix.Fadvise(SEQUENTIAL|WILLNEED): %s", err)
 	}
 }

app/vmselect/netstorage/fadvise_linux.go

@@ -7,9 +7,9 @@ import (
 	"golang.org/x/sys/unix"
 )
 
-func mustFadviseRandomRead(f *os.File) {
+func mustFadviseSequentialRead(f *os.File) {
 	fd := int(f.Fd())
-	if err := unix.Fadvise(int(fd), 0, 0, unix.FADV_RANDOM|unix.FADV_WILLNEED); err != nil {
-		logger.Panicf("FATAL: error returned from unix.Fadvise(RANDOM|WILLNEED): %s", err)
+	if err := unix.Fadvise(int(fd), 0, 0, unix.FADV_SEQUENTIAL|unix.FADV_WILLNEED); err != nil {
+		logger.Panicf("FATAL: error returned from unix.Fadvise(SEQUENTIAL|WILLNEED): %s", err)
 	}
 }

app/vmselect/netstorage/netstorage.go

@@ -92,6 +92,7 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 	doneCh := make(chan error)
 
 	// Start workers.
+	rowsProcessedTotal := uint64(0)
 	for i := 0; i < workersCount; i++ {
 		go func(workerID uint) {
 			rs := getResult()
@@ -99,6 +100,7 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 			maxWorkersCount := gomaxprocs / workersCount
 
 			var err error
+			rowsProcessed := 0
 			for pts := range workCh {
 				if time.Until(rss.deadline.Deadline) < 0 {
 					err = fmt.Errorf("timeout exceeded during query execution: %s", rss.deadline.Timeout)
@@ -111,8 +113,10 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 					// Skip empty blocks.
 					continue
 				}
+				rowsProcessed += len(rs.Values)
 				f(rs, workerID)
 			}
+			atomic.AddUint64(&rowsProcessedTotal, uint64(rowsProcessed))
 			// Drain the remaining work
 			for range workCh {
 			}
@@ -124,6 +128,7 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 	for i := range rss.packedTimeseries {
 		workCh <- &rss.packedTimeseries[i]
 	}
+	seriesProcessedTotal := len(rss.packedTimeseries)
 	rss.packedTimeseries = rss.packedTimeseries[:0]
 	close(workCh)
 
@@ -134,6 +139,8 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 			errors = append(errors, err)
 		}
 	}
+	perQueryRowsProcessed.Update(float64(rowsProcessedTotal))
+	perQuerySeriesProcessed.Update(float64(seriesProcessedTotal))
 	if len(errors) > 0 {
 		// Return just the first error, since other errors
 		// is likely duplicate the first error.
@@ -142,6 +149,9 @@ func (rss *Results) RunParallel(f func(rs *Result, workerID uint)) error {
 	return nil
 }
 
+var perQueryRowsProcessed = metrics.NewHistogram(`vm_per_query_rows_processed_count`)
+var perQuerySeriesProcessed = metrics.NewHistogram(`vm_per_query_series_processed_count`)
+
 var gomaxprocs = runtime.GOMAXPROCS(-1)
 
 type packedTimeseries struct {
@@ -452,16 +462,12 @@ func getStorageSearch() *storage.Search {
 }
 
 func putStorageSearch(sr *storage.Search) {
-	n := atomic.LoadUint64(&sr.MissingMetricNamesForMetricID)
-	missingMetricNamesForMetricID.Add(int(n))
 	sr.MustClose()
 	ssPool.Put(sr)
 }
 
 var ssPool sync.Pool
 
-var missingMetricNamesForMetricID = metrics.NewCounter(`vm_missing_metric_names_for_metric_id_total`)
-
 // ProcessSearchQuery performs sq on storage nodes until the given deadline.
 func ProcessSearchQuery(sq *storage.SearchQuery, fetchData bool, deadline Deadline) (*Results, error) {
 	// Setup search.
@@ -484,9 +490,12 @@ func ProcessSearchQuery(sq *storage.SearchQuery, fetchData bool, deadline Deadli
 	tbf := getTmpBlocksFile()
 	m := make(map[string][]tmpBlockAddr)
 	blocksRead := 0
+	bb := tmpBufPool.Get()
+	defer tmpBufPool.Put(bb)
 	for sr.NextMetricBlock() {
 		blocksRead++
-		addr, err := tbf.WriteBlock(sr.MetricBlock.Block)
+		bb.B = storage.MarshalBlock(bb.B[:0], sr.MetricBlock.Block)
+		addr, err := tbf.WriteBlockData(bb.B)
 		if err != nil {
 			putTmpBlocksFile(tbf)
 			return nil, fmt.Errorf("cannot write data block #%d to temporary blocks file: %s", blocksRead, err)
@@ -520,6 +529,15 @@ func ProcessSearchQuery(sq *storage.SearchQuery, fetchData bool, deadline Deadli
 		pts.metricName = metricName
 		pts.addrs = addrs
 	}
+
+	// Sort rss.packedTimeseries by the first addr offset in order
+	// to reduce the number of disk seeks during unpacking in RunParallel.
+	// In this case tmpBlocksFile must be read almost sequentially.
+	sort.Slice(rss.packedTimeseries, func(i, j int) bool {
+		pts := rss.packedTimeseries
+		return pts[i].addrs[0].offset < pts[j].addrs[0].offset
+	})
+
 	return &rss, nil
 }
 

app/vmselect/netstorage/tmp_blocks_file.go

@@ -82,22 +82,18 @@ func (addr tmpBlockAddr) String() string {
 
 var tmpBlocksFilesCreated = metrics.NewCounter(`vm_tmp_blocks_files_created_total`)
 
-// WriteBlock writes b to tbf.
+// WriteBlockData writes b to tbf.
 //
 // It returns errors since the operation may fail on space shortage
 // and this must be handled.
-func (tbf *tmpBlocksFile) WriteBlock(b *storage.Block) (tmpBlockAddr, error) {
-	bb := tmpBufPool.Get()
-	defer tmpBufPool.Put(bb)
-	bb.B = storage.MarshalBlock(bb.B[:0], b)
-
+func (tbf *tmpBlocksFile) WriteBlockData(b []byte) (tmpBlockAddr, error) {
 	var addr tmpBlockAddr
 	addr.offset = tbf.offset
-	addr.size = len(bb.B)
+	addr.size = len(b)
 	tbf.offset += uint64(addr.size)
-	if len(tbf.buf)+len(bb.B) <= cap(tbf.buf) {
+	if len(tbf.buf)+len(b) <= cap(tbf.buf) {
 		// Fast path - the data fits tbf.buf
-		tbf.buf = append(tbf.buf, bb.B...)
+		tbf.buf = append(tbf.buf, b...)
 		return addr, nil
 	}
 
@@ -111,7 +107,7 @@ func (tbf *tmpBlocksFile) WriteBlock(b *storage.Block) (tmpBlockAddr, error) {
 		tmpBlocksFilesCreated.Inc()
 	}
 	_, err := tbf.f.Write(tbf.buf)
-	tbf.buf = append(tbf.buf[:0], bb.B...)
+	tbf.buf = append(tbf.buf[:0], b...)
 	if err != nil {
 		return addr, fmt.Errorf("cannot write block to %q: %s", tbf.f.Name(), err)
 	}
@@ -129,7 +125,10 @@ func (tbf *tmpBlocksFile) Finalize() error {
 	if _, err := tbf.f.Seek(0, 0); err != nil {
 		logger.Panicf("FATAL: cannot seek to the start of file: %s", err)
 	}
-	mustFadviseRandomRead(tbf.f)
+	// Hint the OS that the file is read almost sequentiallly.
+	// This should reduce the number of disk seeks, which is important
+	// for HDDs.
+	mustFadviseSequentialRead(tbf.f)
 	return nil
 }
 

app/vmselect/netstorage/tmp_blocks_file_test.go

@@ -77,9 +77,12 @@ func testTmpBlocksFile() error {
 			// Write blocks until their summary size exceeds `size`.
 			var addrs []tmpBlockAddr
 			var blocks []*storage.Block
+			bb := tmpBufPool.Get()
+			defer tmpBufPool.Put(bb)
 			for tbf.offset < uint64(size) {
 				b := createBlock()
-				addr, err := tbf.WriteBlock(b)
+				bb.B = storage.MarshalBlock(bb.B[:0], b)
+				addr, err := tbf.WriteBlockData(bb.B)
 				if err != nil {
 					return fmt.Errorf("cannot write block at offset %d: %s", tbf.offset, err)
 				}

app/vmselect/prometheus/export.qtpl

@@ -13,7 +13,7 @@
 	{% for i, ts := range rs.Timestamps %}
 		{%z= bb.B %}{% space %}
 		{%f= rs.Values[i] %}{% space %}
-		{%d= int(ts) %}{% newline %}
+		{%dl= ts %}{% newline %}
 	{% endfor %}
 	{% code quicktemplate.ReleaseByteBuffer(bb) %}
 {% endfunc %}
@@ -35,10 +35,10 @@
 		"timestamps":[
 			{% if len(rs.Timestamps) > 0 %}
 				{% code timestamps := rs.Timestamps %}
-				{%d= int(timestamps[0]) %}
+				{%dl= timestamps[0] %}
 				{% code timestamps = timestamps[1:] %}
 				{% for _, ts := range timestamps %}
-					,{%d= int(ts) %}
+					,{%dl= ts %}
 				{% endfor %}
 			{% endif %}
 		]

app/vmselect/prometheus/export.qtpl.go

@@ -49,7 +49,7 @@ func StreamExportPrometheusLine(qw422016 *qt422016.Writer, rs *netstorage.Result
 //line app/vmselect/prometheus/export.qtpl:15
 		qw422016.N().S(` `)
 //line app/vmselect/prometheus/export.qtpl:16
-		qw422016.N().D(int(ts))
+		qw422016.N().DL(ts)
 //line app/vmselect/prometheus/export.qtpl:16
 		qw422016.N().S(`
 `)
@@ -129,7 +129,7 @@ func StreamExportJSONLine(qw422016 *qt422016.Writer, rs *netstorage.Result) {
 		timestamps := rs.Timestamps
 
 //line app/vmselect/prometheus/export.qtpl:38
-		qw422016.N().D(int(timestamps[0]))
+		qw422016.N().DL(timestamps[0])
 //line app/vmselect/prometheus/export.qtpl:39
 		timestamps = timestamps[1:]
 
@@ -138,7 +138,7 @@ func StreamExportJSONLine(qw422016 *qt422016.Writer, rs *netstorage.Result) {
 //line app/vmselect/prometheus/export.qtpl:40
 			qw422016.N().S(`,`)
 //line app/vmselect/prometheus/export.qtpl:41
-			qw422016.N().D(int(ts))
+			qw422016.N().DL(ts)
 //line app/vmselect/prometheus/export.qtpl:42
 		}
 //line app/vmselect/prometheus/export.qtpl:43

app/vmselect/prometheus/federate.qtpl

@@ -10,7 +10,7 @@
 	{% if len(rs.Timestamps) == 0 || len(rs.Values) == 0 %}{% return %}{% endif %}
 	{%= prometheusMetricName(&rs.MetricName) %}{% space %}
 	{%f= rs.Values[len(rs.Values)-1] %}{% space %}
-	{%d= int(rs.Timestamps[len(rs.Timestamps)-1]) %}{% newline %}
+	{%dl= rs.Timestamps[len(rs.Timestamps)-1] %}{% newline %}
 {% endfunc %}
 
 {% endstripspace %}

app/vmselect/prometheus/federate.qtpl.go

@@ -41,7 +41,7 @@ func StreamFederate(qw422016 *qt422016.Writer, rs *netstorage.Result) {
 //line app/vmselect/prometheus/federate.qtpl:12
 	qw422016.N().S(` `)
 //line app/vmselect/prometheus/federate.qtpl:13
-	qw422016.N().D(int(rs.Timestamps[len(rs.Timestamps)-1]))
+	qw422016.N().DL(rs.Timestamps[len(rs.Timestamps)-1])
 //line app/vmselect/prometheus/federate.qtpl:13
 	qw422016.N().S(`
 `)

app/vmselect/prometheus/prometheus.go

@@ -21,17 +21,17 @@ import (
 )
 
 var (
+	latencyOffset = flag.Duration("search.latencyOffset", time.Second*30, "The time when data points become visible in query results after the colection. "+
+		"Too small value can result in incomplete last points for query results")
 	maxQueryDuration = flag.Duration("search.maxQueryDuration", time.Second*30, "The maximum time for search query execution")
 	maxQueryLen      = flag.Int("search.maxQueryLen", 16*1024, "The maximum search query length in bytes")
+	maxLookback      = flag.Duration("search.maxLookback", 0, "Synonim to `-search.lookback-delta` from Prometheus. "+
+		"The value is dynamically detected from interval between time series datapoints if not set. It can be overridden on per-query basis via `max_lookback` arg")
 )
 
 // Default step used if not set.
 const defaultStep = 5 * 60 * 1000
 
-// Latency for data processing pipeline, i.e. the time between data is ignested
-// into the system and the time it becomes visible to search.
-const latencyOffset = 60 * 1000
-
 // FederateHandler implements /federate . See https://prometheus.io/docs/prometheus/latest/federation/
 func FederateHandler(w http.ResponseWriter, r *http.Request) error {
 	startTime := time.Now()
@@ -43,11 +43,14 @@ func FederateHandler(w http.ResponseWriter, r *http.Request) error {
 	if len(matches) == 0 {
 		return fmt.Errorf("missing `match[]` arg")
 	}
-	maxLookback, err := getDuration(r, "max_lookback", defaultStep)
+	lookbackDelta, err := getMaxLookback(r)
 	if err != nil {
 		return err
 	}
-	start, err := getTime(r, "start", ct-maxLookback)
+	if lookbackDelta <= 0 {
+		lookbackDelta = defaultStep
+	}
+	start, err := getTime(r, "start", ct-lookbackDelta)
 	if err != nil {
 		return err
 	}
@@ -128,7 +131,7 @@ func ExportHandler(w http.ResponseWriter, r *http.Request) error {
 	format := r.FormValue("format")
 	deadline := getDeadline(r)
 	if start >= end {
-		start = end - defaultStep
+		end = start + defaultStep
 	}
 	if err := exportHandler(w, matches, start, end, format, deadline); err != nil {
 		return err
@@ -142,7 +145,7 @@ var exportDuration = metrics.NewSummary(`vm_request_duration_seconds{path="/api/
 func exportHandler(w http.ResponseWriter, matches []string, start, end int64, format string, deadline netstorage.Deadline) error {
 	writeResponseFunc := WriteExportStdResponse
 	writeLineFunc := WriteExportJSONLine
-	contentType := "application/json"
+	contentType := "application/stream+json"
 	if format == "prometheus" {
 		contentType = "text/plain"
 		writeLineFunc = WriteExportPrometheusLine
@@ -283,7 +286,7 @@ func labelValuesWithMatches(labelName string, matches []string, start, end int64
 		return nil, err
 	}
 	if start >= end {
-		start = end - defaultStep
+		end = start + defaultStep
 	}
 	sq := &storage.SearchQuery{
 		MinTimestamp: start,
@@ -406,7 +409,7 @@ func SeriesHandler(w http.ResponseWriter, r *http.Request) error {
 		return err
 	}
 	if start >= end {
-		start = end - defaultStep
+		end = start + defaultStep
 	}
 	sq := &storage.SearchQuery{
 		MinTimestamp: start,
@@ -463,17 +466,24 @@ func QueryHandler(w http.ResponseWriter, r *http.Request) error {
 	if err != nil {
 		return err
 	}
-	step, err := getDuration(r, "step", latencyOffset)
+	queryOffset := getLatencyOffsetMilliseconds()
+	step, err := getDuration(r, "step", queryOffset)
 	if err != nil {
 		return err
 	}
 	deadline := getDeadline(r)
+	lookbackDelta, err := getMaxLookback(r)
+	if err != nil {
+		return err
+	}
 
 	if len(query) > *maxQueryLen {
 		return fmt.Errorf(`too long query; got %d bytes; mustn't exceed %d bytes`, len(query), *maxQueryLen)
 	}
-	if ct-start < latencyOffset {
-		start -= latencyOffset
+	if !getBool(r, "nocache") && ct-start < queryOffset {
+		// Adjust start time only if `nocache` arg isn't set.
+		// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/241
+		start = ct - queryOffset
 	}
 	if childQuery, windowStr, offsetStr := promql.IsMetricSelectorWithRollup(query); childQuery != "" {
 		var window int64
@@ -503,10 +513,11 @@ func QueryHandler(w http.ResponseWriter, r *http.Request) error {
 	}
 
 	ec := promql.EvalConfig{
-		Start:    start,
-		End:      start,
-		Step:     step,
-		Deadline: deadline,
+		Start:         start,
+		End:           start,
+		Step:          step,
+		Deadline:      deadline,
+		LookbackDelta: lookbackDelta,
 	}
 	result, err := promql.Exec(&ec, query, true)
 	if err != nil {
@@ -546,13 +557,17 @@ func QueryRangeHandler(w http.ResponseWriter, r *http.Request) error {
 	}
 	deadline := getDeadline(r)
 	mayCache := !getBool(r, "nocache")
+	lookbackDelta, err := getMaxLookback(r)
+	if err != nil {
+		return err
+	}
 
 	// Validate input args.
 	if len(query) > *maxQueryLen {
 		return fmt.Errorf(`too long query; got %d bytes; mustn't exceed %d bytes`, len(query), *maxQueryLen)
 	}
 	if start > end {
-		start = end
+		end = start + defaultStep
 	}
 	if err := promql.ValidateMaxPointsPerTimeseries(start, end, step); err != nil {
 		return err
@@ -562,17 +577,19 @@ func QueryRangeHandler(w http.ResponseWriter, r *http.Request) error {
 	}
 
 	ec := promql.EvalConfig{
-		Start:    start,
-		End:      end,
-		Step:     step,
-		Deadline: deadline,
-		MayCache: mayCache,
+		Start:         start,
+		End:           end,
+		Step:          step,
+		Deadline:      deadline,
+		MayCache:      mayCache,
+		LookbackDelta: lookbackDelta,
 	}
 	result, err := promql.Exec(&ec, query, false)
 	if err != nil {
 		return fmt.Errorf("cannot execute %q: %s", query, err)
 	}
-	if ct-end < latencyOffset {
+	queryOffset := getLatencyOffsetMilliseconds()
+	if ct-end < queryOffset {
 		result = adjustLastPoints(result)
 	}
 
@@ -726,6 +743,11 @@ func getDuration(r *http.Request, argKey string, defaultValue int64) (int64, err
 
 const maxDurationMsecs = 100 * 365 * 24 * 3600 * 1000
 
+func getMaxLookback(r *http.Request) (int64, error) {
+	d := int64(*maxLookback / time.Millisecond)
+	return getDuration(r, "max_lookback", d)
+}
+
 func getDeadline(r *http.Request) netstorage.Deadline {
 	d, err := getDuration(r, "timeout", 0)
 	if err != nil {
@@ -764,3 +786,11 @@ func getTagFilterssFromMatches(matches []string) ([][]storage.TagFilter, error)
 	}
 	return tagFilterss, nil
 }
+
+func getLatencyOffsetMilliseconds() int64 {
+	d := int64(*latencyOffset / time.Millisecond)
+	if d <= 1000 {
+		d = 1000
+	}
+	return d
+}

app/vmselect/prometheus/series_count_response.qtpl

@@ -3,7 +3,7 @@ SeriesCountResponse generates response for /api/v1/series/count .
 {% func SeriesCountResponse(n uint64) %}
 {
 	"status":"success",
-	"data":[{%d int(n) %}]
+	"data":[{%dl int64(n) %}]
 }
 {% endfunc %}
 {% endstripspace %}

app/vmselect/prometheus/series_count_response.qtpl.go

@@ -24,7 +24,7 @@ func StreamSeriesCountResponse(qw422016 *qt422016.Writer, n uint64) {
 //line app/vmselect/prometheus/series_count_response.qtpl:3
 	qw422016.N().S(`{"status":"success","data":[`)
 //line app/vmselect/prometheus/series_count_response.qtpl:6
-	qw422016.N().D(int(n))
+	qw422016.N().DL(int64(n))
 //line app/vmselect/prometheus/series_count_response.qtpl:6
 	qw422016.N().S(`]}`)
 //line app/vmselect/prometheus/series_count_response.qtpl:8

app/vmselect/promql/aggr.go

@@ -9,6 +9,8 @@ import (
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
+	"github.com/VictoriaMetrics/metrics"
+	"github.com/valyala/histogram"
 )
 
 var aggrFuncs = map[string]aggrFunc{
@@ -26,11 +28,20 @@ var aggrFuncs = map[string]aggrFunc{
 	"quantile":     aggrFuncQuantile,
 
 	// Extended PromQL funcs
-	"median":   aggrFuncMedian,
-	"limitk":   aggrFuncLimitK,
-	"distinct": newAggrFunc(aggrFuncDistinct),
-	"sum2":     newAggrFunc(aggrFuncSum2),
-	"geomean":  newAggrFunc(aggrFuncGeomean),
+	"median":         aggrFuncMedian,
+	"limitk":         aggrFuncLimitK,
+	"distinct":       newAggrFunc(aggrFuncDistinct),
+	"sum2":           newAggrFunc(aggrFuncSum2),
+	"geomean":        newAggrFunc(aggrFuncGeomean),
+	"histogram":      newAggrFunc(aggrFuncHistogram),
+	"topk_min":       newAggrFuncRangeTopK(minValue, false),
+	"topk_max":       newAggrFuncRangeTopK(maxValue, false),
+	"topk_avg":       newAggrFuncRangeTopK(avgValue, false),
+	"topk_median":    newAggrFuncRangeTopK(medianValue, false),
+	"bottomk_min":    newAggrFuncRangeTopK(minValue, true),
+	"bottomk_max":    newAggrFuncRangeTopK(maxValue, true),
+	"bottomk_avg":    newAggrFuncRangeTopK(avgValue, true),
+	"bottomk_median": newAggrFuncRangeTopK(medianValue, true),
 }
 
 type aggrFunc func(afa *aggrFuncArg) ([]*timeseries, error)
@@ -184,6 +195,38 @@ func aggrFuncGeomean(tss []*timeseries) []*timeseries {
 	return tss[:1]
 }
 
+func aggrFuncHistogram(tss []*timeseries) []*timeseries {
+	var h metrics.Histogram
+	m := make(map[string]*timeseries)
+	for i := range tss[0].Values {
+		h.Reset()
+		for _, ts := range tss {
+			v := ts.Values[i]
+			h.Update(v)
+		}
+		h.VisitNonZeroBuckets(func(vmrange string, count uint64) {
+			ts := m[vmrange]
+			if ts == nil {
+				ts = &timeseries{}
+				ts.CopyFromShallowTimestamps(tss[0])
+				ts.MetricName.RemoveTag("vmrange")
+				ts.MetricName.AddTag("vmrange", vmrange)
+				values := ts.Values
+				for k := range values {
+					values[k] = 0
+				}
+				m[vmrange] = ts
+			}
+			ts.Values[i] = float64(count)
+		})
+	}
+	rvs := make([]*timeseries, 0, len(m))
+	for _, ts := range m {
+		rvs = append(rvs, ts)
+	}
+	return vmrangeBucketsToLE(rvs)
+}
+
 func aggrFuncMin(tss []*timeseries) []*timeseries {
 	if len(tss) == 1 {
 		// Fast path - nothing to min.
@@ -425,37 +468,138 @@ func newAggrFuncTopK(isReverse bool) aggrFunc {
 			return nil, err
 		}
 		afe := func(tss []*timeseries) []*timeseries {
-			rvs := tss
-			for n := range rvs[0].Values {
-				sort.Slice(rvs, func(i, j int) bool {
-					a := rvs[i].Values[n]
-					b := rvs[j].Values[n]
-					cmp := lessWithNaNs(a, b)
+			for n := range tss[0].Values {
+				sort.Slice(tss, func(i, j int) bool {
+					a := tss[i].Values[n]
+					b := tss[j].Values[n]
 					if isReverse {
-						cmp = !cmp
+						a, b = b, a
 					}
-					return cmp
+					return lessWithNaNs(a, b)
 				})
-				if math.IsNaN(ks[n]) {
-					ks[n] = 0
-				}
-				k := int(ks[n])
-				if k < 0 {
-					k = 0
-				}
-				if k > len(rvs) {
-					k = len(rvs)
-				}
-				for _, ts := range rvs[:len(rvs)-k] {
-					ts.Values[n] = nan
-				}
+				fillNaNsAtIdx(n, ks[n], tss)
 			}
-			return removeNaNs(rvs)
+			return removeNaNs(tss)
 		}
 		return aggrFuncExt(afe, args[1], &afa.ae.Modifier, true)
 	}
 }
 
+type tsWithValue struct {
+	ts    *timeseries
+	value float64
+}
+
+func newAggrFuncRangeTopK(f func(values []float64) float64, isReverse bool) aggrFunc {
+	return func(afa *aggrFuncArg) ([]*timeseries, error) {
+		args := afa.args
+		if err := expectTransformArgsNum(args, 2); err != nil {
+			return nil, err
+		}
+		ks, err := getScalar(args[0], 0)
+		if err != nil {
+			return nil, err
+		}
+		afe := func(tss []*timeseries) []*timeseries {
+			maxs := make([]tsWithValue, len(tss))
+			for i, ts := range tss {
+				value := f(ts.Values)
+				maxs[i] = tsWithValue{
+					ts:    ts,
+					value: value,
+				}
+			}
+			sort.Slice(maxs, func(i, j int) bool {
+				a := maxs[i].value
+				b := maxs[j].value
+				if isReverse {
+					a, b = b, a
+				}
+				return lessWithNaNs(a, b)
+			})
+			for i := range maxs {
+				tss[i] = maxs[i].ts
+			}
+			for i, k := range ks {
+				fillNaNsAtIdx(i, k, tss)
+			}
+			return removeNaNs(tss)
+		}
+		return aggrFuncExt(afe, args[1], &afa.ae.Modifier, true)
+	}
+}
+
+func fillNaNsAtIdx(idx int, k float64, tss []*timeseries) {
+	if math.IsNaN(k) {
+		k = 0
+	}
+	kn := int(k)
+	if kn < 0 {
+		kn = 0
+	}
+	if kn > len(tss) {
+		kn = len(tss)
+	}
+	for _, ts := range tss[:len(tss)-kn] {
+		ts.Values[idx] = nan
+	}
+}
+
+func minValue(values []float64) float64 {
+	if len(values) == 0 {
+		return nan
+	}
+	min := values[0]
+	for _, v := range values[1:] {
+		if v < min {
+			min = v
+		}
+	}
+	return min
+}
+
+func maxValue(values []float64) float64 {
+	if len(values) == 0 {
+		return nan
+	}
+	max := values[0]
+	for _, v := range values[1:] {
+		if v > max {
+			max = v
+		}
+	}
+	return max
+}
+
+func avgValue(values []float64) float64 {
+	sum := float64(0)
+	count := 0
+	for _, v := range values {
+		if math.IsNaN(v) {
+			continue
+		}
+		count++
+		sum += v
+	}
+	if count == 0 {
+		return nan
+	}
+	return sum / float64(count)
+}
+
+func medianValue(values []float64) float64 {
+	h := histogram.GetFast()
+	for _, v := range values {
+		if math.IsNaN(v) {
+			continue
+		}
+		h.Update(v)
+	}
+	value := h.Quantile(0.5)
+	histogram.PutFast(h)
+	return value
+}
+
 func aggrFuncLimitK(afa *aggrFuncArg) ([]*timeseries, error) {
 	args := afa.args
 	if err := expectTransformArgsNum(args, 2); err != nil {

app/vmselect/promql/arch.go

@@ -0,0 +1,5 @@
+package promql
+
+import "unsafe"
+
+const maxByteSliceLen = 1<<(31+9*(unsafe.Sizeof(int(0))/8)) - 1

app/vmselect/promql/arch_amd64.go

@@ -1,3 +0,0 @@
-package promql
-
-const maxByteSliceLen = 1 << 40

app/vmselect/promql/arch_arm.go

@@ -1,3 +0,0 @@
-package promql
-
-const maxByteSliceLen = 1<<31 - 1

app/vmselect/promql/arch_arm64.go

@@ -1,3 +0,0 @@
-package promql
-
-const maxByteSliceLen = 1 << 40

app/vmselect/promql/binary_op.go

@@ -292,24 +292,14 @@ func adjustBinaryOpTags(be *binaryOpExpr, left, right []*timeseries) ([]*timeser
 	}
 
 	// Slow path: `vector op vector` or `a op {on|ignoring} {group_left|group_right} b`
-	ensureOneX := func(side string, tss []*timeseries) error {
-		if len(tss) == 0 {
-			logger.Panicf("BUG: tss must contain at least one value")
-		}
-		if len(tss) == 1 {
-			return nil
-		}
-		if mergeNonOverlappingTimeseries(tss) {
-			return nil
-		}
-		return fmt.Errorf(`duplicate timeseries on the %s side of %s %s: %s and %s`, side, be.Op, be.GroupModifier.AppendString(nil),
-			stringMetricTags(&tss[0].MetricName), stringMetricTags(&tss[1].MetricName))
-	}
-
 	var rvsLeft, rvsRight []*timeseries
 	mLeft, mRight := createTimeseriesMapByTagSet(be, left, right)
 	joinOp := strings.ToLower(be.JoinModifier.Op)
-	joinTags := be.JoinModifier.Args
+	groupOp := strings.ToLower(be.GroupModifier.Op)
+	if len(groupOp) == 0 {
+		groupOp = "ignoring"
+	}
+	groupTags := be.GroupModifier.Args
 	for k, tssLeft := range mLeft {
 		tssRight := mRight[k]
 		if len(tssRight) == 0 {
@@ -317,39 +307,38 @@ func adjustBinaryOpTags(be *binaryOpExpr, left, right []*timeseries) ([]*timeser
 		}
 		switch joinOp {
 		case "group_left":
-			if err := ensureOneX("right", tssRight); err != nil {
+			var err error
+			rvsLeft, rvsRight, err = groupJoin("right", be, rvsLeft, rvsRight, tssLeft, tssRight)
+			if err != nil {
 				return nil, nil, nil, err
 			}
-			src := tssRight[0]
-			for _, ts := range tssLeft {
-				resetMetricGroupIfRequired(be, ts)
-				ts.MetricName.AddMissingTags(joinTags, &src.MetricName)
-				rvsLeft = append(rvsLeft, ts)
-				rvsRight = append(rvsRight, src)
-			}
 		case "group_right":
-			if err := ensureOneX("left", tssLeft); err != nil {
+			var err error
+			rvsRight, rvsLeft, err = groupJoin("left", be, rvsRight, rvsLeft, tssRight, tssLeft)
+			if err != nil {
 				return nil, nil, nil, err
 			}
-			src := tssLeft[0]
-			for _, ts := range tssRight {
-				resetMetricGroupIfRequired(be, ts)
-				ts.MetricName.AddMissingTags(joinTags, &src.MetricName)
-				rvsLeft = append(rvsLeft, src)
-				rvsRight = append(rvsRight, ts)
-			}
 		case "":
-			if err := ensureOneX("left", tssLeft); err != nil {
+			if err := ensureSingleTimeseries("left", be, tssLeft); err != nil {
 				return nil, nil, nil, err
 			}
-			if err := ensureOneX("right", tssRight); err != nil {
+			if err := ensureSingleTimeseries("right", be, tssRight); err != nil {
 				return nil, nil, nil, err
 			}
-			resetMetricGroupIfRequired(be, tssLeft[0])
-			rvsLeft = append(rvsLeft, tssLeft[0])
+			tsLeft := tssLeft[0]
+			resetMetricGroupIfRequired(be, tsLeft)
+			switch groupOp {
+			case "on":
+				tsLeft.MetricName.RemoveTagsOn(groupTags)
+			case "ignoring":
+				tsLeft.MetricName.RemoveTagsIgnoring(groupTags)
+			default:
+				logger.Panicf("BUG: unexpected binary op modifier %q", groupOp)
+			}
+			rvsLeft = append(rvsLeft, tsLeft)
 			rvsRight = append(rvsRight, tssRight[0])
 		default:
-			return nil, nil, nil, fmt.Errorf(`unexpected join modifier %q`, joinOp)
+			logger.Panicf("BUG: unexpected join modifier %q", joinOp)
 		}
 	}
 	dst := rvsLeft
@@ -359,6 +348,90 @@ func adjustBinaryOpTags(be *binaryOpExpr, left, right []*timeseries) ([]*timeser
 	return rvsLeft, rvsRight, dst, nil
 }
 
+func ensureSingleTimeseries(side string, be *binaryOpExpr, tss []*timeseries) error {
+	if len(tss) == 0 {
+		logger.Panicf("BUG: tss must contain at least one value")
+	}
+	for len(tss) > 1 {
+		if !mergeNonOverlappingTimeseries(tss[0], tss[len(tss)-1]) {
+			return fmt.Errorf(`duplicate time series on the %s side of %s %s: %s and %s`, side, be.Op, be.GroupModifier.AppendString(nil),
+				stringMetricTags(&tss[0].MetricName), stringMetricTags(&tss[len(tss)-1].MetricName))
+		}
+		tss = tss[:len(tss)-1]
+	}
+	return nil
+}
+
+func groupJoin(singleTimeseriesSide string, be *binaryOpExpr, rvsLeft, rvsRight, tssLeft, tssRight []*timeseries) ([]*timeseries, []*timeseries, error) {
+	joinTags := be.JoinModifier.Args
+	var m map[string]*timeseries
+	for _, tsLeft := range tssLeft {
+		resetMetricGroupIfRequired(be, tsLeft)
+		if len(tssRight) == 1 {
+			// Easy case - right part contains only a single matching time series.
+			tsLeft.MetricName.AddMissingTags(joinTags, &tssRight[0].MetricName)
+			rvsLeft = append(rvsLeft, tsLeft)
+			rvsRight = append(rvsRight, tssRight[0])
+			continue
+		}
+
+		// Hard case - right part contains multiple matching time series.
+		// Verify it doesn't result in duplicate MetricName values after adding missing tags.
+		if m == nil {
+			m = make(map[string]*timeseries, len(tssRight))
+		} else {
+			for k := range m {
+				delete(m, k)
+			}
+		}
+		bb := bbPool.Get()
+		for _, tsRight := range tssRight {
+			var tsCopy timeseries
+			tsCopy.CopyFromShallowTimestamps(tsLeft)
+			tsCopy.MetricName.AddMissingTags(joinTags, &tsRight.MetricName)
+			bb.B = marshalMetricTagsSorted(bb.B[:0], &tsCopy.MetricName)
+			if tsExisting := m[string(bb.B)]; tsExisting != nil {
+				// Try merging tsExisting with tsRight if they don't overlap.
+				if mergeNonOverlappingTimeseries(tsExisting, tsRight) {
+					continue
+				}
+				return nil, nil, fmt.Errorf("duplicate time series on the %s side of `%s %s %s`: %s and %s",
+					singleTimeseriesSide, be.Op, be.GroupModifier.AppendString(nil), be.JoinModifier.AppendString(nil),
+					stringMetricTags(&tsExisting.MetricName), stringMetricTags(&tsRight.MetricName))
+			}
+			m[string(bb.B)] = tsRight
+			rvsLeft = append(rvsLeft, &tsCopy)
+			rvsRight = append(rvsRight, tsRight)
+		}
+		bbPool.Put(bb)
+	}
+	return rvsLeft, rvsRight, nil
+}
+
+func mergeNonOverlappingTimeseries(dst, src *timeseries) bool {
+	// Verify whether the time series can be merged.
+	srcValues := src.Values
+	dstValues := dst.Values
+	_ = dstValues[len(srcValues)-1]
+	for i, v := range srcValues {
+		if math.IsNaN(v) {
+			continue
+		}
+		if !math.IsNaN(dstValues[i]) {
+			return false
+		}
+	}
+
+	// Time series can be merged. Merge them.
+	for i, v := range srcValues {
+		if math.IsNaN(v) {
+			continue
+		}
+		dstValues[i] = v
+	}
+	return true
+}
+
 func resetMetricGroupIfRequired(be *binaryOpExpr, ts *timeseries) {
 	if isBinaryOpCmp(be.Op) && !be.Bool {
 		// Do not reset MetricGroup for non-boolean `compare` binary ops like Prometheus does.
@@ -535,26 +608,3 @@ func isScalar(arg []*timeseries) bool {
 	}
 	return len(mn.Tags) == 0
 }
-
-func mergeNonOverlappingTimeseries(tss []*timeseries) bool {
-	if len(tss) < 2 {
-		logger.Panicf("BUG: expecting at least two timeseries. Got %d", len(tss))
-	}
-
-	// Check whether time series in tss overlap.
-	var dst timeseries
-	dst.CopyFromShallowTimestamps(tss[0])
-	dstValues := dst.Values
-	for _, ts := range tss[1:] {
-		for i, value := range ts.Values {
-			if math.IsNaN(dstValues[i]) {
-				dstValues[i] = value
-			} else if !math.IsNaN(value) {
-				// Time series overlap.
-				return false
-			}
-		}
-	}
-	tss[0].CopyFromShallowTimestamps(&dst)
-	return true
-}

app/vmselect/promql/eval.go

@@ -70,6 +70,9 @@ type EvalConfig struct {
 
 	MayCache bool
 
+	// LookbackDelta is analog to `-query.lookback-delta` from Prometheus.
+	LookbackDelta int64
+
 	timestamps     []int64
 	timestampsOnce sync.Once
 }
@@ -82,6 +85,7 @@ func newEvalConfig(src *EvalConfig) *EvalConfig {
 	ec.Step = src.Step
 	ec.Deadline = src.Deadline
 	ec.MayCache = src.MayCache
+	ec.LookbackDelta = src.LookbackDelta
 
 	// do not copy src.timestamps - they must be generated again.
 	return &ec
@@ -290,10 +294,10 @@ func tryGetArgRollupFuncWithMetricExpr(ae *aggrFuncExpr) (*funcExpr, newRollupFu
 		return fe, nrf
 	}
 	if re, ok := e.(*rollupExpr); ok {
-		if me, ok := re.Expr.(*metricExpr); !ok || me.IsEmpty() {
+		if me, ok := re.Expr.(*metricExpr); !ok || me.IsEmpty() || re.ForSubquery() {
 			return nil, nil
 		}
-		// e = rollupExpr(metricExpr)
+		// e = metricExpr[d]
 		fe := &funcExpr{
 			Name: "default_rollup",
 			Args: []expr{re},
@@ -315,15 +319,17 @@ func tryGetArgRollupFuncWithMetricExpr(ae *aggrFuncExpr) (*funcExpr, newRollupFu
 		if me.IsEmpty() {
 			return nil, nil
 		}
+		// e = rollupFunc(metricExpr)
 		return &funcExpr{
 			Name: fe.Name,
 			Args: []expr{me},
 		}, nrf
 	}
 	if re, ok := arg.(*rollupExpr); ok {
-		if me, ok := re.Expr.(*metricExpr); !ok || me.IsEmpty() {
+		if me, ok := re.Expr.(*metricExpr); !ok || me.IsEmpty() || re.ForSubquery() {
 			return nil, nil
 		}
+		// e = rollupFunc(metricExpr[d])
 		return fe, nrf
 	}
 	return nil, nil
@@ -368,8 +374,8 @@ func getRollupExprArg(arg expr) *rollupExpr {
 			Expr: arg,
 		}
 	}
-	if len(re.Step) == 0 && !re.InheritStep {
-		// Return standard rollup if it doesn't set step.
+	if !re.ForSubquery() {
+		// Return standard rollup if it doesn't contain subquery.
 		return re
 	}
 	me, ok := re.Expr.(*metricExpr)
@@ -463,7 +469,7 @@ func evalRollupFuncWithSubquery(ec *EvalConfig, name string, rf rollupFunc, re *
 	}
 
 	sharedTimestamps := getTimestamps(ec.Start, ec.End, ec.Step)
-	preFunc, rcs := getRollupConfigs(name, rf, ec.Start, ec.End, ec.Step, window, sharedTimestamps)
+	preFunc, rcs := getRollupConfigs(name, rf, ec.Start, ec.End, ec.Step, window, ec.LookbackDelta, sharedTimestamps)
 	tss := make([]*timeseries, 0, len(tssSQ)*len(rcs))
 	var tssLock sync.Mutex
 	removeMetricGroup := !rollupFuncsKeepMetricGroup[name]
@@ -584,12 +590,23 @@ func evalRollupFuncWithMetricExpr(ec *EvalConfig, name string, rf rollupFunc, me
 		return tss, nil
 	}
 	sharedTimestamps := getTimestamps(start, ec.End, ec.Step)
-	preFunc, rcs := getRollupConfigs(name, rf, start, ec.End, ec.Step, window, sharedTimestamps)
+	preFunc, rcs := getRollupConfigs(name, rf, start, ec.End, ec.Step, window, ec.LookbackDelta, sharedTimestamps)
 
 	// Verify timeseries fit available memory after the rollup.
 	// Take into account points from tssCached.
 	pointsPerTimeseries := 1 + (ec.End-ec.Start)/ec.Step
-	rollupPoints := mulNoOverflow(pointsPerTimeseries, int64(rssLen*len(rcs)))
+	timeseriesLen := rssLen
+	if iafc != nil {
+		// Incremental aggregates require hold only GOMAXPROCS timeseries in memory.
+		timeseriesLen = runtime.GOMAXPROCS(-1)
+		if iafc.ae.Modifier.Op != "" {
+			// Increase the number of timeseries for non-empty group list: `aggr() by (something)`,
+			// since each group can have own set of time series in memory.
+			// Estimate the number of such groups is lower than 100 :)
+			timeseriesLen *= 100
+		}
+	}
+	rollupPoints := mulNoOverflow(pointsPerTimeseries, int64(timeseriesLen*len(rcs)))
 	rollupMemorySize := mulNoOverflow(rollupPoints, 16)
 	rml := getRollupMemoryLimiter()
 	if !rml.Get(uint64(rollupMemorySize)) {
@@ -687,7 +704,8 @@ func doRollupForTimeseries(rc *rollupConfig, tsDst *timeseries, mnSrc *storage.M
 	tsDst.denyReuse = true
 }
 
-func getRollupConfigs(name string, rf rollupFunc, start, end, step, window int64, sharedTimestamps []int64) (func(values []float64, timestamps []int64), []*rollupConfig) {
+func getRollupConfigs(name string, rf rollupFunc, start, end, step, window int64, lookbackDelta int64, sharedTimestamps []int64) (
+	func(values []float64, timestamps []int64), []*rollupConfig) {
 	preFunc := func(values []float64, timestamps []int64) {}
 	if rollupFuncsRemoveCounterResets[name] {
 		preFunc = func(values []float64, timestamps []int64) {
@@ -703,6 +721,7 @@ func getRollupConfigs(name string, rf rollupFunc, start, end, step, window int64
 			Step:            step,
 			Window:          window,
 			MayAdjustWindow: rollupFuncsMayAdjustWindow[name],
+			LookbackDelta:   lookbackDelta,
 			Timestamps:      sharedTimestamps,
 		}
 	}

app/vmselect/promql/exec.go

@@ -110,7 +110,7 @@ func timeseriesToResult(tss []*timeseries, maySort bool) ([]netstorage.Result, e
 	for i, ts := range tss {
 		bb.B = marshalMetricNameSorted(bb.B[:0], &ts.MetricName)
 		if _, ok := m[string(bb.B)]; ok {
-			return nil, fmt.Errorf(`duplicate output timeseries: %s%s`, ts.MetricName.MetricGroup, stringMetricName(&ts.MetricName))
+			return nil, fmt.Errorf(`duplicate output timeseries: %s`, stringMetricName(&ts.MetricName))
 		}
 		m[string(bb.B)] = struct{}{}
 
@@ -194,11 +194,14 @@ type parseCacheValue struct {
 }
 
 type parseCache struct {
-	m  map[string]*parseCacheValue
-	mu sync.RWMutex
+	// Move atomic counters to the top of struct for 8-byte alignment on 32-bit arch.
+	// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/212
 
 	requests uint64
 	misses   uint64
+
+	m  map[string]*parseCacheValue
+	mu sync.RWMutex
 }
 
 func (pc *parseCache) Requests() uint64 {

app/vmselect/promql/exec_test.go

@@ -369,6 +369,17 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run("timestamp(time()>=1600)", func(t *testing.T) {
+		t.Parallel()
+		q := `timestamp(time()>=1600)`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, nan, 1600, 1800, 2000},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
 	t.Run("time()/100", func(t *testing.T) {
 		t.Parallel()
 		q := `time()/100`
@@ -1826,10 +1837,6 @@ func TestExecSuccess(t *testing.T) {
 			Timestamps: timestampsExpected,
 		}
 		r.MetricName.Tags = []storage.Tag{
-			{
-				Key:   []byte("aa"),
-				Value: []byte("bb"),
-			},
 			{
 				Key:   []byte("foo"),
 				Value: []byte("bar"),
@@ -1851,12 +1858,75 @@ func TestExecSuccess(t *testing.T) {
 				Key:   []byte("foo"),
 				Value: []byte("bar"),
 			},
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
+	t.Run(`vector * on(foo) group_left(additional_tag) duplicate_timeseries_differ_by_additional_tag`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(label_set(time()/10, "foo", "bar", "xx", "yy", "__name__", "qwert") + on(foo) group_left(op) (
+			label_set(time() < 1400, "foo", "bar", "op", "le"),
+			label_set(time() >= 1400, "foo", "bar", "op", "ge"),
+		))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1100, 1320, nan, nan, nan, nan},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("op"),
+				Value: []byte("le"),
+			},
 			{
 				Key:   []byte("xx"),
 				Value: []byte("yy"),
 			},
 		}
-		resultExpected := []netstorage.Result{r}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, 1540, 1760, 1980, 2200},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("op"),
+				Value: []byte("ge"),
+			},
+			{
+				Key:   []byte("xx"),
+				Value: []byte("yy"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1, r2}
+		f(q, resultExpected)
+	})
+	t.Run(`vector * on(foo) duplicate_nonoverlapping_timeseries`, func(t *testing.T) {
+		t.Parallel()
+		q := `label_set(time()/10, "foo", "bar", "xx", "yy", "__name__", "qwert") + on(foo) (
+			label_set(time() < 1400, "foo", "bar", "op", "le"),
+			label_set(time() >= 1400, "foo", "bar", "op", "ge"),
+		)`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1100, 1320, 1540, 1760, 1980, 2200},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1}
 		f(q, resultExpected)
 	})
 	t.Run(`vector * on(foo) group_left() duplicate_nonoverlapping_timeseries`, func(t *testing.T) {
@@ -2043,10 +2113,6 @@ func TestExecSuccess(t *testing.T) {
 			Timestamps: timestampsExpected,
 		}
 		r.MetricName.Tags = []storage.Tag{
-			{
-				Key:   []byte("t1"),
-				Value: []byte("v123"),
-			},
 			{
 				Key:   []byte("t2"),
 				Value: []byte("v3"),
@@ -2152,10 +2218,6 @@ func TestExecSuccess(t *testing.T) {
 			Timestamps: timestampsExpected,
 		}
 		r.MetricName.Tags = []storage.Tag{
-			{
-				Key:   []byte("t1"),
-				Value: []byte("v123"),
-			},
 			{
 				Key:   []byte("t2"),
 				Value: []byte("v3"),
@@ -2227,6 +2289,45 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`histogram_quantile(single-value-valid-le-max-phi)`, func(t *testing.T) {
+		t.Parallel()
+		q := `histogram_quantile(1, (
+			label_set(100, "le", "200"),
+			label_set(0, "le", "55"),
+		))`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{200, 200, 200, 200, 200, 200},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
+	t.Run(`histogram_quantile(single-value-valid-le-min-phi)`, func(t *testing.T) {
+		t.Parallel()
+		q := `histogram_quantile(0, (
+			label_set(100, "le", "200"),
+			label_set(0, "le", "55"),
+		))`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{55, 55, 55, 55, 55, 55},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
+	t.Run(`histogram_quantile(single-value-valid-le-min-phi-no-zero-bucket)`, func(t *testing.T) {
+		t.Parallel()
+		q := `histogram_quantile(0, label_set(100, "le", "200"))`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{0, 0, 0, 0, 0, 0},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
 	t.Run(`histogram_quantile(scalar-phi)`, func(t *testing.T) {
 		t.Parallel()
 		q := `histogram_quantile(time() / 2 / 1e3, label_set(100, "le", "200"))`
@@ -2287,7 +2388,7 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
-	t.Run(`histogram_quantile(nan-bucket-count)`, func(t *testing.T) {
+	t.Run(`histogram_quantile(nan-bucket-count-some)`, func(t *testing.T) {
 		t.Parallel()
 		q := `histogram_quantile(0.6,
 			label_set(90, "foo", "bar", "le", "10")
@@ -2296,7 +2397,7 @@ func TestExecSuccess(t *testing.T) {
 		)`
 		r := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{30, 30, 30, 30, 30, 30},
+			Values:     []float64{10, 10, 10, 10, 10, 10},
 			Timestamps: timestampsExpected,
 		}
 		r.MetricName.Tags = []storage.Tag{{
@@ -2306,7 +2407,7 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
-	t.Run(`histogram_quantile(nan-bucket-count)`, func(t *testing.T) {
+	t.Run(`histogram_quantile(normal-bucket-count)`, func(t *testing.T) {
 		t.Parallel()
 		q := `histogram_quantile(0.2,
 			label_set(0, "foo", "bar", "le", "10")
@@ -2335,7 +2436,7 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{}
 		f(q, resultExpected)
 	})
-	t.Run(`histogram_quantile(nan-bucket-count)`, func(t *testing.T) {
+	t.Run(`histogram_quantile(nan-bucket-count-all)`, func(t *testing.T) {
 		t.Parallel()
 		q := `histogram_quantile(0.6,
 			label_set(nan, "foo", "bar", "le", "10")
@@ -2345,6 +2446,190 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{}
 		f(q, resultExpected)
 	})
+	t.Run(`prometheus_buckets(missing-vmrange)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(prometheus_buckets((
+			alias(label_set(time()/20, "foo", "bar", "le", "0.2"), "xyz"),
+			alias(label_set(time()/100, "foo", "bar", "vmrange", "foobar"), "xxx"),
+			alias(label_set(time()/100, "foo", "bar", "vmrange", "30...foobar"), "xxx"),
+			alias(label_set(time()/100, "foo", "bar", "vmrange", "30...40"), "xxx"),
+			alias(label_set(time()/80, "foo", "bar", "vmrange", "0...900", "le", "54"), "yyy"),
+			alias(label_set(time()/40, "foo", "bar", "vmrange", "900...+Inf", "le", "2343"), "yyy"),
+		)))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{0, 0, 0, 0, 0, 0},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.MetricGroup = []byte("xxx")
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("30"),
+			},
+		}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 12, 14, 16, 18, 20},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.MetricGroup = []byte("xxx")
+		r2.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("40"),
+			},
+		}
+		r3 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 12, 14, 16, 18, 20},
+			Timestamps: timestampsExpected,
+		}
+		r3.MetricName.MetricGroup = []byte("xxx")
+		r3.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("+Inf"),
+			},
+		}
+		r4 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{12.5, 15, 17.5, 20, 22.5, 25},
+			Timestamps: timestampsExpected,
+		}
+		r4.MetricName.MetricGroup = []byte("yyy")
+		r4.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("900"),
+			},
+		}
+		r5 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{37.5, 45, 52.5, 60, 67.5, 75},
+			Timestamps: timestampsExpected,
+		}
+		r5.MetricName.MetricGroup = []byte("yyy")
+		r5.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("+Inf"),
+			},
+		}
+		r6 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{50, 60, 70, 80, 90, 100},
+			Timestamps: timestampsExpected,
+		}
+		r6.MetricName.MetricGroup = []byte("xyz")
+		r6.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("0.2"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1, r2, r3, r4, r5, r6}
+		f(q, resultExpected)
+	})
+	t.Run(`prometheus_buckets(valid)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(prometheus_buckets((
+			alias(label_set(90, "foo", "bar", "vmrange", "0...0"), "xxx"),
+			alias(label_set(time()/20, "foo", "bar", "vmrange", "0...0.2"), "xxx"),
+			alias(label_set(time()/100, "foo", "bar", "vmrange", "0.2...40"), "xxx"),
+			alias(label_set(time()/10, "foo", "bar", "vmrange", "40...Inf"), "xxx"),
+		)))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{90, 90, 90, 90, 90, 90},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.MetricGroup = []byte("xxx")
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("0"),
+			},
+		}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{140, 150, 160, 170, 180, 190},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.MetricGroup = []byte("xxx")
+		r2.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("0.2"),
+			},
+		}
+		r3 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{150, 162, 174, 186, 198, 210},
+			Timestamps: timestampsExpected,
+		}
+		r3.MetricName.MetricGroup = []byte("xxx")
+		r3.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("40"),
+			},
+		}
+		r4 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{250, 282, 314, 346, 378, 410},
+			Timestamps: timestampsExpected,
+		}
+		r4.MetricName.MetricGroup = []byte("xxx")
+		r4.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("foo"),
+				Value: []byte("bar"),
+			},
+			{
+				Key:   []byte("le"),
+				Value: []byte("Inf"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1, r2, r3, r4}
+		f(q, resultExpected)
+	})
 	t.Run(`median_over_time()`, func(t *testing.T) {
 		t.Parallel()
 		q := `median_over_time({})`
@@ -2395,6 +2680,108 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`histogram(scalar)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(histogram(123)+(
+			label_set(0, "le", "1.0e2"),
+			label_set(0, "le", "1.5e2"),
+			label_set(1, "le", "+Inf"),
+		))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{0, 0, 0, 0, 0, 0},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("1.0e2"),
+			},
+		}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1, 1, 1, 1, 1, 1},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("1.5e2"),
+			},
+		}
+		r3 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{2, 2, 2, 2, 2, 2},
+			Timestamps: timestampsExpected,
+		}
+		r3.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("+Inf"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1, r2, r3}
+		f(q, resultExpected)
+	})
+	t.Run(`histogram(vector)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(histogram((
+			label_set(1, "foo", "bar"),
+			label_set(1.1, "xx", "yy"),
+			alias(1.15, "foobar"),
+		))+(
+			label_set(0, "le", "9.5e-1"),
+			label_set(0, "le", "1.0e0"),
+			label_set(0, "le", "1.5e0"),
+			label_set(1, "le", "+Inf"),
+		))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{0, 0, 0, 0, 0, 0},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("9.5e-1"),
+			},
+		}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1, 1, 1, 1, 1, 1},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("1.0e0"),
+			},
+		}
+		r3 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{3, 3, 3, 3, 3, 3},
+			Timestamps: timestampsExpected,
+		}
+		r3.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("1.5e0"),
+			},
+		}
+		r4 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{4, 4, 4, 4, 4, 4},
+			Timestamps: timestampsExpected,
+		}
+		r4.MetricName.Tags = []storage.Tag{
+			{
+				Key:   []byte("le"),
+				Value: []byte("+Inf"),
+			},
+		}
+		resultExpected := []netstorage.Result{r1, r2, r3, r4}
+		f(q, resultExpected)
+	})
 	t.Run(`avg(scalar) wiTHout (xx, yy)`, func(t *testing.T) {
 		t.Parallel()
 		q := `avg wiTHout (xx, yy) (123)`
@@ -2620,6 +3007,28 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`increases_over_time`, func(t *testing.T) {
+		t.Parallel()
+		q := `increases_over_time(rand(0)[200s:10s])`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{11, 9, 9, 12, 9, 8},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
+	t.Run(`decreases_over_time`, func(t *testing.T) {
+		t.Parallel()
+		q := `decreases_over_time(rand(0)[200s:10s])`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{9, 11, 11, 8, 11, 12},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
 	t.Run(`limitk(-1)`, func(t *testing.T) {
 		t.Parallel()
 		q := `limitk(-1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss"))`
@@ -2695,6 +3104,126 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r1, r2}
 		f(q, resultExpected)
 	})
+	t.Run(`topk_min(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(topk_min(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 10, 10, nan, nan, nan},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("foo"),
+			Value: []byte("bar"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`bottomk_min(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(bottomk_min(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, nan, 10.666666666666666, 12, 13.333333333333334},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`topk_max(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(topk_max(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, nan, 10.666666666666666, 12, 13.333333333333334},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`bottomk_max(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(bottomk_max(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 10, 10, nan, nan, nan},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("foo"),
+			Value: []byte("bar"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`topk_avg(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(topk_avg(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, nan, 10.666666666666666, 12, 13.333333333333334},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`bottomk_avg(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(bottomk_avg(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 10, 10, nan, nan, nan},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("foo"),
+			Value: []byte("bar"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`topk_median(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(topk_median(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{nan, nan, nan, 10.666666666666666, 12, 13.333333333333334},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
+	t.Run(`bottomk_median(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort(bottomk_median(1, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss")))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{10, 10, 10, nan, nan, nan},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("foo"),
+			Value: []byte("bar"),
+		}}
+		resultExpected := []netstorage.Result{r1}
+		f(q, resultExpected)
+	})
 	t.Run(`topk(1, nan_timeseries)`, func(t *testing.T) {
 		t.Parallel()
 		q := `topk(1, label_set(NaN, "foo", "bar") or label_set(time()/150, "baz", "sss")) default 0`
@@ -3472,7 +4001,7 @@ func TestExecSuccess(t *testing.T) {
 		}}
 		r4 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{0.85, 0.94, 0.97, 0.93, 0.98, 0.92},
+			Values:     []float64{0.9, 0.94, 0.97, 0.93, 0.98, 0.92},
 			Timestamps: timestampsExpected,
 		}
 		r4.MetricName.Tags = []storage.Tag{{
@@ -3520,7 +4049,7 @@ func TestExecSuccess(t *testing.T) {
 		q := `sort(rollup(time()[:50s]))`
 		r1 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{850, 1050, 1250, 1450, 1650, 1850},
+			Values:     []float64{800, 1000, 1200, 1400, 1600, 1800},
 			Timestamps: timestampsExpected,
 		}
 		r1.MetricName.Tags = []storage.Tag{{
@@ -3626,6 +4155,17 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`lag()`, func(t *testing.T) {
+		t.Parallel()
+		q := `lag(time()[60s:17s])`
+		r := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{14, 10, 6, 2, 15, 11},
+			Timestamps: timestampsExpected,
+		}
+		resultExpected := []netstorage.Result{r}
+		f(q, resultExpected)
+	})
 	t.Run(`()`, func(t *testing.T) {
 		t.Parallel()
 		q := `()`
@@ -3774,6 +4314,35 @@ func TestExecSuccess(t *testing.T) {
 		resultExpected := []netstorage.Result{r1, r2}
 		f(q, resultExpected)
 	})
+	t.Run(`((1),(2,3))`, func(t *testing.T) {
+		t.Parallel()
+		q := `((
+			alias(1, "x1"),
+		),(
+			alias(2, "x2"),
+			alias(3, "x3"),
+		))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1, 1, 1, 1, 1, 1},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.MetricGroup = []byte("x1")
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{2, 2, 2, 2, 2, 2},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.MetricGroup = []byte("x2")
+		r3 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{3, 3, 3, 3, 3, 3},
+			Timestamps: timestampsExpected,
+		}
+		r3.MetricName.MetricGroup = []byte("x3")
+		resultExpected := []netstorage.Result{r1, r2, r3}
+		f(q, resultExpected)
+	})
 	t.Run(`union(more-than-two)`, func(t *testing.T) {
 		t.Parallel()
 		q := `union(
@@ -4059,8 +4628,16 @@ func TestExecError(t *testing.T) {
 	f(`count_values()`)
 	f(`quantile()`)
 	f(`topk()`)
+	f(`topk_min()`)
+	f(`topk_max()`)
+	f(`topk_avg()`)
+	f(`topk_median()`)
 	f(`limitk()`)
 	f(`bottomk()`)
+	f(`bottomk_min()`)
+	f(`bottomk_max()`)
+	f(`bottomk_avg()`)
+	f(`bottomk_median()`)
 	f(`time(123)`)
 	f(`start(1)`)
 	f(`end(1)`)
@@ -4091,6 +4668,8 @@ func TestExecError(t *testing.T) {
 	f(`alias()`)
 	f(`alias(1)`)
 	f(`alias(1, "foo", "bar")`)
+	f(`lifetime()`)
+	f(`lag()`)
 
 	// Invalid argument type
 	f(`median_over_time({}, 2)`)
@@ -4101,6 +4680,7 @@ func TestExecError(t *testing.T) {
 	f(`clamp_max(1, 1 or label_set(2, "xx", "foo"))`)
 	f(`clamp_min(1, 1 or label_set(2, "xx", "foo"))`)
 	f(`topk(label_set(2, "xx", "foo") or 1, 12)`)
+	f(`topk_avg(label_set(2, "xx", "foo") or 1, 12)`)
 	f(`limitk(label_set(2, "xx", "foo") or 1, 12)`)
 	f(`round(1, 1 or label_set(2, "xx", "foo"))`)
 	f(`histogram_quantile(1 or label_set(2, "xx", "foo"), 1)`)
@@ -4176,27 +4756,27 @@ func testResultsEqual(t *testing.T, result, resultExpected []netstorage.Result)
 	for i := range result {
 		r := &result[i]
 		rExpected := &resultExpected[i]
-		testMetricNamesEqual(t, &r.MetricName, &rExpected.MetricName)
+		testMetricNamesEqual(t, &r.MetricName, &rExpected.MetricName, i)
 		testRowsEqual(t, r.Values, r.Timestamps, rExpected.Values, rExpected.Timestamps)
 	}
 }
 
-func testMetricNamesEqual(t *testing.T, mn, mnExpected *storage.MetricName) {
+func testMetricNamesEqual(t *testing.T, mn, mnExpected *storage.MetricName, pos int) {
 	t.Helper()
 	if string(mn.MetricGroup) != string(mnExpected.MetricGroup) {
-		t.Fatalf(`unexpected MetricGroup; got %q; want %q`, mn.MetricGroup, mnExpected.MetricGroup)
+		t.Fatalf(`unexpected MetricGroup at #%d; got %q; want %q`, pos, mn.MetricGroup, mnExpected.MetricGroup)
 	}
 	if len(mn.Tags) != len(mnExpected.Tags) {
-		t.Fatalf(`unexpected tags count; got %d; want %d`, len(mn.Tags), len(mnExpected.Tags))
+		t.Fatalf(`unexpected tags count at #%d; got %d; want %d`, pos, len(mn.Tags), len(mnExpected.Tags))
 	}
 	for i := range mn.Tags {
 		tag := &mn.Tags[i]
 		tagExpected := &mnExpected.Tags[i]
 		if string(tag.Key) != string(tagExpected.Key) {
-			t.Fatalf(`unexpected tag key; got %q; want %q`, tag.Key, tagExpected.Key)
+			t.Fatalf(`unexpected tag key at #%d,%d; got %q; want %q`, pos, i, tag.Key, tagExpected.Key)
 		}
 		if string(tag.Value) != string(tagExpected.Value) {
-			t.Fatalf(`unexpected tag value; got %q; want %q`, tag.Value, tagExpected.Value)
+			t.Fatalf(`unexpected tag value for key %q at #%d,%d; got %q; want %q`, tag.Key, pos, i, tag.Value, tagExpected.Value)
 		}
 	}
 }

app/vmselect/promql/parser.go

@@ -116,13 +116,17 @@ func removeParensExpr(e expr) expr {
 		return fe
 	}
 	if pe, ok := e.(*parensExpr); ok {
+		args := *pe
+		for i, arg := range args {
+			args[i] = removeParensExpr(arg)
+		}
 		if len(*pe) == 1 {
-			return removeParensExpr((*pe)[0])
+			return args[0]
 		}
 		// Treat parensExpr as a function with empty name, i.e. union()
 		fe := &funcExpr{
 			Name: "",
-			Args: *pe,
+			Args: args,
 		}
 		return fe
 	}
@@ -1550,6 +1554,10 @@ type rollupExpr struct {
 	InheritStep bool
 }
 
+func (re *rollupExpr) ForSubquery() bool {
+	return len(re.Step) > 0 || re.InheritStep
+}
+
 func (re *rollupExpr) AppendString(dst []byte) []byte {
 	needParens := func() bool {
 		if _, ok := re.Expr.(*rollupExpr); ok {

app/vmselect/promql/parser_test.go

@@ -252,6 +252,8 @@ func TestParsePromQLSuccess(t *testing.T) {
 	another(`(-foo + ((bar) / (baz))) + ((23))`, `((0 - foo) + (bar / baz)) + 23`)
 	another(`(FOO + ((Bar) / (baZ))) + ((23))`, `(FOO + (Bar / baZ)) + 23`)
 	same(`(foo, bar)`)
+	another(`((foo, bar),(baz))`, `((foo, bar), baz)`)
+	same(`(foo, (bar, baz), ((x, y), (z, y), xx))`)
 	another(`1+(foo, bar,)`, `1 + (foo, bar)`)
 	another(`((foo(bar,baz)), (1+(2)+(3,4)+()))`, `(foo(bar, baz), (3 + (3, 4)) + ())`)
 	same(`()`)

app/vmselect/promql/regexp_cache.go

@@ -51,11 +51,14 @@ type regexpCacheValue struct {
 }
 
 type regexpCache struct {
-	m  map[string]*regexpCacheValue
-	mu sync.RWMutex
+	// Move atomic counters to the top of struct for 8-byte alignment on 32-bit arch.
+	// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/212
 
 	requests uint64
 	misses   uint64
+
+	m  map[string]*regexpCacheValue
+	mu sync.RWMutex
 }
 
 func (rc *regexpCache) Requests() uint64 {

app/vmselect/promql/rollup.go

@@ -3,7 +3,6 @@ package promql
 import (
 	"fmt"
 	"math"
-	"sort"
 	"strings"
 	"sync"
 
@@ -23,8 +22,8 @@ var rollupFuncs = map[string]newRollupFunc{
 	"deriv_fast":         newRollupFuncOneArg(rollupDerivFast),
 	"holt_winters":       newRollupHoltWinters,
 	"idelta":             newRollupFuncOneArg(rollupIdelta),
-	"increase":           newRollupFuncOneArg(rollupDelta),  // + rollupFuncsRemoveCounterResets
-	"irate":              newRollupFuncOneArg(rollupIderiv), // + rollupFuncsRemoveCounterResets
+	"increase":           newRollupFuncOneArg(rollupIncrease), // + rollupFuncsRemoveCounterResets
+	"irate":              newRollupFuncOneArg(rollupIderiv),   // + rollupFuncsRemoveCounterResets
 	"predict_linear":     newRollupPredictLinear,
 	"rate":               newRollupFuncOneArg(rollupDerivFast), // + rollupFuncsRemoveCounterResets
 	"resets":             newRollupFuncOneArg(rollupResets),
@@ -38,21 +37,24 @@ var rollupFuncs = map[string]newRollupFunc{
 	"stdvar_over_time":   newRollupFuncOneArg(rollupStdvar),
 
 	// Additional rollup funcs.
-	"sum2_over_time":     newRollupFuncOneArg(rollupSum2),
-	"geomean_over_time":  newRollupFuncOneArg(rollupGeomean),
-	"first_over_time":    newRollupFuncOneArg(rollupFirst),
-	"last_over_time":     newRollupFuncOneArg(rollupLast),
-	"distinct_over_time": newRollupFuncOneArg(rollupDistinct),
-	"integrate":          newRollupFuncOneArg(rollupIntegrate),
-	"ideriv":             newRollupFuncOneArg(rollupIderiv),
-	"lifetime":           newRollupFuncOneArg(rollupLifetime),
-	"scrape_interval":    newRollupFuncOneArg(rollupScrapeInterval),
-	"rollup":             newRollupFuncOneArg(rollupFake),
-	"rollup_rate":        newRollupFuncOneArg(rollupFake), // + rollupFuncsRemoveCounterResets
-	"rollup_deriv":       newRollupFuncOneArg(rollupFake),
-	"rollup_delta":       newRollupFuncOneArg(rollupFake),
-	"rollup_increase":    newRollupFuncOneArg(rollupFake), // + rollupFuncsRemoveCounterResets
-	"rollup_candlestick": newRollupFuncOneArg(rollupFake),
+	"sum2_over_time":      newRollupFuncOneArg(rollupSum2),
+	"geomean_over_time":   newRollupFuncOneArg(rollupGeomean),
+	"first_over_time":     newRollupFuncOneArg(rollupFirst),
+	"last_over_time":      newRollupFuncOneArg(rollupLast),
+	"distinct_over_time":  newRollupFuncOneArg(rollupDistinct),
+	"increases_over_time": newRollupFuncOneArg(rollupIncreases),
+	"decreases_over_time": newRollupFuncOneArg(rollupDecreases),
+	"integrate":           newRollupFuncOneArg(rollupIntegrate),
+	"ideriv":              newRollupFuncOneArg(rollupIderiv),
+	"lifetime":            newRollupFuncOneArg(rollupLifetime),
+	"lag":                 newRollupFuncOneArg(rollupLag),
+	"scrape_interval":     newRollupFuncOneArg(rollupScrapeInterval),
+	"rollup":              newRollupFuncOneArg(rollupFake),
+	"rollup_rate":         newRollupFuncOneArg(rollupFake), // + rollupFuncsRemoveCounterResets
+	"rollup_deriv":        newRollupFuncOneArg(rollupFake),
+	"rollup_delta":        newRollupFuncOneArg(rollupFake),
+	"rollup_increase":     newRollupFuncOneArg(rollupFake), // + rollupFuncsRemoveCounterResets
+	"rollup_candlestick":  newRollupFuncOneArg(rollupFake),
 }
 
 var rollupFuncsMayAdjustWindow = map[string]bool{
@@ -111,8 +113,10 @@ type rollupFuncArg struct {
 	values        []float64
 	timestamps    []int64
 
-	idx  int
-	step int64
+	currTimestamp int64
+	idx           int
+	step          int64
+	realPrevValue float64
 }
 
 func (rfa *rollupFuncArg) reset() {
@@ -120,8 +124,10 @@ func (rfa *rollupFuncArg) reset() {
 	rfa.prevTimestamp = 0
 	rfa.values = nil
 	rfa.timestamps = nil
+	rfa.currTimestamp = 0
 	rfa.idx = 0
 	rfa.step = 0
+	rfa.realPrevValue = nan
 }
 
 // rollupFunc must return rollup value for the given rfa.
@@ -147,6 +153,9 @@ type rollupConfig struct {
 	MayAdjustWindow bool
 
 	Timestamps []int64
+
+	// LoookbackDelta is the analog to `-query.lookback-delta` from Prometheus world.
+	LookbackDelta int64
 }
 
 var (
@@ -184,6 +193,9 @@ func (rc *rollupConfig) Do(dstValues []float64, values []float64, timestamps []i
 	dstValues = decimal.ExtendFloat64sCapacity(dstValues, len(rc.Timestamps))
 
 	maxPrevInterval := getMaxPrevInterval(timestamps)
+	if rc.LookbackDelta > 0 && maxPrevInterval > rc.LookbackDelta {
+		maxPrevInterval = rc.LookbackDelta
+	}
 	window := rc.Window
 	if window <= 0 {
 		window = rc.Step
@@ -194,6 +206,7 @@ func (rc *rollupConfig) Do(dstValues []float64, values []float64, timestamps []i
 	rfa := getRollupFuncArg()
 	rfa.idx = 0
 	rfa.step = rc.Step
+	rfa.realPrevValue = nan
 
 	i := 0
 	j := 0
@@ -218,6 +231,10 @@ func (rc *rollupConfig) Do(dstValues []float64, values []float64, timestamps []i
 
 		rfa.values = values[i:j]
 		rfa.timestamps = timestamps[i:j]
+		rfa.currTimestamp = tEnd
+		if i > 0 {
+			rfa.realPrevValue = values[i-1]
+		}
 		value := rc.Func(rfa)
 		rfa.idx++
 		dstValues = append(dstValues, value)
@@ -261,17 +278,42 @@ func seekFirstTimestampIdxAfter(timestamps []int64, seekTimestamp int64, nHint i
 		return startIdx + len(timestamps)
 	}
 	// Slow path: too big len(timestamps), so use binary search.
-	i := sort.Search(len(timestamps), func(n int) bool {
-		return n >= 0 && n < len(timestamps) && timestamps[n] > seekTimestamp
-	})
-	return startIdx + i
+	i := binarySearchInt64(timestamps, seekTimestamp+1)
+	return startIdx + int(i)
+}
+
+func binarySearchInt64(a []int64, v int64) uint {
+	// Copy-pasted sort.Search from https://golang.org/src/sort/search.go?s=2246:2286#L49
+	i, j := uint(0), uint(len(a))
+	for i < j {
+		h := (i + j) >> 1
+		if h < uint(len(a)) && a[h] < v {
+			i = h + 1
+		} else {
+			j = h
+		}
+	}
+	return i
 }
 
 func getMaxPrevInterval(timestamps []int64) int64 {
 	if len(timestamps) < 2 {
 		return int64(maxSilenceInterval)
 	}
-	d := (timestamps[len(timestamps)-1] - timestamps[0]) / int64(len(timestamps)-1)
+
+	// Estimate scrape interval as 0.6 quantile for the first 100 intervals.
+	h := histogram.GetFast()
+	tsPrev := timestamps[0]
+	timestamps = timestamps[1:]
+	if len(timestamps) > 100 {
+		timestamps = timestamps[:100]
+	}
+	for _, ts := range timestamps {
+		h.Update(float64(ts - tsPrev))
+		tsPrev = ts
+	}
+	d := int64(h.Quantile(0.6))
+	histogram.PutFast(h)
 	if d <= 0 {
 		return int64(maxSilenceInterval)
 	}
@@ -531,11 +573,14 @@ func rollupAvg(rfa *rollupFuncArg) float64 {
 func rollupMin(rfa *rollupFuncArg) float64 {
 	// There is no need in handling NaNs here, since they must be cleaned up
 	// before calling rollup funcs.
+	minValue := rfa.prevValue
 	values := rfa.values
-	if len(values) == 0 {
-		return rfa.prevValue
+	if math.IsNaN(minValue) {
+		if len(values) == 0 {
+			return nan
+		}
+		minValue = values[0]
 	}
-	minValue := values[0]
 	for _, v := range values {
 		if v < minValue {
 			minValue = v
@@ -547,11 +592,14 @@ func rollupMin(rfa *rollupFuncArg) float64 {
 func rollupMax(rfa *rollupFuncArg) float64 {
 	// There is no need in handling NaNs here, since they must be cleaned up
 	// before calling rollup funcs.
+	maxValue := rfa.prevValue
 	values := rfa.values
-	if len(values) == 0 {
-		return rfa.prevValue
+	if math.IsNaN(maxValue) {
+		if len(values) == 0 {
+			return nan
+		}
+		maxValue = values[0]
 	}
-	maxValue := values[0]
 	for _, v := range values {
 		if v > maxValue {
 			maxValue = v
@@ -565,7 +613,10 @@ func rollupSum(rfa *rollupFuncArg) float64 {
 	// before calling rollup funcs.
 	values := rfa.values
 	if len(values) == 0 {
-		return rfa.prevValue
+		if math.IsNaN(rfa.prevValue) {
+			return nan
+		}
+		return 0
 	}
 	var sum float64
 	for _, v := range values {
@@ -649,6 +700,14 @@ func rollupStdvar(rfa *rollupFuncArg) float64 {
 }
 
 func rollupDelta(rfa *rollupFuncArg) float64 {
+	return rollupDeltaInternal(rfa, false)
+}
+
+func rollupIncrease(rfa *rollupFuncArg) float64 {
+	return rollupDeltaInternal(rfa, true)
+}
+
+func rollupDeltaInternal(rfa *rollupFuncArg, canUseRealPrevValue bool) float64 {
 	// There is no need in handling NaNs here, since they must be cleaned up
 	// before calling rollup funcs.
 	values := rfa.values
@@ -658,6 +717,10 @@ func rollupDelta(rfa *rollupFuncArg) float64 {
 			return nan
 		}
 		if len(values) == 1 {
+			if canUseRealPrevValue && !math.IsNaN(rfa.realPrevValue) {
+				// Fix against removeCounterResets.
+				return values[0] - rfa.realPrevValue
+			}
 			// Assume that the previous non-existing value was 0.
 			return values[0]
 		}
@@ -782,6 +845,18 @@ func rollupLifetime(rfa *rollupFuncArg) float64 {
 	return float64(timestamps[len(timestamps)-1]-rfa.prevTimestamp) * 1e-3
 }
 
+func rollupLag(rfa *rollupFuncArg) float64 {
+	// Calculate the duration between the current timestamp and the last data point.
+	timestamps := rfa.timestamps
+	if len(timestamps) == 0 {
+		if math.IsNaN(rfa.prevValue) {
+			return nan
+		}
+		return float64(rfa.currTimestamp-rfa.prevTimestamp) * 1e-3
+	}
+	return float64(rfa.currTimestamp-timestamps[len(timestamps)-1]) * 1e-3
+}
+
 func rollupScrapeInterval(rfa *rollupFuncArg) float64 {
 	// Calculate the average interval between data points.
 	timestamps := rfa.timestamps
@@ -820,6 +895,37 @@ func rollupChanges(rfa *rollupFuncArg) float64 {
 	return float64(n)
 }
 
+func rollupIncreases(rfa *rollupFuncArg) float64 {
+	// There is no need in handling NaNs here, since they must be cleaned up
+	// before calling rollup funcs.
+	values := rfa.values
+	if len(values) == 0 {
+		if math.IsNaN(rfa.prevValue) {
+			return nan
+		}
+		return 0
+	}
+	prevValue := rfa.prevValue
+	if math.IsNaN(prevValue) {
+		prevValue = values[0]
+		values = values[1:]
+	}
+	if len(values) == 0 {
+		return 0
+	}
+	n := 0
+	for _, v := range values {
+		if v > prevValue {
+			n++
+		}
+		prevValue = v
+	}
+	return float64(n)
+}
+
+// `decreases_over_time` logic is the same as `resets` logic.
+var rollupDecreases = rollupResets
+
 func rollupResets(rfa *rollupFuncArg) float64 {
 	// There is no need in handling NaNs here, since they must be cleaned up
 	// before calling rollup funcs.
@@ -922,6 +1028,8 @@ func rollupIntegrate(rfa *rollupFuncArg) float64 {
 		timestamp := timestamps[i]
 		dt := float64(timestamp-prevTimestamp) * 1e-3
 		sum += 0.5 * (v + prevValue) * dt
+		prevTimestamp = timestamp
+		prevValue = v
 	}
 	return sum
 }

app/vmselect/promql/rollup_result_cache_test.go

@@ -388,7 +388,7 @@ func testTimeseriesEqual(t *testing.T, tss, tssExpected []*timeseries) {
 	}
 	for i, ts := range tss {
 		tsExpected := tssExpected[i]
-		testMetricNamesEqual(t, &ts.MetricName, &tsExpected.MetricName)
+		testMetricNamesEqual(t, &ts.MetricName, &tsExpected.MetricName, i)
 		testRowsEqual(t, ts.Values, ts.Timestamps, tsExpected.Values, tsExpected.Timestamps)
 	}
 }

app/vmselect/promql/rollup_test.go

@@ -291,9 +291,11 @@ func TestRollupNewRollupFuncSuccess(t *testing.T) {
 	f("stdvar_over_time", 945.7430555555555)
 	f("first_over_time", 123)
 	f("last_over_time", 34)
-	f("integrate", 61.0275)
+	f("integrate", 5.4705)
 	f("distinct_over_time", 8)
 	f("ideriv", 0)
+	f("decreases_over_time", 5)
+	f("increases_over_time", 5)
 }
 
 func TestRollupNewRollupFuncError(t *testing.T) {
@@ -486,6 +488,51 @@ func TestRollupWindowPartialPoints(t *testing.T) {
 	})
 }
 
+func TestRollupFuncsLookbackDelta(t *testing.T) {
+	t.Run("1", func(t *testing.T) {
+		rc := rollupConfig{
+			Func:          rollupFirst,
+			Start:         80,
+			End:           140,
+			Step:          10,
+			LookbackDelta: 1,
+		}
+		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
+		values := rc.Do(nil, testValues, testTimestamps)
+		valuesExpected := []float64{99, 12, 44, nan, 32, 34, nan}
+		timestampsExpected := []int64{80, 90, 100, 110, 120, 130, 140}
+		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
+	})
+	t.Run("7", func(t *testing.T) {
+		rc := rollupConfig{
+			Func:          rollupFirst,
+			Start:         80,
+			End:           140,
+			Step:          10,
+			LookbackDelta: 7,
+		}
+		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
+		values := rc.Do(nil, testValues, testTimestamps)
+		valuesExpected := []float64{99, 12, 44, 44, 32, 34, nan}
+		timestampsExpected := []int64{80, 90, 100, 110, 120, 130, 140}
+		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
+	})
+	t.Run("0", func(t *testing.T) {
+		rc := rollupConfig{
+			Func:          rollupFirst,
+			Start:         80,
+			End:           140,
+			Step:          10,
+			LookbackDelta: 0,
+		}
+		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
+		values := rc.Do(nil, testValues, testTimestamps)
+		valuesExpected := []float64{34, 12, 12, 44, 44, 34, nan}
+		timestampsExpected := []int64{80, 90, 100, 110, 120, 130, 140}
+		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
+	})
+}
+
 func TestRollupFuncsNoWindow(t *testing.T) {
 	t.Run("first", func(t *testing.T) {
 		rc := rollupConfig{
@@ -525,7 +572,7 @@ func TestRollupFuncsNoWindow(t *testing.T) {
 		}
 		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
 		values := rc.Do(nil, testValues, testTimestamps)
-		valuesExpected := []float64{nan, 21, 12, 32, 34}
+		valuesExpected := []float64{nan, 21, 12, 12, 34}
 		timestampsExpected := []int64{0, 40, 80, 120, 160}
 		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
 	})
@@ -585,6 +632,20 @@ func TestRollupFuncsNoWindow(t *testing.T) {
 		timestampsExpected := []int64{10, 50, 90, 130}
 		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
 	})
+	t.Run("lag", func(t *testing.T) {
+		rc := rollupConfig{
+			Func:   rollupLag,
+			Start:  0,
+			End:    160,
+			Step:   40,
+			Window: 0,
+		}
+		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
+		values := rc.Do(nil, testValues, testTimestamps)
+		valuesExpected := []float64{nan, 0.004, 0, 0, 0.03}
+		timestampsExpected := []int64{0, 40, 80, 120, 160}
+		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
+	})
 	t.Run("lifetime_1", func(t *testing.T) {
 		rc := rollupConfig{
 			Func:   rollupLifetime,
@@ -749,7 +810,7 @@ func TestRollupFuncsNoWindow(t *testing.T) {
 		}
 		rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
 		values := rc.Do(nil, testValues, testTimestamps)
-		valuesExpected := []float64{nan, 4.6035, 4.3934999999999995, 2.166, 0.34}
+		valuesExpected := []float64{nan, 1.526, 2.2795, 1.325, 0.34}
 		timestampsExpected := []int64{0, 40, 80, 120, 160}
 		testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
 	})
@@ -783,6 +844,27 @@ func TestRollupFuncsNoWindow(t *testing.T) {
 	})
 }
 
+func TestRollupBigNumberOfValues(t *testing.T) {
+	const srcValuesCount = 1e4
+	rc := rollupConfig{
+		Func:   rollupDefault,
+		End:    srcValuesCount,
+		Step:   srcValuesCount / 5,
+		Window: srcValuesCount / 4,
+	}
+	rc.Timestamps = getTimestamps(rc.Start, rc.End, rc.Step)
+	srcValues := make([]float64, srcValuesCount)
+	srcTimestamps := make([]int64, srcValuesCount)
+	for i := 0; i < srcValuesCount; i++ {
+		srcValues[i] = float64(i)
+		srcTimestamps[i] = int64(i / 2)
+	}
+	values := rc.Do(nil, srcValues, srcTimestamps)
+	valuesExpected := []float64{1, 4001, 8001, 9999, nan, nan}
+	timestampsExpected := []int64{0, 2000, 4000, 6000, 8000, 10000}
+	testRowsEqual(t, values, rc.Timestamps, valuesExpected, timestampsExpected)
+}
+
 func testRowsEqual(t *testing.T, values []float64, timestamps []int64, valuesExpected []float64, timestampsExpected []int64) {
 	t.Helper()
 	if len(values) != len(valuesExpected) {
@@ -811,7 +893,7 @@ func testRowsEqual(t *testing.T, values []float64, timestamps []int64, valuesExp
 			}
 			continue
 		}
-		if v != vExpected {
+		if math.Abs(v-vExpected) > 1e-15 {
 			t.Fatalf("unexpected value at values[%d]; got %f; want %f\nvalues=\n%v\nvaluesExpected=\n%v",
 				i, v, vExpected, values, valuesExpected)
 		}

app/vmselect/promql/timeseries.go

@@ -288,7 +288,6 @@ func marshalMetricTagsFast(dst []byte, tags []storage.Tag) []byte {
 }
 
 func marshalMetricNameSorted(dst []byte, mn *storage.MetricName) []byte {
-	// Do not marshal AccountID and ProjectID, since they are unused.
 	dst = marshalBytesFast(dst, mn.MetricGroup)
 	sortMetricTags(mn.Tags)
 	dst = marshalMetricTagsFast(dst, mn.Tags)

app/vmselect/promql/transform.go

@@ -91,6 +91,7 @@ var transformFuncs = map[string]transformFunc{
 	"cos":                newTransformFuncOneArg(transformCos),
 	"asin":               newTransformFuncOneArg(transformAsin),
 	"acos":               newTransformFuncOneArg(transformAcos),
+	"prometheus_buckets": transformPrometheusBuckets,
 }
 
 func getTransformFunc(s string) transformFunc {
@@ -272,6 +273,131 @@ func transformFloor(v float64) float64 {
 	return math.Floor(v)
 }
 
+func transformPrometheusBuckets(tfa *transformFuncArg) ([]*timeseries, error) {
+	args := tfa.args
+	if err := expectTransformArgsNum(args, 1); err != nil {
+		return nil, err
+	}
+	rvs := vmrangeBucketsToLE(args[0])
+	return rvs, nil
+}
+
+func vmrangeBucketsToLE(tss []*timeseries) []*timeseries {
+	rvs := make([]*timeseries, 0, len(tss))
+
+	// Group timeseries by MetricGroup+tags excluding `vmrange` tag.
+	type x struct {
+		startStr string
+		endStr   string
+		start    float64
+		end      float64
+		ts       *timeseries
+	}
+	m := make(map[string][]x)
+	bb := bbPool.Get()
+	defer bbPool.Put(bb)
+	for _, ts := range tss {
+		vmrange := ts.MetricName.GetTagValue("vmrange")
+		if len(vmrange) == 0 {
+			if le := ts.MetricName.GetTagValue("le"); len(le) > 0 {
+				// Keep Prometheus-compatible buckets.
+				rvs = append(rvs, ts)
+			}
+			continue
+		}
+		n := strings.Index(bytesutil.ToUnsafeString(vmrange), "...")
+		if n < 0 {
+			continue
+		}
+		startStr := string(vmrange[:n])
+		start, err := strconv.ParseFloat(startStr, 64)
+		if err != nil {
+			continue
+		}
+		endStr := string(vmrange[n+len("..."):])
+		end, err := strconv.ParseFloat(endStr, 64)
+		if err != nil {
+			continue
+		}
+		ts.MetricName.RemoveTag("le")
+		ts.MetricName.RemoveTag("vmrange")
+		bb.B = marshalMetricNameSorted(bb.B[:0], &ts.MetricName)
+		m[string(bb.B)] = append(m[string(bb.B)], x{
+			startStr: startStr,
+			endStr:   endStr,
+			start:    start,
+			end:      end,
+			ts:       ts,
+		})
+	}
+
+	// Convert `vmrange` label in each group of time series to `le` label.
+	copyTS := func(src *timeseries, leStr string) *timeseries {
+		var ts timeseries
+		ts.CopyFromShallowTimestamps(src)
+		values := ts.Values
+		for i := range values {
+			values[i] = 0
+		}
+		ts.MetricName.RemoveTag("le")
+		ts.MetricName.AddTag("le", leStr)
+		return &ts
+	}
+	isZeroTS := func(ts *timeseries) bool {
+		for _, v := range ts.Values {
+			if v > 0 {
+				return false
+			}
+		}
+		return true
+	}
+	for _, xss := range m {
+		sort.Slice(xss, func(i, j int) bool { return xss[i].end < xss[j].end })
+		xssNew := make([]x, 0, len(xss)+2)
+		var xsPrev x
+		for _, xs := range xss {
+			ts := xs.ts
+			if isZeroTS(ts) {
+				// Skip time series with zeros. They are substituted by xssNew below.
+				continue
+			}
+			if xs.start != xsPrev.end {
+				xssNew = append(xssNew, x{
+					endStr: xs.startStr,
+					end:    xs.start,
+					ts:     copyTS(ts, xs.startStr),
+				})
+			}
+			ts.MetricName.AddTag("le", xs.endStr)
+			xssNew = append(xssNew, xs)
+			xsPrev = xs
+		}
+		if !math.IsInf(xsPrev.end, 1) {
+			xssNew = append(xssNew, x{
+				endStr: "+Inf",
+				end:    math.Inf(1),
+				ts:     copyTS(xsPrev.ts, "+Inf"),
+			})
+		}
+		xss = xssNew
+		for i := range xss[0].ts.Values {
+			count := float64(0)
+			for _, xs := range xss {
+				ts := xs.ts
+				v := ts.Values[i]
+				if !math.IsNaN(v) && v > 0 {
+					count += v
+				}
+				ts.Values[i] = count
+			}
+		}
+		for _, xs := range xss {
+			rvs = append(rvs, xs.ts)
+		}
+	}
+	return rvs
+}
+
 func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 	args := tfa.args
 	if err := expectTransformArgsNum(args, 2); err != nil {
@@ -282,6 +408,9 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 		return nil, err
 	}
 
+	// Convert buckets with `vmrange` labels to buckets with `le` labels.
+	tss := vmrangeBucketsToLE(args[1])
+
 	// Group metrics by all tags excluding "le"
 	type x struct {
 		le float64
@@ -289,7 +418,7 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 	}
 	m := make(map[string][]x)
 	bb := bbPool.Get()
-	for _, ts := range args[1] {
+	for _, ts := range tss {
 		tagValue := ts.MetricName.GetTagValue("le")
 		if len(tagValue) == 0 {
 			continue
@@ -313,18 +442,16 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 	lastNonInf := func(i int, xss []x) float64 {
 		for len(xss) > 0 {
 			xsLast := xss[len(xss)-1]
-			if xsLast.ts.Values[i] == 0 {
+			v := xsLast.ts.Values[i]
+			if v == 0 {
 				return nan
 			}
-			if !math.IsInf(xsLast.le, 0) {
-				break
+			if !math.IsNaN(v) && !math.IsInf(xsLast.le, 0) {
+				return xsLast.le
 			}
 			xss = xss[:len(xss)-1]
 		}
-		if len(xss) == 0 {
-			return nan
-		}
-		return xss[len(xss)-1].le
+		return nan
 	}
 	quantile := func(i int, phis []float64, xss []x) float64 {
 		phi := phis[i]
@@ -337,13 +464,21 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 		vPrev := float64(0)
 		for _, xs := range xss {
 			v := xs.ts.Values[i]
-			if math.IsNaN(v) || v < vPrev {
+			if v < vPrev {
 				xs.ts.Values[i] = vPrev
-			} else {
+			} else if !math.IsNaN(v) {
 				vPrev = v
 			}
 		}
-		if len(xss) == 0 {
+		vLast := nan
+		for len(xss) > 0 {
+			vLast = xss[len(xss)-1].ts.Values[i]
+			if !math.IsNaN(vLast) {
+				break
+			}
+			xss = xss[:len(xss)-1]
+		}
+		if vLast == 0 || math.IsNaN(vLast) {
 			return nan
 		}
 		if phi < 0 {
@@ -352,16 +487,22 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 		if phi > 1 {
 			return inf
 		}
-		vLast := xss[len(xss)-1].ts.Values[i]
-		if vLast == 0 {
-			return nan
-		}
 		vReq := vLast * phi
 		vPrev = 0
 		lePrev := float64(0)
 		for _, xs := range xss {
 			v := xs.ts.Values[i]
+			if math.IsNaN(v) {
+				// Skip NaNs - they may appear if the selected time range
+				// contains multiple different bucket sets.
+				continue
+			}
 			le := xs.le
+			if v <= 0 {
+				// Skip zero buckets.
+				lePrev = le
+				continue
+			}
 			if v < vReq {
 				vPrev = v
 				lePrev = le
@@ -388,7 +529,6 @@ func transformHistogramQuantile(tfa *transformFuncArg) ([]*timeseries, error) {
 		}
 		rvs = append(rvs, dst)
 	}
-
 	return rvs, nil
 }
 
@@ -1121,7 +1261,10 @@ func transformTimestamp(tfa *transformFuncArg) ([]*timeseries, error) {
 		ts.MetricName.ResetMetricGroup()
 		values := ts.Values
 		for i, t := range ts.Timestamps {
-			values[i] = float64(t) / 1e3
+			v := values[i]
+			if !math.IsNaN(v) {
+				values[i] = float64(t) / 1e3
+			}
 		}
 	}
 	return rvs, nil

app/vmstorage/main.go

@@ -24,6 +24,9 @@ var (
 
 	// DataPath is a path to storage data.
 	DataPath = flag.String("storageDataPath", "victoria-metrics-data", "Path to storage data")
+
+	bigMergeConcurrency   = flag.Int("bigMergeConcurrency", 0, "The maximum number of CPU cores to use for big merges. Default value is used if set to 0")
+	smallMergeConcurrency = flag.Int("smallMergeConcurrency", 0, "The maximum number of CPU cores to use for small merges. Default value is used if set to 0")
 )
 
 // Init initializes vmstorage.
@@ -39,6 +42,10 @@ func InitWithoutMetrics() {
 	if err := encoding.CheckPrecisionBits(uint8(*precisionBits)); err != nil {
 		logger.Fatalf("invalid `-precisionBits`: %s", err)
 	}
+
+	storage.SetBigMergeWorkersCount(*bigMergeConcurrency)
+	storage.SetSmallMergeWorkersCount(*smallMergeConcurrency)
+
 	logger.Infof("opening storage at %q with retention period %d months", *DataPath, *retentionPeriod)
 	startTime := time.Now()
 	WG = syncwg.WaitGroup{}
@@ -298,6 +305,9 @@ func registerStorageMetrics() {
 		return float64(idbm().PartsRefCount)
 	})
 
+	metrics.NewGauge(`vm_new_timeseries_created_total`, func() float64 {
+		return float64(idbm().NewTimeseriesCreated)
+	})
 	metrics.NewGauge(`vm_missing_tsids_for_metric_id_total`, func() float64 {
 		return float64(idbm().MissingTSIDsForMetricID)
 	})
@@ -313,6 +323,12 @@ func registerStorageMetrics() {
 	metrics.NewGauge(`vm_date_metric_ids_search_hits_total`, func() float64 {
 		return float64(idbm().DateMetricIDsSearchHits)
 	})
+	metrics.NewGauge(`vm_index_blocks_with_metric_ids_processed_total`, func() float64 {
+		return float64(idbm().IndexBlocksWithMetricIDsProcessed)
+	})
+	metrics.NewGauge(`vm_index_blocks_with_metric_ids_incorrect_order_total`, func() float64 {
+		return float64(idbm().IndexBlocksWithMetricIDsIncorrectOrder)
+	})
 
 	metrics.NewGauge(`vm_assisted_merges_total{type="storage/small"}`, func() float64 {
 		return float64(tm().SmallAssistedMerges)
@@ -391,6 +407,24 @@ func registerStorageMetrics() {
 		return float64(idbm().ItemsCount)
 	})
 
+	metrics.NewGauge(`vm_date_range_search_calls_total`, func() float64 {
+		return float64(idbm().DateRangeSearchCalls)
+	})
+	metrics.NewGauge(`vm_date_range_hits_total`, func() float64 {
+		return float64(idbm().DateRangeSearchHits)
+	})
+
+	metrics.NewGauge(`vm_missing_metric_names_for_metric_id_total`, func() float64 {
+		return float64(idbm().MissingMetricNamesForMetricID)
+	})
+
+	metrics.NewGauge(`vm_date_metric_id_cache_syncs_total`, func() float64 {
+		return float64(m().DateMetricIDCacheSyncsCount)
+	})
+	metrics.NewGauge(`vm_date_metric_id_cache_resets_total`, func() float64 {
+		return float64(m().DateMetricIDCacheResetsCount)
+	})
+
 	metrics.NewGauge(`vm_cache_entries{type="storage/tsid"}`, func() float64 {
 		return float64(m().TSIDCacheSize)
 	})
@@ -440,6 +474,9 @@ func registerStorageMetrics() {
 	metrics.NewGauge(`vm_cache_size_bytes{type="storage/date_metricID"}`, func() float64 {
 		return float64(m().DateMetricIDCacheSizeBytes)
 	})
+	metrics.NewGauge(`vm_cache_size_bytes{type="storage/hour_metric_ids"}`, func() float64 {
+		return float64(m().HourMetricIDCacheSizeBytes)
+	})
 	metrics.NewGauge(`vm_cache_size_bytes{type="indexdb/tagFilters"}`, func() float64 {
 		return float64(idbm().TagCacheSizeBytes)
 	})
@@ -456,9 +493,6 @@ func registerStorageMetrics() {
 	metrics.NewGauge(`vm_cache_requests_total{type="storage/metricName"}`, func() float64 {
 		return float64(m().MetricNameCacheRequests)
 	})
-	metrics.NewGauge(`vm_cache_requests_total{type="storage/date_metricID"}`, func() float64 {
-		return float64(m().DateMetricIDCacheRequests)
-	})
 	metrics.NewGauge(`vm_cache_requests_total{type="storage/bigIndexBlocks"}`, func() float64 {
 		return float64(tm().BigIndexBlocksCacheRequests)
 	})
@@ -490,9 +524,6 @@ func registerStorageMetrics() {
 	metrics.NewGauge(`vm_cache_misses_total{type="storage/metricName"}`, func() float64 {
 		return float64(m().MetricNameCacheMisses)
 	})
-	metrics.NewGauge(`vm_cache_misses_total{type="storage/date_metricID"}`, func() float64 {
-		return float64(m().DateMetricIDCacheMisses)
-	})
 	metrics.NewGauge(`vm_cache_misses_total{type="storage/bigIndexBlocks"}`, func() float64 {
 		return float64(tm().BigIndexBlocksCacheMisses)
 	})
@@ -525,7 +556,4 @@ func registerStorageMetrics() {
 	metrics.NewGauge(`vm_cache_collisions_total{type="storage/metricName"}`, func() float64 {
 		return float64(m().MetricNameCacheCollisions)
 	})
-	metrics.NewGauge(`vm_cache_collisions_total{type="storage/date_metricID"}`, func() float64 {
-		return float64(m().DateMetricIDCacheCollisions)
-	})
 }

deployment/docker/Makefile

@@ -1,6 +1,6 @@
 DOCKER_NAMESPACE := victoriametrics
-BUILDER_IMAGE := local/builder:go1.13.0
-CERTS_IMAGE := local/certs:1.0.2
+BUILDER_IMAGE := local/builder:go1.13.4
+CERTS_IMAGE := local/certs:1.0.3
 
 package-certs:
 	(docker image ls --format '{{.Repository}}:{{.Tag}}' | grep -q '$(CERTS_IMAGE)') \
@@ -21,7 +21,8 @@ app-via-docker: package-certs package-builder
 		--env GO111MODULE=on \
 		$(DOCKER_OPTS) \
 		$(BUILDER_IMAGE) \
-		go build $(RACE) -mod=vendor -ldflags "-s -w -extldflags '-static' $(GO_BUILDINFO)" -tags 'netgo osusergo' -o bin/$(APP_NAME)$(APP_SUFFIX)-prod $(PKG_PREFIX)/app/$(APP_NAME)
+		go build $(RACE) -mod=vendor -trimpath -ldflags "-s -w -extldflags '-static' $(GO_BUILDINFO)" -tags 'netgo osusergo' \
+			-o bin/$(APP_NAME)$(APP_SUFFIX)-prod $(PKG_PREFIX)/app/$(APP_NAME)
 
 package-via-docker:
 	(docker image ls --format '{{.Repository}}:{{.Tag}}' | grep -q '$(DOCKER_NAMESPACE)/$(APP_NAME):$(PKG_TAG)$(RACE)') || (\

deployment/docker/builder/Dockerfile

@@ -1,2 +1,2 @@
-FROM golang:1.13.0
+FROM golang:1.13.4
 STOPSIGNAL SIGINT

deployment/docker/certs/Dockerfile

@@ -1,3 +1,3 @@
 # See https://medium.com/on-docker/use-multi-stage-builds-to-inject-ca-certs-ad1e8f01de1b
-FROM alpine:3.9 as certs
+FROM alpine:3.10 as certs
 RUN apk --update add ca-certificates

deployment/docker/docker-compose.yml

@@ -2,7 +2,7 @@ version: '3.5'
 services:
   prometheus:
     container_name: prometheus
-    image: prom/prometheus:v2.10.0
+    image: prom/prometheus:v2.14.0
     depends_on:
       - "victoriametrics"
     ports:
@@ -35,7 +35,7 @@ services:
     restart: always
   grafana:
     container_name: grafana
-    image: grafana/grafana:6.2.1
+    image: grafana/grafana:6.5.0
     entrypoint: >
       /bin/sh -c "
       cd /var/lib/grafana &&

deployment/docker/provisioning/datasources/datasource.yml

@@ -5,10 +5,10 @@ datasources:
       type: prometheus
       access: proxy
       url: http://prometheus:9090
-      isDefault: false
+      isDefault: true
 
     - name: VictoriaMetrics
       type: prometheus
       access: proxy
       url: http://victoriametrics:8428
-      isDefault: true
+      isDefault: false

docs/Articles.md

@@ -0,0 +1,22 @@
+# Articles
+
+* [Open-sourcing VictoriaMetrics](https://medium.com/@valyala/open-sourcing-victoriametrics-f31e34485c2b)
+* [How we created VictoriaMetrics](https://medium.com/devopslinks/victoriametrics-creating-the-best-remote-storage-for-prometheus-5d92d66787ac)
+* [VictoriaMetrics vs TimescaleDB vs InfluxDB benchmarks on 40K unique time series](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4)
+* [VictoriaMetrics vs TimescaleDB vs InfluxDB benchmarks on 400K, 4M and 40M unique time series](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
+* [Insert benchmarks for VictoriaMetrics vs InfluxDB on high-cardinality data](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893)
+* [Measuring vertical scalability for time series databases in Google Cloud](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
+* [How VictoriaMetrics creates instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
+* [Prometheus Subqueries in VictoriaMetrics](https://medium.com/@valyala/prometheus-subqueries-in-victoriametrics-9b1492b720b3)
+* [Why irate from Prometheus doesn't capture spikes](https://medium.com/@valyala/why-irate-from-prometheus-doesnt-capture-spikes-45f9896d7832)
+* [Why mmap'ed files in Go may hurt performance](https://medium.com/@valyala/mmap-in-go-considered-harmful-d92a25cb161d)
+* [WAL Usage Looks Broken in Modern TSDBs](https://medium.com/@valyala/wal-usage-looks-broken-in-modern-time-series-databases-b62a627ab704)
+* [Analyzing Prometheus data with external tools](https://medium.com/@valyala/analyzing-prometheus-data-with-external-tools-5f3e5e147639)
+* [Stripping dependency bloat in VictoriaMetrics Docker image](https://medium.com/@valyala/stripping-dependency-bloat-in-victoriametrics-docker-image-983fb5912b0d)
+* [PromQL tutorial for beginners](https://medium.com/@valyala/promql-tutorial-for-beginners-9ab455142085)
+* [Achieving better compression for time series data than Gorilla](https://medium.com/@valyala/victoriametrics-achieving-better-compression-for-time-series-data-than-gorilla-317bc1f95932)
+* [Comparing Thanos to VictoriaMetrics cluster](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683)
+* [Speeding up backups for big time series databases](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883)
+* [Evaluation performance and correctness: VictoriaMetrics response](https://medium.com/@valyala/evaluating-performance-and-correctness-victoriametrics-response-e27315627e87)
+* [Improving histogram usability for Prometheus and Grafana](https://medium.com/@valyala/improving-histogram-usability-for-prometheus-and-grafana-bc7e5df0e350)
+* [Prometheus storage: tech terms for humans](https://medium.com/@valyala/prometheus-storage-technical-terms-for-humans-4ab4de6c3d48)

docs/Cluster-VictoriaMetrics.md

@@ -0,0 +1,326 @@
+# Cluster version
+
+VictoriaMetrics is fast, cost-effective and scalable time series database. It can be used as a long-term remote storage for Prometheus.
+
+It is recommended using [single-node version](https://github.com/VictoriaMetrics/VictoriaMetrics) instead of cluster version
+for ingestion rates lower than 10 million of data points per second.
+Single-node version [scales perfectly](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
+with the number of CPU cores, RAM and available storage space.
+Single-node version is easier to configure and operate comparing to cluster version, so think twice before sticking to cluster version.
+
+Join [our Slack](http://slack.victoriametrics.com/) or [contact us](mailto:info@victoriametrics.com) with consulting and support questions.
+
+
+## Prominent features
+
+- Supports all the features of [single-node version](https://github.com/VictoriaMetrics/VictoriaMetrics).
+- Performance and capacity scales horizontally.
+- Supports multiple independent namespaces for time series data (aka multi-tenancy).
+
+
+## Architecture overview
+
+VictoriaMetrics cluster consists of the following services:
+
+- `vmstorage` - stores the data
+- `vminsert` - proxies the ingested data to `vmstorage` shards using consistent hashing
+- `vmselect` - performs incoming queries using the data from `vmstorage`
+
+Each service may scale independently and may run on the most suitable hardware.
+
+<img src="https://docs.google.com/drawings/d/e/2PACX-1vTvk2raU9kFgZ84oF-OKolrGwHaePhHRsZEcfQ1I_EC5AB_XPWwB392XshxPramLJ8E4bqptTnFn5LL/pub?w=1104&amp;h=746">
+
+
+## Binaries
+
+Compiled binaries for cluster version are available in the `assets` section of [releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases).
+See archives containing `cluster` word.
+
+Docker images for cluster version are available here:
+
+- `vminsert` - https://hub.docker.com/r/victoriametrics/vminsert/tags
+- `vmselect` - https://hub.docker.com/r/victoriametrics/vmselect/tags
+- `vmstorage` - https://hub.docker.com/r/victoriametrics/vmstorage/tags
+
+
+## Building from sources
+
+Source code for cluster version is available at [cluster branch](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+
+### Development Builds
+
+1. [Install go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make` from the repository root. It should build `vmstorage`, `vmselect`
+   and `vminsert` binaries and put them into the `bin` folder.
+
+
+### Production builds
+
+There is no need in installing Go on a host system since binaries are built
+inside [the official docker container for Go](https://hub.docker.com/_/golang).
+This makes reproducible builds.
+So [install docker](https://docs.docker.com/install/) and run the following command:
+
+```
+make vminsert-prod vmselect-prod vmstorage-prod
+```
+
+Production binaries are built into statically linked binaries for `GOARCH=amd64`, `GOOS=linux`.
+They are put into `bin` folder with `-prod` suffixes:
+```
+$ make vminsert-prod vmselect-prod vmstorage-prod
+$ ls -1 bin
+vminsert-prod
+vmselect-prod
+vmstorage-prod
+```
+
+### Building docker images
+
+Run `make package`. It will build the following docker images locally:
+
+* `victoriametrics/vminsert:<PKG_TAG>`
+* `victoriametrics/vmselect:<PKG_TAG>`
+* `victoriametrics/vmstorage:<PKG_TAG>`
+
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package`.
+
+
+
+## Operation
+
+### Cluster setup
+
+A minimal cluster must contain the following nodes:
+
+* a single `vmstorage` node with `-retentionPeriod` and `-storageDataPath` flags
+* a single `vminsert` node with `-storageNode=<vmstorage_host>:8400`
+* a single `vmselect` node with `-storageNode=<vmstorage_host>:8401`
+
+It is recommended to run at least two nodes for each service
+for high availability purposes.
+
+An http load balancer must be put in front of `vminsert` and `vmselect` nodes:
+- requests starting with `/insert` must be routed to port `8480` on `vminsert` nodes.
+- requests starting with `/select` must be routed to port `8481` on `vmselect` nodes.
+
+Ports may be altered by setting `-httpListenAddr` on the corresponding nodes.
+
+It is recommended setting up [monitoring](#monitoring) for the cluster.
+
+
+### Monitoring
+
+All the cluster components expose various metrics in Prometheus-compatible format at `/metrics` page on the TCP port set in `-httpListenAddr` command-line flag.
+By default the following TCP ports are used:
+- `vminsert` - 8480
+- `vmselect` - 8481
+- `vmstorage` - 8482
+
+It is recommended setting up Prometheus to scrape `/metrics` pages from all the cluster components, so they can be monitored and analyzed
+with [the official Grafana dashboard for VictoriaMetrics cluster](https://grafana.com/grafana/dashboards/11176).
+
+
+### URL format
+
+* URLs for data ingestion: `http://<vminsert>:8480/insert/<accountID>/<suffix>`, where:
+  - `<accountID>` is an arbitrary number identifying namespace for data ingestion (aka tenant)
+  - `<suffix>` may have the following values:
+     - `prometheus` - for inserting data with [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
+     - `influx/write` or `influx/api/v2/write` - for inserting data with [Influx line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/)
+
+* URLs for querying: `http://<vmselect>:8481/select/<accountID>/prometheus/<suffix>`, where:
+  - `<accountID>` is an arbitrary number identifying data namespace for the query (aka tenant)
+  - `<suffix>` may have the following values:
+    - `api/v1/query` - performs [PromQL instant query](https://prometheus.io/docs/prometheus/latest/querying/api/#instant-queries)
+    - `api/v1/query_range` - performs [PromQL range query](https://prometheus.io/docs/prometheus/latest/querying/api/#range-queries)
+    - `api/v1/series` - performs [series query](https://prometheus.io/docs/prometheus/latest/querying/api/#finding-series-by-label-matchers)
+    - `api/v1/labels` - returns a [list of label names](https://prometheus.io/docs/prometheus/latest/querying/api/#getting-label-names)
+    - `api/v1/label/<label_name>/values` - returns values for the given `<label_name>` according [to API](https://prometheus.io/docs/prometheus/latest/querying/api/#querying-label-values)
+    - `federate` - returns [federated metrics](https://prometheus.io/docs/prometheus/latest/federation/)
+    - `api/v1/export` - exports raw data. See [this article](https://medium.com/@valyala/analyzing-prometheus-data-with-external-tools-5f3e5e147639) for details
+
+* URL for time series deletion: `http://<vmselect>:8481/delete/<accountID>/prometheus/api/v1/admin/tsdb/delete_series?match[]=<timeseries_selector_for_delete>`.
+  Note that the `delete_series` handler should be used only in exceptional cases such as deletion of accidentally ingested incorrect time series. It shouldn't
+  be used on a regular basis, since it carries non-zero overhead.
+
+* `vmstorage` nodes provide the following HTTP endpoints on `8482` port:
+  - `/snapshot/create` - create [instant snapshot](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282),
+    which can be used for backups in background. Snapshots are created in `<storageDataPath>/snapshots` folder, where `<storageDataPath>` is the corresponding
+    command-line flag value.
+  - `/snapshot/list` - list available snasphots.
+  - `/snapshot/delete?snapshot=<id>` - delete the given snapshot.
+  - `/snapshot/delete_all` - delete all the snapshots.
+
+  Snapshots may be created independently on each `vmstorage` node. There is no need in synchronizing snapshots' creation
+  across `vmstorage` nodes.
+
+
+### Cluster resizing and scalability.
+
+Cluster performance and capacity scales with adding new nodes.
+
+* `vminsert` and `vmselect` nodes are stateless and may be added / removed at any time.
+  Do not forget updating the list of these nodes on http load balancer.
+  Adding more `vminsert` nodes scales data ingestion rate. See [this comment](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/175#issuecomment-536925841)
+  about ingestion rate scalability.
+  Adding more `vmselect` nodes scales select queries rate.
+* `vmstorage` nodes own the ingested data, so they cannot be removed without data loss.
+  Adding more `vmstorage` nodes scales cluster capacity.
+
+Steps to add `vmstorage` node:
+
+1. Start new `vmstorage` node with the same `-retentionPeriod` as existing nodes in the cluster.
+2. Gradually restart all the `vmselect` nodes with new `-storageNode` arg containing `<new_vmstorage_host>:8401`.
+3. Gradually restart all the `vminsert` nodes with new `-storageNode` arg containing `<new_vmstorage_host>:8400`.
+
+
+### Cluster availability
+
+* HTTP load balancer must stop routing requests to unavailable `vminsert` and `vmselect` nodes.
+* The cluster remains available if at least a single `vmstorage` node exists:
+
+  - `vminsert` re-routes incoming data from unavailable `vmstorage` nodes to healthy `vmstorage` nodes
+  - `vmselect` continues serving partial responses if at least a single `vmstorage` node is available.
+
+
+### Updating / reconfiguring cluster nodes
+
+All the node types - `vminsert`, `vmselect` and `vmstorage` - may be updated via graceful shutdown.
+Send `SIGINT` signal to the corresponding process, wait until it finishes and then start new version
+with new configs.
+
+Cluster should remain in working state if at least a single node of each type remains available during
+the update process. See [cluster availability](#cluster-availability) section for details.
+
+
+### Capacity planning
+
+Each instance type - `vminsert`, `vmselect` and `vmstorage` - can run on the most suitable hardware.
+
+#### vminsert
+
+* The recommended total number of vCPU cores for all the `vminsert` instances can be calculated from the ingestion rate: `vCPUs = ingestion_rate / 150K`.
+* The recommended number of vCPU cores per each `vminsert` instance should equal to the number of `vmstorage` instances in the cluster.
+* The amount of RAM per each `vminsert` instance should be 1GB or more. RAM is used as a buffer for spikes in ingestion rate.
+* Sometimes `-rpc.disableCompression` command-line flag on `vminsert` instances could increase ingestion capacity at the cost
+  of higher network bandwidth usage between `vminsert` and `vmstorage`.
+
+#### vmstorage
+
+* The recommended total number of vCPU cores for all the `vmstorage` instances can be calculated from the ingestion rate: `vCPUs = ingestion_rate / 150K`.
+* The recommended total amount of RAM for all the `vmstorage` instances can be calculated from the number of active time series: `RAM = active_time_series * 1KB`.
+  Time series is active if it received at least a single data point during the last hour or if it has been queried during the last hour.
+* The recommended total amount of storage space for all the `vmstorage` instances can be calculated
+  from the ingestion rate and retention: `storage_space = ingestion_rate * retention_seconds`.
+
+#### vmselect
+
+The recommended hardware for `vmselect` instances highly depends on the type of queries. Lightweight queries over small number of time series usually require
+small number of vCPU cores and small amount of RAM on `vmselect`, while heavy queries over big number of time series (>10K) usually require
+bigger number of vCPU cores and bigger amounts of RAM.
+
+
+### Helm
+
+Helm chart simplifies managing cluster version of VictoriaMetrics in Kubernetes.
+It is available in the [helm-charts](https://github.com/VictoriaMetrics/helm-charts) repository.
+
+Upgrade follows `Cluster resizing procedure` under the hood.
+
+
+### Replication and data safety
+
+VictoriaMetrics offloads replication to the underlying storage pointed by `-storageDataPath`.
+It is recommended storing data on [Google Compute Engine persistent disks](https://cloud.google.com/compute/docs/disks/#pdspecs),
+since they are protected from data loss and data corruption. They also provide consistently high performance
+and [may be resized](https://cloud.google.com/compute/docs/disks/add-persistent-disk) without downtime.
+HDD-based persistent disks should be enough for the majority of use cases.
+
+It is recommended using durable replicated persistent volumes in Kubernetes.
+
+Note that [replication doesn't save from disaster](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883).
+
+
+### Backups
+
+It is recommended performing periodical backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
+for protecting from user errors such as accidental data deletion.
+
+The following steps must be performed for each `vmstorage` node for creating a backup:
+
+1. Create an instant snapshot by navigating to `/snapshot/create` HTTP handler. It will create snapshot and return its name.
+2. Archive the created snapshot from `<-storageDataPath>/snapshots/<snapshot_name>` folder using [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/app/vmbackup/README.md).
+   The archival process doesn't interfere with `vmstorage` work, so it may be performed at any suitable time.
+3. Delete unused snapshots via `/snapshot/delete?snapshot=<snapshot_name>` or `/snapshot/delete_all` in order to free up occupied storage space.
+
+There is no need in synchronizing backups among all the `vmstorage` nodes.
+
+Restoring from backup:
+
+1. Stop `vmstorage` node with `kill -INT`.
+2. Restore data from backup using [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/app/vmrestore/README.md) into `-storageDataPath` directory.
+3. Start `vmstorage` node.
+
+
+## Community and contributions
+
+We are open to third-party pull requests provided they follow [KISS design principle](https://en.wikipedia.org/wiki/KISS_principle):
+
+- Prefer simple code and architecture.
+- Avoid complex abstractions.
+- Avoid magic code and fancy algorithms.
+- Avoid [big external dependencies](https://medium.com/@valyala/stripping-dependency-bloat-in-victoriametrics-docker-image-983fb5912b0d).
+- Minimize the number of moving parts in the distributed system.
+- Avoid automated decisions, which may hurt cluster availability, consistency or performance.
+
+Adhering `KISS` principle simplifies the resulting code and architecture, so it can be reviewed, understood and verified by many people.
+
+Due to `KISS` cluster version of VictoriaMetrics has no the following "features" popular in distributed computing world:
+
+- Fragile gossip protocols. See [failed attempt in Thanos](https://github.com/improbable-eng/thanos/blob/030bc345c12c446962225221795f4973848caab5/docs/proposals/completed/201809_gossip-removal.md).
+- Hard-to-understand-and-implement-properly [Paxos protocols](https://www.quora.com/In-distributed-systems-what-is-a-simple-explanation-of-the-Paxos-algorithm).
+- Complex replication schemes, which may go nuts in unforesseen edge cases. The replication is offloaded to the underlying durable replicated storage
+  such as [persistent disks in Google Compute Engine](https://cloud.google.com/compute/docs/disks/#pdspecs).
+- Automatic data reshuffling between storage nodes, which may hurt cluster performance and availability.
+- Automatic cluster resizing, which may cost you a lot of money if improperly configured.
+- Automatic discovering and addition of new nodes in the cluster, which may mix data between dev and prod clusters :)
+- Automatic leader election, which may result in split brain disaster on network errors.
+
+
+## Reporting bugs
+
+Report bugs and propose new features [here](https://github.com/VictoriaMetrics/VictoriaMetrics/issues).
+
+
+## Victoria Metrics Logo
+
+[Zip](VM_logo.zip) contains three folders with different image orientation (main color and inverted version).
+
+Files included in each folder:
+
+* 2 JPEG Preview files
+* 2 PNG Preview files with transparent background
+* 2 EPS Adobe Illustrator EPS10 files
+
+
+### Logo Usage Guidelines
+
+#### Font used:
+
+* Lato Black
+* Lato Regular
+
+#### Color Palette:
+
+* HEX [#110f0f](https://www.color-hex.com/color/110f0f)
+* HEX [#ffffff](https://www.color-hex.com/color/ffffff)
+
+### We kindly ask:
+
+- Please don't use any other font instead of suggested.
+- There should be sufficient clear space around the logo.
+- Do not change spacing, alignment, or relative locations of the design elements.
+- Do not change the proportions of any of the design elements or the design itself. You    may resize as needed but must retain all proportions.

docs/ExtendedPromQL.md

@@ -0,0 +1,73 @@
+# Extended PromQL
+
+VictoriaMetrics supports [standard PromQL](https://prometheus.io/docs/prometheus/latest/querying/basics/)
+including [subqueries](https://prometheus.io/blog/2019/01/28/subquery-support/).
+Additionally it supports useful extensions mentioned below.
+Try these extensions on [an editable Grafana dashboard](http://play-grafana.victoriametrics.com:3000/d/4ome8yJmz/node-exporter-on-victoriametrics-demo).
+
+- [`WITH` templates](https://play.victoriametrics.com/promql/expand-with-exprs). This feature simplifies writing and managing complex queries. Go to [`WITH` templates playground](https://victoriametrics.com/promql/expand-with-exprs) and try it.
+- Metric names and metric labels may contain escaped chars. For instance, `foo\-bar{baz\=aa="b"}` is valid expression. It returns time series with name `foo-bar` containing label `baz=aa` with value `b`. Additionally, `\xXX` escape sequence is supported, where `XX` is hexadecimal representation of escaped char.
+- `offset`, range duration and step value for range vector may refer to the current step aka `$__interval` value from Grafana.
+  For instance, `rate(metric[10i] offset 5i)` would return per-second rate over a range covering 10 previous steps with the offset of 5 steps.
+- `default` binary operator. `q1 default q2` substitutes `NaN` values from `q1` with the corresponding values from `q2`.
+- `if` binary operator. `q1 if q2` removes values from `q1` for `NaN` values from `q2`.
+- `ifnot` binary operator. `q1 ifnot q2` removes values from `q1` for non-`NaN` values from `q2`.
+- `offset` may be put anywere in the query. For instance, `sum(foo) offset 24h`.
+- Trailing commas on all the lists are allowed - label filters, function args and with expressions. For instance, the following queries are valid: `m{foo="bar",}`, `f(a, b,)`, `WITH (x=y,) x`. This simplifies maintenance of multi-line queries.
+- String literals may be concatenated. This is useful with `WITH` templates: `WITH (commonPrefix="long_metric_prefix_") {__name__=commonPrefix+"suffix1"} / {__name__=commonPrefix+"suffix2"}`.
+- Range duration in functions such as [rate](https://prometheus.io/docs/prometheus/latest/querying/functions/#rate()) may be omitted. VictoriaMetrics automatically selects range duration depending on the current step used for building the graph. For instance, the following query is valid in VictoriaMetrics: `rate(node_network_receive_bytes_total)`.
+- [Range duration](https://prometheus.io/docs/prometheus/latest/querying/basics/#range-vector-selectors) and [offset](https://prometheus.io/docs/prometheus/latest/querying/basics/#offset-modifier) may be fractional. For instance, `rate(node_network_receive_bytes_total[1.5m] offset 0.5d)`.
+- Comments starting with `#` and ending with newline. For instance, `up # this is a comment for 'up' metric`.
+- Rollup functions - `rollup(m[d])`, `rollup_rate(m[d])`, `rollup_deriv(m[d])`, `rollup_increase(m[d])`, `rollup_delta(m[d])` - return `min`, `max` and `avg`
+  values for all the `m` data points over `d` duration.
+- `rollup_candlestick(m[d])` - returns `open`, `close`, `low` and `high` values (OHLC) for all the `m` data points over `d` duration. This function is useful for financial applications.
+- `union(q1, ... qN)` function for building multiple graphs for `q1`, ... `qN` subqueries with a single query. The `union` function name may be skipped -
+  the following queries are equivalent: `union(q1, q2)` and `(q1, q2)`.
+- `ru(freeResources, maxResources)` function for returning resource utilization percentage in the range `0% - 100%`. For instance, `ru(node_memory_MemFree_bytes, node_memory_MemTotal_bytes)` returns memory utilization over [node_exporter](https://github.com/prometheus/node_exporter) metrics.
+- `ttf(slowlyChangingFreeResources)` function for returning the time in seconds when the given `slowlyChangingFreeResources` expression reaches zero. For instance, `ttf(node_filesystem_avail_byte)` returns the time to storage space exhaustion. This function may be useful for capacity planning.
+- Functions for label manipulation:
+  - `alias(q, name)` for setting metric name across all the time series `q`.
+  - `label_set(q, label1, value1, ... labelN, valueN)` for setting the given values for the given labels on `q`.
+  - `label_del(q, label1, ... labelN)` for deleting the given labels from `q`.
+  - `label_keep(q, label1, ... labelN)` for deleting all the labels except the given labels from `q`.
+  - `label_copy(q, src_label1, dst_label1, ... src_labelN, dst_labelN)` for copying label values from `src_*` to `dst_*`.
+  - `label_move(q, src_label1, dst_label1, ... src_labelN, dst_labelN)` for moving label values from `src_*` to `dst_*`.
+  - `label_transform(q, label, regexp, replacement)` for replacing all the `regexp` occurences with `replacement` in the `label` values from `q`.
+  - `label_value(q, label)` - returns numeric values for the given `label` from `q`.
+- `step()` function for returning the step in seconds used in the query.
+- `start()` and `end()` functions for returning the start and end timestamps of the `[start ... end]` range used in the query.
+- `integrate(m[d])` for returning integral over the given duration `d` for the given metric `m`.
+- `ideriv(m)` - for calculating `instant` derivative for `m`.
+- `deriv_fast(m[d])` - for calculating `fast` derivative for `m` based on the first and the last points from duration `d`.
+- `running_` functions - `running_sum`, `running_min`, `running_max`, `running_avg` - for calculating [running values](https://en.wikipedia.org/wiki/Running_total) on the selected time range.
+- `range_` functions - `range_sum`, `range_min`, `range_max`, `range_avg`, `range_first`, `range_last`, `range_median`, `range_quantile` - for calculating global value over the selected time range.
+- `smooth_exponential(q, sf)` - smooths `q` using [exponential moving average](https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average) with the given smooth factor `sf`.
+- `remove_resets(q)` - removes counter resets from `q`.
+- `lag(q[d])` - returns lag between the current timestamp and the timestamp from the previous data point in `q` over `d`.
+- `lifetime(q[d])` - returns lifetime of `q` over `d` in seconds. It is expected that `d` exceeds the lifetime of `q`.
+- `scrape_interval(q[d])` - returns the average interval in seconds between data points of `q` over `d` aka `scrape interval`.
+- Trigonometric functions - `sin(q)`, `cos(q)`, `asin(q)`, `acos(q)` and `pi()`.
+- `median_over_time(m[d])` - calculates median values for `m` over `d` time window. Shorthand to `quantile_over_time(0.5, m[d])`.
+- `median(q)` - median aggregate. Shorthand to `quantile(0.5, q)`.
+- `limitk(k, q)` - limits the number of time series returned from `q` to `k`.
+- `keep_last_value(q)` - fills missing data (gaps) in `q` with the previous value.
+- `distinct_over_time(m[d])` - returns distinct number of values for `m` data points over `d` duration.
+- `distinct(q)` - returns a time series with the number of unique values for each timestamp in `q`.
+- `sum2_over_time(m[d])` - returns sum of squares for all the `m` values over `d` duration.
+- `sum2(q)` - returns a time series with sum of square values for each timestamp in `q`.
+- `geomean_over_time(m[d])` - returns [geomean](https://en.wikipedia.org/wiki/Geometric_mean) value for all the `m` value over `d` duration.
+- `geomean(q)` - returns a time series with [geomean](https://en.wikipedia.org/wiki/Geometric_mean) value for each timestamp in `q`.
+- `rand()`, `rand_normal()` and `rand_exponential()` functions - for generating pseudo-random series with even, normal and exponential distribution.
+- `increases_over_time(m[d])` and `decreases_over_time(m[d])` - returns the number of `m` increases or decreases over the given duration `d`.
+- `prometheus_buckets(q)` - converts [VictoriaMetrics histogram](https://godoc.org/github.com/VictoriaMetrics/metrics#Histogram) buckets to Prometheus buckets with `le` labels.
+- `histogram(q)` - calculates aggregate histogram over `q` time series for each point on the graph.
+- `topk_*` and `bottomk_*` aggregate functions, which return up to K time series. Note that the standard `topk` function may return more than K time series -
+   see [this article](https://www.robustperception.io/graph-top-n-time-series-in-grafana) for details.
+   - `topk_min(k, q)` - returns top K time series with the max minimums on the given time range
+   - `topk_max(k, q)` - returns top K time series with the max maximums on the given time range
+   - `topk_avg(k, q)` - returns top K time series with the max averages on the given time range
+   - `topk_median(k, q)` - returns top K time series with the max medians on the given time range
+   - `bottomk_min(k, q)` - returns bottom K time series with the min minimums on the given time range
+   - `bottomk_max(k, q)` - returns bottom K time series with the min maximums on the given time range
+   - `bottomk_avg(k, q)` - returns bottom K time series with the min averages on the given time range
+   - `bottomk_median(k, q)` - returns bottom K time series with the min medians on the given time range

docs/FAQ.md

@@ -0,0 +1,160 @@
+# FAQ
+
+### What is the main purpose of VictoriaMetrics?
+
+To provide the best long-term [remote storage](https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage) solution for [Prometheus](https://prometheus.io/).
+
+
+### Which features does VictoriaMetrics have?
+
+* Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
+  Additionally, VictoriaMetrics extends PromQL with opt-in [useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
+* High performance and good scalability for both [inserts](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
+  and [selects](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4).
+  [Outperforms InfluxDB and TimescaleDB by up to 20x](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+* [Uses 10x less RAM than InfluxDB](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) when working with millions of unique time series (aka high cardinality).
+* High data compression, so [up to 70x more data points](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4)
+  may be crammed into a limited storage comparing to TimescaleDB.
+* Optimized for storage with high-latency IO and low iops (HDD and network storage in AWS, Google Cloud, Microsoft Azure, etc). See [graphs from these benchmarks](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b).
+* A single-node VictoriaMetrics may substitute moderately sized clusters built with competing solutions such as Thanos, Uber M3, Cortex, InfluxDB or TimescaleDB.
+  See [vertical scalability benchmarks](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
+  and [comparing Thanos to VictoriaMetrics](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683).
+* Easy operation:
+  * VictoriaMetrics consists of a single executable without external dependencies.
+  * All the configuration is done via explicit command-line flags with reasonable defaults.
+  * All the data is stored in a single directory pointed by `-storageDataPath` flag.
+  * Easy backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
+* Storage is protected from corruption on unclean shutdown (i.e. hardware reset or `kill -9`) thanks to [the storage architecture](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
+* Supports metrics' ingestion and backfilling via the following protocols:
+  * [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
+  * [InfluxDB line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/)
+  * [Graphite plaintext protocol](https://graphite.readthedocs.io/en/latest/feeding-carbon.html) with [tags](https://graphite.readthedocs.io/en/latest/tags.html#carbon)
+    if `-graphiteListenAddr` is set.
+  * [OpenTSDB put message](http://opentsdb.net/docs/build/html/api_telnet/put.html) if `-opentsdbListenAddr` is set.
+* Ideally works with big amounts of time series data from IoT sensors, connected car sensors and industrial sensors.
+* Has open source [cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+
+### Which clients do you target?
+
+The following Prometheus users may be interested in VictoriaMetrics:
+- Users who don't want to bother with Prometheus' local storage operational burden - backups, replication, capacity planning, scalability, etc.
+- Users with multiple Prometheus instances who want performing arbitrary queries over all the metrics collected by their Prometheus instances (aka `global querying view`).
+- Users who want reducing costs for storing huge amounts of time series data.
+
+
+### How to start using VictoriaMetrics?
+
+Start with [single-node version](Single-server-VictoriaMetrics). It is easy to configure and operate. It should fit the majority of use cases.
+
+
+### Is it safe to enable [remote write storage](https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage) in Prometheus?
+
+Yes. Prometheus continues writing data to local storage after enabling remote storage write, so all the existing local storage data
+and new data is available for querying via Prometheus as usual.
+
+
+### How does VictoriaMetrics compare to other clustered TSDBs on top of Prometheus such as [M3 from Uber](https://eng.uber.com/m3/), [Thanos](https://github.com/improbable-eng/thanos), [Cortex](https://github.com/cortexproject/cortex), etc.?
+
+VictoriaMetrics is simpler, faster, more cost-effective and it provides [useful extensions for PromQL](ExtendedPromQL). The simplicity is twofold:
+- It is simpler to configure and operate. There is no need in configuring third-party [sidecars](https://github.com/improbable-eng/thanos/blob/master/docs/components/sidecar.md)
+  or fighting with [gossip protocol](https://github.com/improbable-eng/thanos/blob/master/docs/proposals/completed/201809_gossip-removal.md).
+- VictoriaMetrics has simpler architecture, which means less bugs and more useful features in a long run comparing to competing TSDBs.
+
+See [comparing Thanos to VictoriaMetrics cluster](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683).
+
+
+### How does VictoriaMetrics compare to [InfluxDB](https://www.influxdata.com/time-series-platform/influxdb/)?
+
+VictoriaMetrics requires [10x less RAM](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) and it [works faster](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+It is easier to configure and operate. It provides [better query language](https://medium.com/@valyala/promql-tutorial-for-beginners-9ab455142085) than InfluxQL or Flux.
+
+
+### How does VictoriaMetrics compare to [TimescaleDB](https://www.timescale.com/)?
+
+TimescaleDB insists on using SQL as a query language. While SQL is more powerful than PromQL, this power is rarely required during typical TSDB usage. Real-world queries usually [look clearer and simpler when written in PromQL than in SQL](https://medium.com/@valyala/promql-tutorial-for-beginners-9ab455142085).
+Additionally, VictoriaMetrics requires [up to 70x less storage space comparing to TimescaleDB](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4) for storing the same amount of time series data.
+
+
+### Does VictoriaMetrics use Prometheus technologies like other clustered TSDBs built on top of Prometheus such as [M3 from Uber](https://eng.uber.com/m3/), [Thanos](https://github.com/improbable-eng/thanos), [Cortex](https://github.com/cortexproject/cortex)?
+
+No. VictoriaMetrics core is written in Go from scratch by [fasthttp](https://github.com/valyala/fasthttp) [author](https://github.com/valyala).
+The architecture is [optimized for storing and querying large amounts of time series data with high cardinality](https://medium.com/devopslinks/victoriametrics-creating-the-best-remote-storage-for-prometheus-5d92d66787ac). VictoriaMetrics storage uses [certain ideas from ClickHouse](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282). Special thanks to [Alexey Milovidov](https://github.com/alexey-milovidov).
+
+
+### Are there performance comparisons with other solutions?
+
+Yes:
+
+* [Measuring vertical scalability for time series databases: VictoriaMetrics vs InfluxDB vs TimescaleDB](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+* [Measuring insert performance on high-cardinality time series: VictoriaMetrics vs InfluxDB](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893)
+* [TSBS benchmark on high-cardinality time series: VictoriaMetrics vs InfluxDB vs TimescaleDB](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
+* [Standard TSBS benchmark: VictoriaMetrics vs InfluxDB vs TimescaleDB](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4)
+
+
+### What is the pricing for VictoriaMetrics?
+
+The following versions are open source and free:
+* [Single-node version](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics).
+* [Cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+We provide commercial support for both versions. [Contact us](mailto:info@victoriametrics.com) for the pricing.
+
+The following versions are commercial:
+* Managed cluster in the Cloud.
+* SaaS version.
+
+[Contact us](mailto:info@victoriametrics.com) for the pricing.
+
+
+### Why VictoriaMetrics doesn't support [Prometheus remote read API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#%3Cremote_read%3E)?
+
+Remote read API requires transferring all the raw data for all the requested metrics over the given time range. For instance,
+if a query covers 1000 metrics with 10K values each, then the remote read API had to return `1000*10K`=10M metric values to Prometheus.
+This is slow and expensive.
+Prometheus remote read API isn't intended for querying foreign data aka `global query view`. See [this issue](https://github.com/prometheus/prometheus/issues/4456) for details.
+
+So just query VictoriaMetrics directly via [Prometheus Querying API](https://prometheus.io/docs/prometheus/latest/querying/api/)
+or via [Prometheus datasoruce in Grafana](http://docs.grafana.org/features/datasources/prometheus/).
+
+
+### Does VictoriaMetrics deduplicate data from Prometheus instances scraping the same targets (aka `HA pairs`)?
+
+Data from all the Prometheus instances is saved in VictoriaMetrics without deduplication.
+
+The deduplication for Prometheus HA pair may be easily implemented on top of VictoriaMetrics with the following steps:
+
+1) Run multiple VictoriaMetrics instances in multiple availability zones (datacenters).
+2) Configure each Prometheus from each HA pair to write data to VictoriaMetrics in distinct availability zone.
+3) Put [Promxy](https://github.com/jacksontj/promxy) in front of all the VictoriaMetrics instances.
+4) Send queries to Promxy - it will deduplicate data from VictoriaMetrics instances behind it.
+
+
+### Where is the source code of VictoriaMetrics?
+
+Source code for the following versions is available in the following places:
+* [Single-node version](https://github.com/VictoriaMetrics/VictoriaMetrics).
+* [Cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+
+### Does VictoriaMetrics fit for data from IoT sensors and industrial sensors?
+
+VictoriaMetrics is able to handle data from hundreds of millions of IoT sensors and industrial sensors.
+It supports [high cardinality data](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b),
+perfectly [scales up on a single node](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
+and scales horizontally to multiple nodes.
+
+
+### Where can I ask questions about VictoriaMetrics?
+
+See [VictoriaMetrics-users group](https://groups.google.com/forum/#!forum/victorametrics-users).
+
+
+### Where can I file bugs and feature requests regarding VictoriaMetrics?
+
+File bugs and feature requests [here](https://github.com/VictoriaMetrics/VictoriaMetrics/issues).
+
+
+### Are you looking for investors?
+
+Yes. [Mail us](mailto:info@victoriametrics.com) if you are interested in.

docs/Home.md

@@ -0,0 +1,11 @@
+# Docs
+
+* [Quick start](Quick-Start)
+* [`WITH` templates playground](https://play.victoriametrics.com/promql/expand-with-exprs)
+* [Grafana playground](http://play-grafana.victoriametrics.com:3000/d/4ome8yJmz/node-exporter-on-victoriametrics-demo)
+* [Extended PromQL](ExtendedPromQL)
+* [Single-node version](Single-server-VictoriaMetrics)
+* [FAQ](FAQ)
+* [Cluster version](Cluster-VictoriaMetrics)
+* [Articles](Articles)
+

docs/Quick-Start.md

@@ -0,0 +1,27 @@
+# Quick Start
+
+1. Download the latest VictoriaMetrics release from [releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases),
+   from [Docker hub](https://hub.docker.com/r/valyala/victoria-metrics/)
+   or [build it from sources](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics#how-to-build-from-sources).
+
+2. Run the binary or Docker image with the desired command-line flags. Pass `-help` in order to see description for all the available flags
+   and their default values. Default flag values should fit the majoirty of cases. The minimum required flags to configure are:
+
+   * `-storageDataPath` - path to directory where VictoriaMetrics stores all the data.
+   * `-retentionPeriod` - data retention in months.
+
+   For instance:
+
+   `./victoria-metrics-prod -storageDataPath=/var/lib/victoria-metrics-data -retentionPeriod=3`
+
+   See [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/43) in order to configure VictoriaMetrics as OS service.
+   It is recommended setting up [VictoriaMetrics monitoring](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#monitoring).
+
+3. Configure all the Prometheus instances to write data to VictoriaMetrics.
+   See [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics#prometheus-setup).
+
+4. Configure Grafana to query VictoriaMetrics instead of Prometheus.
+   See [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics#grafana-setup).
+
+
+There is also [cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster) and [SaaS playground](https://play.victoriametrics.com/signIn).

docs/Release-Guide.md

@@ -0,0 +1,44 @@
+Release process guidance
+
+## Release version and Docker images
+
+1. Create release tag with `git tag v1.xx.y`.
+2. Run `make release` for creating `*.tar.gz` release archive with the corresponding `_checksums.txt` inside `bin` directory.
+3. Run `make publish` for creating and publishing Docker images.
+4. Push release tag to https://github.com/VictoriaMetrics/VictoriaMetrics : `git push origin v1.xx.y`.
+5. Go to https://github.com/VictoriaMetrics/VictoriaMetrics/releases , create new release from the pushed tag on step 4
+   and upload `*.tar.gz` archive with the corresponding `_checksums.txt` from step 2.
+
+
+## Helm Charts
+
+The helm chart repository [https://github.com/VictoriaMetrics/helm-charts/](https://github.com/VictoriaMetrics/helm-charts/)
+
+
+### Bump the version of images. 
+In that case, don't need to bump the helm chart version
+
+1. Need to update [`values.yaml`](https://github.com/VictoriaMetrics/helm-charts/blob/master/charts/victoria-metrics-cluster/values.yaml), bump version for `vmselect`, `vminsert` and `vmstorage`
+2. Specify the correct version in [`Chart.yaml`](https://github.com/VictoriaMetrics/helm-charts/blob/master/charts/victoria-metrics-cluster/Chart.yaml)
+3. Update version [README.md](https://github.com/VictoriaMetrics/helm-charts/blob/master/charts/victoria-metrics-cluster/README.md), specify the new version in the documentation
+4. Push changes to master. `master` is a source of truth
+5. Rebase `master` into `gh-pages` branch
+6. Run `make package` which creates or updates zip file with the packed chart
+7. Run `make merge`. It creates or updates metadata for charts in index.yaml 
+8. Push the changes to `gh-pages` branch 
+
+### Updating the chart.
+1. Update chart version in [`Chart.yaml`](https://github.com/VictoriaMetrics/helm-charts/blob/master/charts/victoria-metrics-cluster/Chart.yaml)
+2. Update [README.md](https://github.com/VictoriaMetrics/helm-charts/blob/master/charts/victoria-metrics-cluster/README.md) file, reflect changes in the documentation.
+3. Repeat the procedure from step _4_ previous section.
+
+
+## Wiki pages
+
+All changes from `docs` folder and `.md` extension automatically push to Wiki
+
+**_Note_**: no vice versa, direct changes on Wiki will be overitten after any changes in `docs/*.md` 
+
+## Github pages
+
+All changes in `README.md`, `docs` folder and `.md` extension automatically push to Wiki

docs/Single-server-VictoriaMetrics.md

@@ -0,0 +1,817 @@
+# Single-node
+
+VictoriaMetrics is fast, cost-effective and scalable time-series database. It can be used as long-term remote storage for Prometheus.
+It is available in [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases),
+[docker images](https://hub.docker.com/r/victoriametrics/victoria-metrics/) and
+in [source code](https://github.com/VictoriaMetrics/VictoriaMetrics).
+
+Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+
+## Prominent features
+
+* Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
+  Additionally, VictoriaMetrics extends PromQL with opt-in [useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
+* Supports global query view. Multiple Prometheus instances may write data into VictoriaMetrics. Later this data may be used in a single query.
+* High performance and good scalability for both [inserts](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b)
+  and [selects](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4).
+  [Outperforms InfluxDB and TimescaleDB by up to 20x](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+* [Uses 10x less RAM than InfluxDB](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) when working with millions of unique time series (aka high cardinality).
+* Optimized for time series with high churn rate. Think about [prometheus-operator](https://github.com/coreos/prometheus-operator) metrics from frequent deployments in Kubernetes.
+* High data compression, so [up to 70x more data points](https://medium.com/@valyala/when-size-matters-benchmarking-victoriametrics-vs-timescale-and-influxdb-6035811952d4)
+  may be crammed into limited storage comparing to TimescaleDB.
+* Optimized for storage with high-latency IO and low IOPS (HDD and network storage in AWS, Google Cloud, Microsoft Azure, etc). See [graphs from these benchmarks](https://medium.com/@valyala/high-cardinality-tsdb-benchmarks-victoriametrics-vs-timescaledb-vs-influxdb-13e6ee64dd6b).
+* A single-node VictoriaMetrics may substitute moderately sized clusters built with competing solutions such as Thanos, Uber M3, Cortex, InfluxDB or TimescaleDB.
+  See [vertical scalability benchmarks](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
+  and [comparing Thanos to VictoriaMetrics cluster](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683).
+* Easy operation:
+  * VictoriaMetrics consists of a single [small executable](https://medium.com/@valyala/stripping-dependency-bloat-in-victoriametrics-docker-image-983fb5912b0d) without external dependencies.
+  * All the configuration is done via explicit command-line flags with reasonable defaults.
+  * All the data is stored in a single directory pointed by `-storageDataPath` flag.
+  * Easy and fast backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
+  to S3 or GCS with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md) / [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md).
+  See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
+* Storage is protected from corruption on unclean shutdown (i.e. OOM, hardware reset or `kill -9`) thanks to [the storage architecture](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
+* Supports metrics' ingestion and [backfilling](#backfilling) via the following protocols:
+  * [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
+  * [InfluxDB line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/)
+  * [Graphite plaintext protocol](https://graphite.readthedocs.io/en/latest/feeding-carbon.html) with [tags](https://graphite.readthedocs.io/en/latest/tags.html#carbon)
+    if `-graphiteListenAddr` is set.
+  * [OpenTSDB put message](http://opentsdb.net/docs/build/html/api_telnet/put.html) if `-opentsdbListenAddr` is set.
+  * [HTTP OpenTSDB /api/put requests](http://opentsdb.net/docs/build/html/api_http/put.html) if `-opentsdbHTTPListenAddr` is set.
+* Ideally works with big amounts of time series data from Kubernetes, IoT sensors, connected cars, industrial telemetry, financial data and various Enterprise workloads.
+* Has open source [cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
+
+
+## Operation
+
+
+### Table of contents
+
+  - [How to start VictoriaMetrics](#how-to-start-victoriametrics)
+  - [Prometheus setup](#prometheus-setup)
+  - [Grafana setup](#grafana-setup)
+  - [How to upgrade VictoriaMetrics?](#how-to-upgrade-victoriametrics)
+  - [How to apply new config to VictoriaMetrics?](#how-to-apply-new-config-to-victoriametrics)
+  - [How to send data from InfluxDB-compatible agents such as Telegraf?](#how-to-send-data-from-influxdb-compatible-agents-such-as-telegraf)
+  - [How to send data from Graphite-compatible agents such as StatsD?](#how-to-send-data-from-graphite-compatible-agents-such-as-statsd)
+  - [Querying Graphite data](#querying-graphite-data)
+  - [How to send data from OpenTSDB-compatible agents?](#how-to-send-data-from-opentsdb-compatible-agents)
+  - [How to build from sources](#how-to-build-from-sources)
+    - [Development build](#development-build)
+    - [Production build](#production-build)
+    - [ARM build](#arm-build)
+    - [Pure Go build (CGO_ENABLED=0)](#pure-go-build-cgo_enabled0)
+    - [Building docker images](#building-docker-images)
+  - [Start with docker-compose](#start-with-docker-compose)
+  - [Setting up service](#setting-up-service)
+  - [Third-party contributions](#third-party-contributions)
+  - [How to work with snapshots?](#how-to-work-with-snapshots)
+  - [How to delete time series?](#how-to-delete-time-series)
+  - [How to export time series?](#how-to-export-time-series)
+  - [Federation](#federation)
+  - [Capacity planning](#capacity-planning)
+  - [High availability](#high-availability)
+  - [Multiple retentions](#multiple-retentions)
+  - [Downsampling](#downsampling)
+  - [Multi-tenancy](#multi-tenancy)
+  - [Scalability and cluster version](#scalability-and-cluster-version)
+  - [Alerting](#alerting)
+  - [Security](#security)
+  - [Tuning](#tuning)
+  - [Monitoring](#monitoring)
+  - [Troubleshooting](#troubleshooting)
+  - [Backfilling](#backfilling)
+  - [Profiling](#profiling)
+- [Integrations](#integrations)
+- [Roadmap](#roadmap)
+- [Contacts](#contacts)
+- [Community and contributions](#community-and-contributions)
+- [Reporting bugs](#reporting-bugs)
+- [Victoria Metrics Logo](#victoria-metrics-logo)
+  - [Logo Usage Guidelines](#logo-usage-guidelines)
+    - [Font used:](#font-used)
+    - [Color Palette:](#color-palette)
+  - [We kindly ask:](#we-kindly-ask)
+
+
+### How to start VictoriaMetrics
+
+Just start VictoriaMetrics [executable](https://github.com/VictoriaMetrics/VictoriaMetrics/releases)
+or [docker image](https://hub.docker.com/r/victoriametrics/victoria-metrics/) with the desired command-line flags.
+
+The following command-line flags are used the most:
+
+* `-storageDataPath` - path to data directory. VictoriaMetrics stores all the data in this directory. Default path is `victoria-metrics-data` in current working directory.
+* `-retentionPeriod` - retention period in months for the data. Older data is automatically deleted. Default period is 1 month.
+* `-httpListenAddr` - TCP address to listen to for http requests. By default, it listens port `8428` on all the network interfaces.
+* `-graphiteListenAddr` - TCP and UDP address to listen to for Graphite data. By default, it is disabled.
+* `-opentsdbListenAddr` - TCP and UDP address to listen to for OpenTSDB data over telnet protocol. By default, it is disabled.
+* `-opentsdbHTTPListenAddr` - TCP address to listen to for HTTP OpenTSDB data over `/api/put`. By default, it is disabled.
+
+Pass `-help` to see all the available flags with description and default values.
+
+It is recommended setting up [monitoring](#monitoring) for VictoriaMetrics.
+
+
+### Prometheus setup
+
+Add the following lines to Prometheus config file (it is usually located at `/etc/prometheus/prometheus.yml`):
+
+```yml
+remote_write:
+  - url: http://<victoriametrics-addr>:8428/api/v1/write
+    queue_config:
+      max_samples_per_send: 10000
+      max_shards: 30
+```
+
+Substitute `<victoriametrics-addr>` with the hostname or IP address of VictoriaMetrics.
+Then apply the new config via the following command:
+
+```
+kill -HUP `pidof prometheus`
+```
+
+Prometheus writes incoming data to local storage and replicates it to remote storage in parallel.
+This means the data remains available in local storage for `--storage.tsdb.retention.time` duration
+even if remote storage is unavailable.
+
+If you plan to send data to VictoriaMetrics from multiple Prometheus instances, then add the following lines into `global` section
+of [Prometheus config](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#configuration-file):
+
+```yml
+global:
+  external_labels:
+    datacenter: dc-123
+```
+
+This instructs Prometheus to add `datacenter=dc-123` label to each time series sent to remote storage.
+The label name may be arbitrary - `datacenter` is just an example. The label value must be unique
+across Prometheus instances, so those time series may be filtered and grouped by this label.
+
+
+It is recommended upgrading Prometheus to [v2.12.0](https://github.com/prometheus/prometheus/releases) or newer,
+since the previous versions may have issues with `remote_write`.
+
+
+### Grafana setup
+
+Create [Prometheus datasource](http://docs.grafana.org/features/datasources/prometheus/) in Grafana with the following Url:
+
+```
+http://<victoriametrics-addr>:8428
+```
+
+Substitute `<victoriametrics-addr>` with the hostname or IP address of VictoriaMetrics.
+
+Then build graphs with the created datasource using [Prometheus query language](https://prometheus.io/docs/prometheus/latest/querying/basics/).
+VictoriaMetrics supports native PromQL and [extends it with useful features](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/ExtendedPromQL).
+
+
+### How to upgrade VictoriaMetrics?
+
+It is safe upgrading VictoriaMetrics to new versions unless [release notes](https://github.com/VictoriaMetrics/VictoriaMetrics/releases)
+say otherwise. It is recommended performing regular upgrades to the latest version,
+since it may contain important bug fixes, performance optimizations or new features.
+
+Follow the following steps during the upgrade:
+
+1) Send `SIGINT` signal to VictoriaMetrics process in order to gracefully stop it.
+2) Wait until the process stops. This can take a few seconds.
+3) Start the upgraded VictoriaMetrics.
+
+Prometheus doesn't drop data during VictoriaMetrics restart.
+See [this article](https://grafana.com/blog/2019/03/25/whats-new-in-prometheus-2.8-wal-based-remote-write/) for details.
+
+
+### How to apply new config to VictoriaMetrics?
+
+VictoriaMetrics must be restarted for applying new config:
+
+1) Send `SIGINT` signal to VictoriaMetrics process in order to gracefully stop it.
+2) Wait until the process stops. This can take a few seconds.
+3) Start VictoriaMetrics with the new config.
+
+Prometheus doesn't drop data during VictoriaMetrics restart.
+See [this article](https://grafana.com/blog/2019/03/25/whats-new-in-prometheus-2.8-wal-based-remote-write/) for details.
+
+
+### How to send data from InfluxDB-compatible agents such as [Telegraf](https://www.influxdata.com/time-series-platform/telegraf/)?
+
+Just use `http://<victoriametric-addr>:8428` url instead of InfluxDB url in agents' configs.
+For instance, put the following lines into `Telegraf` config, so it sends data to VictoriaMetrics instead of InfluxDB:
+
+```
+[[outputs.influxdb]]
+  urls = ["http://<victoriametrics-addr>:8428"]
+```
+
+Do not forget substituting `<victoriametrics-addr>` with the real address where VictoriaMetrics runs.
+
+VictoriaMetrics maps Influx data using the following rules:
+* [`db` query arg](https://docs.influxdata.com/influxdb/v1.7/tools/api/#write-http-endpoint) is mapped into `db` label value
+  unless `db` tag exists in the Influx line.
+* Field names are mapped to time series names prefixed with `{measurement}{separator}` value,
+  where `{separator}` equals to `_` by default. It can be changed with `-influxMeasurementFieldSeparator` command-line flag.
+  See also `-influxSkipSingleField` command-line flag. If `{measurement}` is empty, then time series names correspond to field names.
+* Field values are mapped to time series values.
+* Tags are mapped to Prometheus labels as-is.
+
+For example, the following Influx line:
+
+```
+foo,tag1=value1,tag2=value2 field1=12,field2=40
+```
+
+is converted into the following Prometheus data points:
+
+```
+foo_field1{tag1="value1", tag2="value2"} 12
+foo_field2{tag1="value1", tag2="value2"} 40
+```
+
+Example for writing data with [Influx line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/)
+to local VictoriaMetrics using `curl`:
+
+```
+curl -d 'measurement,tag1=value1,tag2=value2 field1=123,field2=1.23' -X POST 'http://localhost:8428/write'
+```
+
+An arbitrary number of lines delimited by '\n' may be sent in a single request.
+After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
+
+```
+curl -G 'http://localhost:8428/api/v1/export' -d 'match={__name__=~"measurement_.*"}'
+```
+
+The `/api/v1/export` endpoint should return the following response:
+
+```
+{"metric":{"__name__":"measurement_field1","tag1":"value1","tag2":"value2"},"values":[123],"timestamps":[1560272508147]}
+{"metric":{"__name__":"measurement_field2","tag1":"value1","tag2":"value2"},"values":[1.23],"timestamps":[1560272508147]}
+```
+
+Note that Influx line protocol expects [timestamps in *nanoseconds* by default](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/#timestamp),
+while VictoriaMetrics stores them with *milliseconds* precision.
+
+
+### How to send data from Graphite-compatible agents such as [StatsD](https://github.com/etsy/statsd)?
+
+1) Enable Graphite receiver in VictoriaMetrics by setting `-graphiteListenAddr` command line flag. For instance,
+the following command will enable Graphite receiver in VictoriaMetrics on TCP and UDP port `2003`:
+
+```
+/path/to/victoria-metrics-prod -graphiteListenAddr=:2003
+```
+
+2) Use the configured address in Graphite-compatible agents. For instance, set `graphiteHost`
+to the VictoriaMetrics host in `StatsD` configs.
+
+
+Example for writing data with Graphite plaintext protocol to local VictoriaMetrics using `nc`:
+
+```
+echo "foo.bar.baz;tag1=value1;tag2=value2 123 `date +%s`" | nc -N localhost 2003
+```
+
+VictoriaMetrics sets the current time if the timestamp is omitted.
+An arbitrary number of lines delimited by `\n` may be sent in one go.
+After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
+
+```
+curl -G 'http://localhost:8428/api/v1/export' -d 'match=foo.bar.baz'
+```
+
+The `/api/v1/export` endpoint should return the following response:
+
+```
+{"metric":{"__name__":"foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123],"timestamps":[1560277406000]}
+```
+
+
+### Querying Graphite data
+
+Data sent to VictoriaMetrics via `Graphite plaintext protocol` may be read either via
+[Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/)
+or via [go-graphite/carbonapi](https://github.com/go-graphite/carbonapi/blob/master/cmd/carbonapi/carbonapi.example.prometheus.yaml).
+
+
+
+### How to send data from OpenTSDB-compatible agents?
+
+VictoriaMetrics supports [telnet put protocol](http://opentsdb.net/docs/build/html/api_telnet/put.html)
+and [HTTP /api/put requests](http://opentsdb.net/docs/build/html/api_http/put.html) for ingesting OpenTSDB data.
+
+#### Sending data via `telnet put` protocol
+
+1) Enable OpenTSDB receiver in VictoriaMetrics by setting `-opentsdbListenAddr` command line flag. For instance,
+the following command enables OpenTSDB receiver in VictoriaMetrics on TCP and UDP port `4242`:
+
+```
+/path/to/victoria-metrics-prod -opentsdbListenAddr=:4242
+```
+
+2) Send data to the given address from OpenTSDB-compatible agents.
+
+
+Example for writing data with OpenTSDB protocol to local VictoriaMetrics using `nc`:
+
+```
+echo "put foo.bar.baz `date +%s` 123 tag1=value1 tag2=value2" | nc -N localhost 4242
+```
+
+An arbitrary number of lines delimited by `\n` may be sent in one go.
+After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
+
+```
+curl -G 'http://localhost:8428/api/v1/export' -d 'match=foo.bar.baz'
+```
+
+The `/api/v1/export` endpoint should return the following response:
+
+```
+{"metric":{"__name__":"foo.bar.baz","tag1":"value1","tag2":"value2"},"values":[123],"timestamps":[1560277292000]}
+```
+
+
+#### Sending OpenTSDB data via HTTP `/api/put` requests
+
+1) Enable HTTP server for OpenTSDB `/api/put` requests by setting `-opentsdbHTTPListenAddr` command line flag. For instance,
+the following command enables OpenTSDB HTTP server on port `4242`:
+
+```
+/path/to/victoria-metrics-prod -opentsdbHTTPListenAddr=:4242
+```
+
+2) Send data to the given address from OpenTSDB-compatible agents.
+
+Example for writing a single data point:
+
+```
+curl -H 'Content-Type: application/json' -d '{"metric":"x.y.z","value":45.34,"tags":{"t1":"v1","t2":"v2"}}' http://localhost:4242/api/put
+```
+
+Example for writing multiple data points in a single request:
+
+```
+curl -H 'Content-Type: application/json' -d '[{"metric":"foo","value":45.34},{"metric":"bar","value":43}]' http://localhost:4242/api/put
+```
+
+After that the data may be read via [/api/v1/export](#how-to-export-time-series) endpoint:
+
+```
+curl -G 'http://localhost:8428/api/v1/export' -d 'match[]=x.y.z' -d 'match[]=foo' -d 'match[]=bar'
+```
+
+The `/api/v1/export` endpoint should return the following response:
+
+```
+{"metric":{"__name__":"foo"},"values":[45.34],"timestamps":[1566464846000]}
+{"metric":{"__name__":"bar"},"values":[43],"timestamps":[1566464846000]}
+{"metric":{"__name__":"x.y.z","t1":"v1","t2":"v2"},"values":[45.34],"timestamps":[1566464763000]}
+```
+
+
+### How to build from sources
+
+We recommend using either [binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) or
+[docker images](https://hub.docker.com/r/victoriametrics/victoria-metrics/) instead of building VictoriaMetrics
+from sources. Building from sources is reasonable when developing additional features specific
+to your needs.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make victoria-metrics` from the root folder of the repository.
+   It builds `victoria-metrics` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make victoria-metrics-prod` from the root folder of the repository.
+   It builds `victoria-metrics-prod` binary and puts it into the `bin` folder.
+
+#### ARM build
+
+ARM build may run on Raspberry Pi or on [energy-efficient ARM servers](https://blog.cloudflare.com/arm-takes-wing/).
+
+#### Development ARM build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make victoria-metrics-arm` or `make victoria-metrics-arm64` from the root folder of the repository.
+   It builds `victoria-metrics-arm` or `victoria-metrics-arm64` binary respectively and puts it into the `bin` folder.
+
+#### Production ARM build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make victoria-metrics-arm-prod` or `make victoria-metrics-arm64-prod` from the root folder of the repository.
+   It builds `victoria-metrics-arm-prod` or `victoria-metrics-arm64-prod` binary respectively and puts it into the `bin` folder.
+
+#### Pure Go build (CGO_ENABLED=0)
+
+`Pure Go` mode builds only Go code without [cgo](https://golang.org/cmd/cgo/) dependencies.
+This is an experimental mode, which may result in a lower compression ratio and slower decompression performance.
+Use it with caution!
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.12.
+2. Run `make victoria-metrics-pure` from the root folder of the repository.
+   It builds `victoria-metrics-pure` binary and puts it into the `bin` folder.
+
+#### Building docker images
+
+Run `make package-victoria-metrics`. It builds `victoriametrics/victoria-metrics:<PKG_TAG>` docker image locally.
+`<PKG_TAG>` is auto-generated image tag, which depends on source code in the repository.
+The `<PKG_TAG>` may be manually set via `PKG_TAG=foobar make package-victoria-metrics`.
+
+
+### Start with docker-compose
+
+[Docker-compose](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/deployment/docker/docker-compose.yml)
+helps to spin up VictoriaMetrics, Prometheus and Grafana with one command.
+More details may be found [here](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/master/deployment/docker#folder-contains-basic-images-and-tools-for-building-and-running-victoria-metrics-in-docker).
+
+
+### Setting up service
+
+Read [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/43) on how to set up VictoriaMetrics as a service in your OS.
+
+
+### Third-party contributions
+
+* [Unofficial yum repository](https://copr.fedorainfracloud.org/coprs/antonpatsev/VictoriaMetrics/) ([source code](https://github.com/patsevanton/victoriametrics-rpm))
+
+
+### How to work with snapshots?
+
+VictoriaMetrics can create [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
+for all the data stored under `-storageDataPath` directory.
+Navigate to `http://<victoriametrics-addr>:8428/snapshot/create` in order to create an instant snapshot.
+The page will return the following JSON response:
+
+```
+{"status":"ok","snapshot":"<snapshot-name>"}
+```
+
+Snapshots are created under `<-storageDataPath>/snapshots` directory, where `<-storageDataPath>`
+is the command-line flag value. Snapshots can be archived to backup storage at any time
+with [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md).
+
+The `http://<victoriametrics-addr>:8428/snapshot/list` page contains the list of available snapshots.
+
+Navigate to `http://<victoriametrics-addr>:8428/snapshot/delete?snapshot=<snapshot-name>` in order
+to delete `<snapshot-name>` snapshot.
+
+Navigate to `http://<victoriametrics-addr>:8428/snapshot/delete_all` in order to delete all the snapshots.
+
+Steps for restoring from a snapshot:
+1. Stop VictoriaMetrics with `kill -INT`.
+2. Restore snapshot contents from backup with [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md)
+   to the directory pointed by `-storageDataPath`.
+3. Start VictoriaMetrics.
+
+
+### How to delete time series?
+
+Send a request to `http://<victoriametrics-addr>:8428/api/v1/admin/tsdb/delete_series?match[]=<timeseries_selector_for_delete>`,
+where `<timeseries_selector_for_delete>` may contain any [time series selector](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors)
+for metrics to delete. After that all the time series matching the given selector are deleted. Storage space for
+the deleted time series isn't freed instantly - it is freed during subsequent merges of data files.
+
+It is recommended verifying which metrics will be deleted with the call to `http://<victoria-metrics-addr>:8428/api/v1/series?match[]=<timeseries_selector_for_delete>`
+before actually deleting the metrics.
+
+
+### How to export time series?
+
+Send a request to `http://<victoriametrics-addr>:8428/api/v1/export?match[]=<timeseries_selector_for_export>`,
+where `<timeseries_selector_for_export>` may contain any [time series selector](https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors)
+for metrics to export. The response would contain all the data for the selected time series in [JSON streaming format](https://en.wikipedia.org/wiki/JSON_streaming#Line-delimited_JSON).
+Each JSON line would contain data for a single time series. An example output:
+
+```
+{"metric":{"__name__":"up","job":"node_exporter","instance":"localhost:9100"},"values":[0,0,0],"timestamps":[1549891472010,1549891487724,1549891503438]}
+{"metric":{"__name__":"up","job":"prometheus","instance":"localhost:9090"},"values":[1,1,1],"timestamps":[1549891461511,1549891476511,1549891491511]}
+```
+
+Optional `start` and `end` args may be added to the request in order to limit the time frame for the exported data. These args may contain either
+unix timestamp in seconds or [RFC3339](https://www.ietf.org/rfc/rfc3339.txt) values.
+
+
+### Federation
+
+VictoriaMetrics exports [Prometheus-compatible federation data](https://prometheus.io/docs/prometheus/latest/federation/)
+at `http://<victoriametrics-addr>:8428/federate?match[]=<timeseries_selector_for_federation>`.
+
+Optional `start` and `end` args may be added to the request in order to scrape the last point for each selected time series on the `[start ... end]` interval.
+`start` and `end` may contain either unix timestamp in seconds or [RFC3339](https://www.ietf.org/rfc/rfc3339.txt) values. By default, the last point
+on the interval `[now - max_lookback ... now]` is scraped for each time series. The default value for `max_lookback` is `5m` (5 minutes), but it can be overridden.
+For instance, `/federate?match[]=up&max_lookback=1h` would return last points on the `[now - 1h ... now]` interval. This may be useful for time series federation
+with scrape intervals exceeding `5m`.
+
+
+### Capacity planning
+
+A rough estimation of the required resources for ingestion path:
+
+* RAM size: less than 1KB per active time series. So, ~1GB of RAM is required for 1M active time series.
+  Time series is considered active if new data points have been added to it recently or if it has been recently queried.
+  The number of active time series may be obtained from `vm_cache_entries{type="storage/hour_metric_ids"}` metric
+  exproted on the `/metrics` page.
+  VictoriaMetrics stores various caches in RAM. Memory size for these caches may be limited by `-memory.allowedPercent` flag.
+
+* CPU cores: a CPU core per 300K inserted data points per second. So, ~4 CPU cores are required for processing
+  the insert stream of 1M data points per second. The ingestion rate may be lower for high cardinality data or for time series with high number of labels.
+  See [this article](https://medium.com/@valyala/insert-benchmarks-with-inch-influxdb-vs-victoriametrics-e31a41ae2893) for details.
+  If you see lower numbers per CPU core, then it is likely active time series info doesn't fit caches,
+  so you need more RAM for lowering CPU usage.
+
+* Storage space: less than a byte per data point on average. So, ~260GB is required for storing a month-long insert stream
+  of 100K data points per second.
+  The actual storage size heavily depends on data randomness (entropy). Higher randomness means higher storage size requirements.
+  Read [this article](https://medium.com/faun/victoriametrics-achieving-better-compression-for-time-series-data-than-gorilla-317bc1f95932)
+  for details.
+
+* Network usage: outbound traffic is negligible. Ingress traffic is ~100 bytes per ingested data point via
+  [Prometheus remote_write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write).
+  The actual ingress bandwidth usage depends on the average number of labels per ingested metric and the average size
+  of label values. The higher number of per-metric labels and longer label values mean the higher ingress bandwidth.
+
+
+The required resources for query path:
+
+* RAM size: depends on the number of time series to scan in each query and the `step`
+  argument passed to [/api/v1/query_range](https://prometheus.io/docs/prometheus/latest/querying/api/#range-queries).
+  The higher number of scanned time series and lower `step` argument results in the higher RAM usage.
+
+* CPU cores: a CPU core per 30 millions of scanned data points per second.
+
+* Network usage: depends on the frequency and the type of incoming requests. Typical Grafana dashboards usually
+  require negligible network bandwidth.
+
+
+### High availability
+
+1) Install multiple VictoriaMetrics instances in distinct datacenters (availability zones).
+2) Add addresses of these instances to `remote_write` section in Prometheus config:
+
+```yml
+remote_write:
+  - url: http://<victoriametrics-addr-1>:8428/api/v1/write
+    queue_config:
+      max_samples_per_send: 10000
+  # ...
+  - url: http://<victoriametrics-addr-N>:8428/api/v1/write
+    queue_config:
+      max_samples_per_send: 10000
+```
+
+3) Apply the updated config:
+
+```
+kill -HUP `pidof prometheus`
+```
+
+4) Now Prometheus should write data into all the configured `remote_write` urls in parallel.
+5) Set up [Promxy](https://github.com/jacksontj/promxy) in front of all the VictoriaMetrics replicas.
+6) Set up Prometheus datasource in Grafana that points to Promxy.
+
+
+If you have Prometheus HA pairs with replicas `r1` and `r2` in each pair, then configure each `r1`
+to write data to `victoriametrics-addr-1`, while each `r2` should write data to `victoriametrics-addr-2`.
+
+
+### Multiple retentions
+
+Just start multiple VictoriaMetrics instances with distinct values for the following flags:
+
+* `-retentionPeriod`
+* `-storageDataPath`, so the data for each retention period is saved in a separate directory
+* `-httpListenAddr`, so clients may reach VictoriaMetrics instance with proper retention
+
+
+### Downsampling
+
+There is no downsampling support at the moment, but:
+- VictoriaMetrics is optimized for querying big amounts of raw data. See benchmark results for heavy queries
+  in [this article](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+- VictoriaMetrics has good compression for on-disk data. See [this article](https://medium.com/@valyala/victoriametrics-achieving-better-compression-for-time-series-data-than-gorilla-317bc1f95932)
+  for details.
+
+These properties reduce the need in downsampling. We plan to implement downsampling in the future.
+See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/36) for details.
+
+
+### Multi-tenancy
+
+Single-node VictoriaMetrics doesn't support multi-tenancy. Use [cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster) instead.
+
+
+### Scalability and cluster version
+
+Though single-node VictoriaMetrics cannot scale to multiple nodes, it is optimized for resource usage - storage size / bandwidth / IOPS, RAM, CPU.
+This means that a single-node VictoriaMetrics may scale vertically and substitute a moderately sized cluster built with competing solutions
+such as Thanos, Uber M3, InfluxDB or TimescaleDB. See [vertical scalability benchmarks](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae).
+
+So try single-node VictoriaMetrics at first and then [switch to cluster version](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster) if you still need
+horizontally scalable long-term remote storage for really large Prometheus deployments.
+[Contact us](mailto:info@victoriametrics.com) for paid support.
+
+
+### Alerting
+
+VictoriaMetrics doesn't support rule evaluation and alerting yet, so these actions must be performed either
+on [Prometheus side](https://prometheus.io/docs/alerting/overview/) or on [Grafana side](https://grafana.com/docs/alerting/rules/).
+
+
+### Security
+
+Do not forget protecting sensitive endpoints in VictoriaMetrics when exposing it to untrusted networks such as the internet.
+Consider setting the following command-line flags:
+
+* `-tls`, `-tlsCertFile` and `-tlsKeyFile` for switching from HTTP to HTTPS.
+* `-httpAuth.username` and `-httpAuth.password` for protecting all the HTTP endpoints
+  with [HTTP Basic Authentication](https://en.wikipedia.org/wiki/Basic_access_authentication).
+* `-deleteAuthKey` for protecting `/api/v1/admin/tsdb/delete_series` endpoint. See [how to delete time series](#how-to-delete-time-series).
+* `-snapshotAuthKey` for protecting `/snapshot*` endpoints. See [how to work with snapshots](#how-to-work-with-snapshots).
+
+Explicitly set internal network interface for TCP and UDP ports for data ingestion with Graphite and OpenTSDB formats.
+For example, substitute `-graphiteListenAddr=:2003` with `-graphiteListenAddr=<internal_iface_ip>:2003`.
+
+
+### Tuning
+
+* There is no need in VictoriaMetrics tuning since it uses reasonable defaults for command-line flags,
+  which are automatically adjusted for the available CPU and RAM resources.
+* There is no need in Operating System tuning since VictoriaMetrics is optimized for default OS settings.
+  The only option is increasing the limit on [the number of open files in the OS](https://medium.com/@muhammadtriwibowo/set-permanently-ulimit-n-open-files-in-ubuntu-4d61064429a),
+  so Prometheus instances could establish more connections to VictoriaMetrics.
+* The recommended filesystem is `ext4`, the recommended persistent storage is [persistent HDD-based disk on GCP](https://cloud.google.com/compute/docs/disks/#pdspecs),
+  since it is protected from hardware failures via internal replication and it can be [resized on the fly](https://cloud.google.com/compute/docs/disks/add-persistent-disk#resize_pd).
+  If you plan storing more than 1TB of data on `ext4` partition or plan extending it to more than 16TB,
+  then the following options are recommended to pass to `mkfs.ext4`:
+
+```
+mkfs.ext4 ... -O 64bit,huge_file,extent -T huge
+```
+
+
+### Monitoring
+
+VictoriaMetrics exports internal metrics in Prometheus format on the `/metrics` page.
+Add this page to Prometheus' scrape config in order to collect VictoriaMetrics metrics.
+There is [an official Grafana dashboard for single-node VictoriaMetrics](https://grafana.com/dashboards/10229).
+
+The most interesting metrics are:
+
+* `vm_cache_entries{type="storage/hour_metric_ids"}` - the number of time series with new data points during the last hour
+  aka active time series.
+* `rate(vm_new_timeseries_created_total[5m])` - time series churn rate.
+* `vm_rows{type="indexdb"}` - the number of rows in inverted index. High value for this number usually mean high churn rate for time series.
+* Sum of `vm_rows{type="storage/big"}` and `vm_rows{type="storage/small"}` - total number of `(timestamp, value)` data points
+  in the database.
+* Sum of all the `vm_cache_size_bytes` metrics - the total size of all the caches in the database.
+* `vm_allowed_memory_bytes` - the maximum allowed size for caches in the database. It is calculated as `system_memory * <-memory.allowedPercent> / 100`,
+  where `system_memory` is the amount of system memory and `-memory.allowedPercent` is the corresponding flag value.
+* `vm_rows_inserted_total` - the total number of inserted rows since VictoriaMetrics start.
+
+
+### Troubleshooting
+
+* It is recommended to use default command-line flag values (i.e. don't set them explicitly) until the need
+  in tweaking these flag values arises.
+
+* If VictoriaMetrics works slowly and eats more than a CPU core per 100K ingested data points per second,
+  then it is likely you have too many active time series for the current amount of RAM.
+  It is recommended increasing the amount of RAM on the node with VictoriaMetrics in order to improve
+  ingestion performance.
+  Another option is to increase `-memory.allowedPercent` command-line flag value. Be careful with this
+  option, since too big value for `-memory.allowedPercent` may result in high I/O usage.
+
+* VictoriaMetrics requires free disk space for [merging data files to bigger ones](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282).
+  It may slow down when there is no enough free space left. So make sure `-storageDataPath` directory
+  has at least 20% of free space comparing to disk size.
+
+* If VictoriaMetrics doesn't work because of certain parts are corrupted due to disk errors,
+  then just remove directoreis with broken parts. This will recover VictoriaMetrics at the cost
+  of data loss stored in the broken parts. In the future, `vmrecover` tool will be created
+  for automatic recovering from such errors.
+
+
+### Backfilling
+
+Make sure that configured `-retentionPeriod` covers timestamps for the backfilled data.
+
+It is recommended disabling query cache with `-search.disableCache` command-line flag when writing
+historical data with timestamps from the past, since the cache assumes that the data is written with
+the current timestamps. Query cache can be enabled after the backfilling is complete.
+
+
+### Profiling
+
+VictoriaMetrics provides handlers for collecting the following [Go profiles](https://blog.golang.org/profiling-go-programs):
+
+- Memory profile. It can be collected with the following command:
+```
+curl -s http://<victoria-metrics-host>:8428/debug/pprof/heap > mem.pprof
+```
+
+- CPU profile. It can be collected with the following command:
+```
+curl -s http://<victoria-metrics-host>:8428/debug/pprof/profile > cpu.pprof
+```
+
+The command for collecting CPU profile waits for 30 seconds before returning.
+
+The collected profiles may be analyzed with [go tool pprof](https://github.com/google/pprof).
+
+
+## Integrations
+
+* [netdata](https://github.com/netdata/netdata) can push data into VictoriaMetrics via `Prometheus remote_write API`.
+  See [these docs](https://github.com/netdata/netdata#integrations).
+* [go-graphite/carbonapi](https://github.com/go-graphite/carbonapi) can use VictoriaMetrics as time series backend.
+  See [this example](/blob/master/cmd/carbonapi/carbonapi.example.prometheus.yaml).
+* [Ansible role for installing VictoriaMetrics](https://github.com/dreamteam-gg/ansible-victoriametrics-role).
+
+
+## Roadmap
+
+- [ ] Replication [#118](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/118)
+- [ ] Support of Object Storages (GCS, S3, Azure Storage) [#38](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/38)
+- [ ] Data downsampling [#36](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/36)
+- [ ] Alert Manager Integration [#119](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/119) 
+- [ ] CLI tool for data migration, re-balancing and adding/removing nodes [#103](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/103)
+
+
+The discussion happens [here](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/129). Feel free to comment any item or add own one.
+
+
+## Contacts
+
+Contact us with any questions regarding VictoriaMetrics at [info@victoriametrics.com](mailto:info@victoriametrics.com).
+
+
+## Community and contributions
+
+Feel free asking any questions regarding VictoriaMetrics:
+
+- [slack](http://slack.victoriametrics.com/)
+- [telegram-en](https://t.me/VictoriaMetrics_en)
+- [telegram-ru](https://t.me/VictoriaMetrics_ru1)
+- [google groups](https://groups.google.com/forum/#!forum/victorametrics-users)
+
+
+If you like VictoriaMetrics and want to contribute, then we need the following:
+
+- Filing issues and feature requests [here](https://github.com/VictoriaMetrics/VictoriaMetrics/issues).
+- Spreading a word about VictoriaMetrics: conference talks, articles, comments, experience sharing with colleagues.
+- Updating documentation.
+
+We are open to third-party pull requests provided they follow [KISS design principle](https://en.wikipedia.org/wiki/KISS_principle):
+
+- Prefer simple code and architecture.
+- Avoid complex abstractions.
+- Avoid magic code and fancy algorithms.
+- Avoid [big external dependencies](https://medium.com/@valyala/stripping-dependency-bloat-in-victoriametrics-docker-image-983fb5912b0d).
+- Minimize the number of moving parts in the distributed system.
+- Avoid automated decisions, which may hurt cluster availability, consistency or performance.
+
+Adhering `KISS` principle simplifies the resulting code and architecture, so it can be reviewed, understood and verified by many people.
+
+
+## Reporting bugs
+
+Report bugs and propose new features [here](https://github.com/VictoriaMetrics/VictoriaMetrics/issues).
+
+
+## Victoria Metrics Logo
+
+[Zip](VM_logo.zip) contains three folders with different image orientation (main color and inverted version).
+
+Files included in each folder:
+
+* 2 JPEG Preview files
+* 2 PNG Preview files with transparent background
+* 2 EPS Adobe Illustrator EPS10 files
+
+
+### Logo Usage Guidelines
+
+#### Font used:
+
+* Lato Black
+* Lato Regular
+
+#### Color Palette:
+
+* HEX [#110f0f](https://www.color-hex.com/color/110f0f)
+* HEX [#ffffff](https://www.color-hex.com/color/ffffff)
+
+### We kindly ask:
+
+- Please don't use any other font instead of suggested.
+- There should be sufficient clear space around the logo.
+- Do not change spacing, alignment, or relative locations of the design elements.
+- Do not change the proportions of any of the design elements or the design itself. You    may resize as needed but must retain all proportions.

go.mod

@@ -1,18 +1,32 @@
 module github.com/VictoriaMetrics/VictoriaMetrics
 
 require (
-	github.com/VictoriaMetrics/fastcache v1.5.1
-	github.com/VictoriaMetrics/metrics v1.7.1
-	github.com/cespare/xxhash/v2 v2.0.1-0.20190104013014-3767db7a7e18
+	cloud.google.com/go v0.49.0 // indirect
+	cloud.google.com/go/storage v1.4.0
+	github.com/VictoriaMetrics/fastcache v1.5.4
+	github.com/VictoriaMetrics/metrics v1.9.2
+	github.com/aws/aws-sdk-go v1.25.43
+	github.com/cespare/xxhash/v2 v2.1.1
+	github.com/golang/groupcache v0.0.0-20191027212112-611e8accdfc9 // indirect
 	github.com/golang/snappy v0.0.1
-	github.com/google/go-cmp v0.3.0 // indirect
-	github.com/klauspost/compress v1.7.6
-	github.com/spaolacci/murmur3 v1.1.0 // indirect
+	github.com/jstemmer/go-junit-report v0.9.1 // indirect
+	github.com/klauspost/compress v1.9.2
 	github.com/valyala/fastjson v1.4.1
-	github.com/valyala/gozstd v1.6.1
+	github.com/valyala/fastrand v1.0.0
+	github.com/valyala/gozstd v1.6.3
 	github.com/valyala/histogram v1.0.1
-	github.com/valyala/quicktemplate v1.2.0
-	golang.org/x/sys v0.0.0-20190813064441-fde4db37ae7a
+	github.com/valyala/quicktemplate v1.4.1
+	go.opencensus.io v0.22.2 // indirect
+	golang.org/x/exp v0.0.0-20191127035308-9964a5a80460 // indirect
+	golang.org/x/lint v0.0.0-20191125180803-fdd1cda4f05f // indirect
+	golang.org/x/net v0.0.0-20191126235420-ef20fe5d7933 // indirect
+	golang.org/x/oauth2 v0.0.0-20191122200657-5d9234df094c // indirect
+	golang.org/x/sys v0.0.0-20191127021746-63cb32ae39b2
+	golang.org/x/tools v0.0.0-20191127064951-724660f1afeb // indirect
+	google.golang.org/api v0.14.0
+	google.golang.org/appengine v1.6.5 // indirect
+	google.golang.org/genproto v0.0.0-20191115221424-83cc0476cb11 // indirect
+	google.golang.org/grpc v1.25.1 // indirect
 )
 
 go 1.12

go.sum

@@ -1,39 +1,99 @@
-github.com/OneOfOne/xxhash v1.2.2/go.mod h1:HSdplMjZKSmBqAxg5vPj2TmRDmfkzw+cTzAElWljhcU=
-github.com/OneOfOne/xxhash v1.2.5 h1:zl/OfRA6nftbBK9qTohYBJ5xvw6C/oNKizR7cZGl3cI=
-github.com/OneOfOne/xxhash v1.2.5/go.mod h1:eZbhyaAYD41SGSSsnmcpxVoRiQ/MPUTjUdIIOT9Um7Q=
-github.com/VictoriaMetrics/fastcache v1.5.1 h1:qHgHjyoNFV7jgucU8QZUuU4gcdhfs8QW1kw68OD2Lag=
-github.com/VictoriaMetrics/fastcache v1.5.1/go.mod h1:+jv9Ckb+za/P1ZRg/sulP5Ni1v49daAVERr0H3CuscE=
-github.com/VictoriaMetrics/metrics v1.7.1 h1:g2qrY6Upn8rvlvR40cGHFY0crwi4hpqF0n9vJMNsCSg=
-github.com/VictoriaMetrics/metrics v1.7.1/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
+cloud.google.com/go v0.26.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
+cloud.google.com/go v0.34.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
+cloud.google.com/go v0.38.0/go.mod h1:990N+gfupTy94rShfmMCWGDn0LpTmnzTp2qbd1dvSRU=
+cloud.google.com/go v0.44.1/go.mod h1:iSa0KzasP4Uvy3f1mN/7PiObzGgflwredwwASm/v6AU=
+cloud.google.com/go v0.44.2/go.mod h1:60680Gw3Yr4ikxnPRS/oxxkBccT6SA1yMk63TGekxKY=
+cloud.google.com/go v0.45.1/go.mod h1:RpBamKRgapWJb87xiFSdk4g1CME7QZg3uwTez+TSTjc=
+cloud.google.com/go v0.46.3/go.mod h1:a6bKKbmY7er1mI7TEI4lsAkts/mkhTSZK8w33B4RAg0=
+cloud.google.com/go v0.49.0 h1:CH+lkubJzcPYB1Ggupcq0+k8Ni2ILdG2lYjDIgavDBQ=
+cloud.google.com/go v0.49.0/go.mod h1:hGvAdzcWNbyuxS3nWhD7H2cIJxjRRTRLQVB0bdputVY=
+cloud.google.com/go/bigquery v1.0.1 h1:hL+ycaJpVE9M7nLoiXb/Pn10ENE2u+oddxbD8uu0ZVU=
+cloud.google.com/go/bigquery v1.0.1/go.mod h1:i/xbL2UlR5RvWAURpBYZTtm/cXjCha9lbfbpx4poX+o=
+cloud.google.com/go/datastore v1.0.0 h1:Kt+gOPPp2LEPWp8CSfxhsM8ik9CcyE/gYu+0r+RnZvM=
+cloud.google.com/go/datastore v1.0.0/go.mod h1:LXYbyblFSglQ5pkeyhO+Qmw7ukd3C+pD7TKLgZqpHYE=
+cloud.google.com/go/pubsub v1.0.1 h1:W9tAK3E57P75u0XLLR82LZyw8VpAnhmyTOxW9qzmyj8=
+cloud.google.com/go/pubsub v1.0.1/go.mod h1:R0Gpsv3s54REJCy4fxDixWD93lHJMoZTyQ2kNxGRt3I=
+cloud.google.com/go/storage v1.0.0/go.mod h1:IhtSnM/ZTZV8YYJWCY8RULGVqBDmpoyjwiyrjsg+URw=
+cloud.google.com/go/storage v1.4.0 h1:KDdqY5VTXBTqpSbctVTt0mVvfanP6JZzNzLE0qNY100=
+cloud.google.com/go/storage v1.4.0/go.mod h1:ZusYJWlOshgSBGbt6K3GnB3MT3H1xs2id9+TCl4fDBA=
+dmitri.shuralyov.com/gpu/mtl v0.0.0-20190408044501-666a987793e9/go.mod h1:H6x//7gZCb22OMCxBHrMx7a5I7Hp++hsVxbQ4BYO7hU=
+github.com/BurntSushi/toml v0.3.1 h1:WXkYYl6Yr3qBf1K79EBnL4mak0OimBfB0XUf9Vl28OQ=
+github.com/BurntSushi/toml v0.3.1/go.mod h1:xHWCNGjB5oqiDr8zfno3MHue2Ht5sIBksp03qcyfWMU=
+github.com/BurntSushi/xgb v0.0.0-20160522181843-27f122750802/go.mod h1:IVnqGOEym/WlBOVXweHU+Q+/VP0lqqI8lqeDx9IjBqo=
+github.com/VictoriaMetrics/fastcache v1.5.4 h1:0BaXbRH01RycJk79OOBwMCXlNryko9z4yEf6RqbP+Xo=
+github.com/VictoriaMetrics/fastcache v1.5.4/go.mod h1:ptDBkNMQI4RtmVo8VS/XwRY6RoTu1dAWCbrk+6WsEM8=
+github.com/VictoriaMetrics/metrics v1.9.2 h1:+CNV5OOPe1PuHff3AsOa5cbN8qCWkWwZZA7eOmFjOwQ=
+github.com/VictoriaMetrics/metrics v1.9.2/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
 github.com/allegro/bigcache v1.2.1-0.20190218064605-e24eb225f156 h1:eMwmnE/GDgah4HI848JfFxHt+iPb26b4zyfspmqY0/8=
 github.com/allegro/bigcache v1.2.1-0.20190218064605-e24eb225f156/go.mod h1:Cb/ax3seSYIx7SuZdm2G2xzfwmv3TPSk2ucNfQESPXM=
-github.com/cespare/xxhash v1.1.0 h1:a6HrQnmkObjyL+Gs60czilIUGqrzKutQD6XZog3p+ko=
-github.com/cespare/xxhash v1.1.0/go.mod h1:XrSqR1VqqWfGrhpAt58auRo0WTKS1nRRg3ghfAqPWnc=
-github.com/cespare/xxhash/v2 v2.0.1-0.20190104013014-3767db7a7e18 h1:pl4eWIqvFe/Kg3zkn7NxevNzILnZYWDCG7qbA1CJik0=
-github.com/cespare/xxhash/v2 v2.0.1-0.20190104013014-3767db7a7e18/go.mod h1:HD5P3vAIAh+Y2GAxg0PrPN1P8WkepXGpjbUPDHJqqKM=
+github.com/aws/aws-sdk-go v1.25.43 h1:R5YqHQFIulYVfgRySz9hvBRTWBjudISa+r0C8XQ1ufg=
+github.com/aws/aws-sdk-go v1.25.43/go.mod h1:KmX6BPdI08NWTb3/sm4ZGu5ShLoqVDhKgpiN924inxo=
+github.com/census-instrumentation/opencensus-proto v0.2.1/go.mod h1:f6KPmirojxKA12rnyqOA5BBL4O983OfeGPqjHWSTneU=
+github.com/cespare/xxhash/v2 v2.1.1 h1:6MnRN8NT7+YBpUIWxHtefFZOKTAPgGjpQSxqLNn0+qY=
+github.com/cespare/xxhash/v2 v2.1.1/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
+github.com/client9/misspell v0.3.4/go.mod h1:qj6jICC3Q7zFZvVWo7KLAzC3yx5G7kyvSDkc90ppPyw=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/envoyproxy/go-control-plane v0.9.0/go.mod h1:YTl/9mNaCwkRvm6d1a2C3ymFceY/DCBVvsKhRF0iEA4=
+github.com/envoyproxy/protoc-gen-validate v0.1.0/go.mod h1:iSmxcyjqTsJpI2R4NaDN7+kN2VEUnK/pcBlmesArF7c=
+github.com/go-gl/glfw v0.0.0-20190409004039-e6da0acd62b1/go.mod h1:vR7hzQXu2zJy9AVAgeJqvqgH9Q5CA+iKCZ2gyEVpxRU=
+github.com/go-gl/glfw/v3.3/glfw v0.0.0-20191125211704-12ad95a8df72/go.mod h1:tQ2UAYgL5IevRw8kRxooKSPJfGvJ9fJQFa0TUsXzTg8=
+github.com/golang/glog v0.0.0-20160126235308-23def4e6c14b h1:VKtxabqXZkF25pY9ekfRL6a582T4P37/31XEstQ5p58=
+github.com/golang/glog v0.0.0-20160126235308-23def4e6c14b/go.mod h1:SBH7ygxi8pfUlaOkMMuAQtPIUF8ecWP5IEl/CR7VP2Q=
+github.com/golang/groupcache v0.0.0-20190702054246-869f871628b6/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc=
+github.com/golang/groupcache v0.0.0-20191027212112-611e8accdfc9 h1:uHTyIjqVhYRhLbJ8nIiOJHkEZZ+5YoOsAbD3sk82NiE=
+github.com/golang/groupcache v0.0.0-20191027212112-611e8accdfc9/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc=
+github.com/golang/mock v1.1.1/go.mod h1:oTYuIxOrZwtPieC+H1uAHpcLFnEyAGVDL/k47Jfbm0A=
+github.com/golang/mock v1.2.0/go.mod h1:oTYuIxOrZwtPieC+H1uAHpcLFnEyAGVDL/k47Jfbm0A=
+github.com/golang/mock v1.3.1/go.mod h1:sBzyDLLjw3U8JLTeZvSv8jJB+tU5PVekmnlKIyFUx0Y=
+github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/golang/protobuf v1.3.2 h1:6nsPYzhq5kReh6QImI3k5qWzO4PEbvbIW2cwSfR/6xs=
+github.com/golang/protobuf v1.3.2/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
 github.com/golang/snappy v0.0.1 h1:Qgr9rKW7uDUkrbSmQeiDsGa8SjGyCOGtuasMWwvp2P4=
 github.com/golang/snappy v0.0.1/go.mod h1:/XxbfmMg8lxefKM7IXC3fBNl/7bRcc72aCRzEWrmP2Q=
+github.com/google/btree v0.0.0-20180813153112-4030bb1f1f0c/go.mod h1:lNA+9X1NB3Zf8V7Ke586lFgjr2dZNuvo3lPJSGZ5JPQ=
+github.com/google/btree v1.0.0/go.mod h1:lNA+9X1NB3Zf8V7Ke586lFgjr2dZNuvo3lPJSGZ5JPQ=
+github.com/google/go-cmp v0.2.0/go.mod h1:oXzfMopK8JAjlY9xF4vHSVASa0yLyX7SntLO5aqRK0M=
 github.com/google/go-cmp v0.3.0 h1:crn/baboCvb5fXaQ0IJ1SGTsTVrWpDsCWC8EGETZijY=
 github.com/google/go-cmp v0.3.0/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU=
+github.com/google/martian v2.1.0+incompatible h1:/CP5g8u/VJHijgedC/Legn3BAbAaWPgecwXBIDzw5no=
+github.com/google/martian v2.1.0+incompatible/go.mod h1:9I4somxYTbIHy5NJKHRl3wXiIaQGbYVAs8BPL6v8lEs=
+github.com/google/pprof v0.0.0-20181206194817-3ea8567a2e57/go.mod h1:zfwlbNMJ+OItoe0UupaVj+oy1omPYYDuagoSzA8v9mc=
+github.com/google/pprof v0.0.0-20190515194954-54271f7e092f/go.mod h1:zfwlbNMJ+OItoe0UupaVj+oy1omPYYDuagoSzA8v9mc=
+github.com/google/renameio v0.1.0/go.mod h1:KWCgfxg9yswjAJkECMjeO8J8rahYeXnNhOm40UhjYkI=
+github.com/googleapis/gax-go/v2 v2.0.4/go.mod h1:0Wqv26UfaUD9n4G6kQubkQ+KchISgw+vpHVxEJEs9eg=
+github.com/googleapis/gax-go/v2 v2.0.5 h1:sjZBwGj9Jlw33ImPtvFviGYvseOtDM7hkSKB7+Tv3SM=
+github.com/googleapis/gax-go/v2 v2.0.5/go.mod h1:DWXyrwAJ9X0FpwwEdw+IPEYBICEFu5mhpdKc/us6bOk=
+github.com/hashicorp/golang-lru v0.5.0/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
+github.com/hashicorp/golang-lru v0.5.1/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
+github.com/jmespath/go-jmespath v0.0.0-20180206201540-c2b33e8439af h1:pmfjZENx5imkbgOkpRUYLnmbU7UEFbjtDA2hxJ1ichM=
+github.com/jmespath/go-jmespath v0.0.0-20180206201540-c2b33e8439af/go.mod h1:Nht3zPeWKUH0NzdCt2Blrr5ys8VGpn0CEB0cQHVjt7k=
+github.com/jstemmer/go-junit-report v0.0.0-20190106144839-af01ea7f8024/go.mod h1:6v2b51hI/fHJwM22ozAgKL4VKDeJcHhJFhtBdhmNjmU=
+github.com/jstemmer/go-junit-report v0.9.1 h1:6QPYqodiu3GuPL+7mfx+NwDdp2eTkp9IfEUpgAwUN0o=
+github.com/jstemmer/go-junit-report v0.9.1/go.mod h1:Brl9GWCQeLvo8nXZwPNNblvFj/XSXhF0NWZEnDohbsk=
+github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
 github.com/klauspost/compress v1.4.0/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A=
 github.com/klauspost/compress v1.4.1/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A=
-github.com/klauspost/compress v1.7.6 h1:GH2karLOcuZtA5a3+KuzSU33A2cvcHGbtEWM6K4t7oU=
-github.com/klauspost/compress v1.7.6/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A=
+github.com/klauspost/compress v1.9.2 h1:LfVyl+ZlLlLDeQ/d2AqfGIIH4qEDu0Ed2S5GyhCWIWY=
+github.com/klauspost/compress v1.9.2/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A=
 github.com/klauspost/cpuid v0.0.0-20180405133222-e7e905edc00e/go.mod h1:Pj4uuM528wm8OyEC2QMXAi2YiTZ96dNQPGgoMS4s3ek=
-github.com/klauspost/cpuid v1.2.0 h1:NMpwD2G9JSFOE1/TJjGSo5zG7Yb2bTe7eq1jH+irmeE=
 github.com/klauspost/cpuid v1.2.0/go.mod h1:Pj4uuM528wm8OyEC2QMXAi2YiTZ96dNQPGgoMS4s3ek=
+github.com/kr/pretty v0.1.0 h1:L/CwN0zerZDmRFUapSPitk6f+Q3+0za1rQkzVuMiMFI=
+github.com/kr/pretty v0.1.0/go.mod h1:dAy3ld7l9f0ibDNOQOHHMYYIIbhfbHSm3C4ZsoJORNo=
+github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
+github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
+github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/spaolacci/murmur3 v0.0.0-20180118202830-f09979ecbc72/go.mod h1:JwIasOWyU6f++ZhiEuf87xNszmSA2myDM2Kzu9HwQUA=
-github.com/spaolacci/murmur3 v1.0.1-0.20190317074736-539464a789e9/go.mod h1:JwIasOWyU6f++ZhiEuf87xNszmSA2myDM2Kzu9HwQUA=
-github.com/spaolacci/murmur3 v1.1.0 h1:7c1g84S4BPRrfL5Xrdp6fOJ206sU9y293DDHaoy0bLI=
-github.com/spaolacci/murmur3 v1.1.0/go.mod h1:JwIasOWyU6f++ZhiEuf87xNszmSA2myDM2Kzu9HwQUA=
+github.com/prometheus/client_model v0.0.0-20190812154241-14fe0d1b01d4/go.mod h1:xMI15A0UPsDsEKsMN9yxemIoYk6Tm2C1GtYGdfGttqA=
+github.com/rogpeppe/go-internal v1.3.0/go.mod h1:M8bDsm7K2OlrFYOpmOWEs/qY81heoFRclV5y23lUDJ4=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.3.0 h1:TivCn/peBQ7UY8ooIcPgZFpTNSz0Q2U6UrFlUfqbe0Q=
 github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI=
+github.com/stretchr/testify v1.4.0 h1:2E4SXV/wtOkTonXsotYi4li6zVWxYlZuYNCXe9XRJyk=
+github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
 github.com/valyala/bytebufferpool v1.0.0 h1:GqA5TC/0021Y/b9FG4Oi9Mr3q7XYx6KllzawFIhcdPw=
 github.com/valyala/bytebufferpool v1.0.0/go.mod h1:6bBcMArwyJ5K/AmCkWv1jt77kVWyCJ6HpOuEn7z0Csc=
 github.com/valyala/fasthttp v1.2.0/go.mod h1:4vX61m6KN+xDduDNwXrhIAVZaZaZiQ1luJk8LWSxF3s=
@@ -41,13 +101,144 @@ github.com/valyala/fastjson v1.4.1 h1:hrltpHpIpkaxll8QltMU8c3QZ5+qIiCL8yKqPFJI/y
 github.com/valyala/fastjson v1.4.1/go.mod h1:nV6MsjxL2IMJQUoHDIrjEI7oLyeqK6aBD7EFWPsvP8o=
 github.com/valyala/fastrand v1.0.0 h1:LUKT9aKer2dVQNUi3waewTbKV+7H17kvWFNKs2ObdkI=
 github.com/valyala/fastrand v1.0.0/go.mod h1:HWqCzkrkg6QXT8V2EXWvXCoow7vLwOFN002oeRzjapQ=
-github.com/valyala/gozstd v1.6.1 h1:oFN2mNW0kOr1fEKJuLpDwakNb6Y9fElVEBZmPEsFTUw=
-github.com/valyala/gozstd v1.6.1/go.mod h1:y5Ew47GLlP37EkTB+B4s7r6A5rdaeB7ftbl9zoYiIPQ=
+github.com/valyala/gozstd v1.6.3 h1:kr3oF/F1RvxYr8wgPjrH04gvHuMEL99pPC9e+5pPQdU=
+github.com/valyala/gozstd v1.6.3/go.mod h1:y5Ew47GLlP37EkTB+B4s7r6A5rdaeB7ftbl9zoYiIPQ=
 github.com/valyala/histogram v1.0.1 h1:FzA7n2Tz/wKRMejgu3PV1vw3htAklTjjuoI6z3d4KDg=
 github.com/valyala/histogram v1.0.1/go.mod h1:lQy0xA4wUz2+IUnf97SivorsJIp8FxsnRd6x25q7Mto=
-github.com/valyala/quicktemplate v1.2.0 h1:BaO1nHTkspYzmAjPXj0QiDJxai96tlcZyKcI9dyEGvM=
-github.com/valyala/quicktemplate v1.2.0/go.mod h1:EH+4AkTd43SvgIbQHYu59/cJyxDoOVRUAfrukLPuGJ4=
+github.com/valyala/quicktemplate v1.4.1 h1:tEtkSN6mTCJlYVT7As5x4wjtkk2hj2thsb0M+AcAVeM=
+github.com/valyala/quicktemplate v1.4.1/go.mod h1:EH+4AkTd43SvgIbQHYu59/cJyxDoOVRUAfrukLPuGJ4=
 github.com/valyala/tcplisten v0.0.0-20161114210144-ceec8f93295a/go.mod h1:v3UYOV9WzVtRmSR+PDvWpU/qWl4Wa5LApYYX4ZtKbio=
+go.opencensus.io v0.21.0/go.mod h1:mSImk1erAIZhrmZN+AvHh14ztQfjbGwt4TtuofqLduU=
+go.opencensus.io v0.22.0/go.mod h1:+kGneAE2xo2IficOXnaByMWTGM9T73dGwxeWcUqIpI8=
+go.opencensus.io v0.22.2 h1:75k/FF0Q2YM8QYo07VPddOLBslDt1MZOdEslOHvmzAs=
+go.opencensus.io v0.22.2/go.mod h1:yxeiOL68Rb0Xd1ddK5vPZ/oVn4vY4Ynel7k9FzqtOIw=
+golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
+golang.org/x/crypto v0.0.0-20190510104115-cbcb75029529/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
+golang.org/x/crypto v0.0.0-20190605123033-f99c8df09eb5/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
+golang.org/x/exp v0.0.0-20190121172915-509febef88a4/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
+golang.org/x/exp v0.0.0-20190306152737-a1d7652674e8/go.mod h1:CJ0aWSM057203Lf6IL+f9T1iT9GByDxfZKAQTCR3kQA=
+golang.org/x/exp v0.0.0-20190510132918-efd6b22b2522/go.mod h1:ZjyILWgesfNpC6sMxTJOJm9Kp84zZh5NQWvqDGG3Qr8=
+golang.org/x/exp v0.0.0-20190829153037-c13cbed26979/go.mod h1:86+5VVa7VpoJ4kLfm080zCjGlMRFzhUhsZKEZO7MGek=
+golang.org/x/exp v0.0.0-20191030013958-a1ab85dbe136 h1:A1gGSx58LAGVHUUsOf7IiR0u8Xb6W51gRwfDBhkdcaw=
+golang.org/x/exp v0.0.0-20191030013958-a1ab85dbe136/go.mod h1:JXzH8nQsPlswgeRAPE3MuO9GYsAcnJvJ4vnMwN/5qkY=
+golang.org/x/exp v0.0.0-20191127035308-9964a5a80460 h1:zNL062UG4d0GC48Bhm+lEI9lTOMsEHNL0WITb/cw7/s=
+golang.org/x/exp v0.0.0-20191127035308-9964a5a80460/go.mod h1:2RIsYlXP63K8oxa1u096TMicItID8zy7Y6sNkU49FU4=
+golang.org/x/image v0.0.0-20190227222117-0694c2d4d067/go.mod h1:kZ7UVZpmo3dzQBMxlp+ypCbDeSB+sBbTgSJuh5dn5js=
+golang.org/x/image v0.0.0-20190802002840-cff245a6509b/go.mod h1:FeLwcggjj3mMvU+oOTbSwawSJRM1uh48EjtB4UJZlP0=
+golang.org/x/lint v0.0.0-20181026193005-c67002cb31c3/go.mod h1:UVdnD1Gm6xHRNCYTkRU2/jEulfH38KcIWyp/GAMgvoE=
+golang.org/x/lint v0.0.0-20190227174305-5b3e6a55c961/go.mod h1:wehouNa3lNwaWXcvxsM5YxQ5yQlVC4a0KAMCusXpPoU=
+golang.org/x/lint v0.0.0-20190301231843-5614ed5bae6f/go.mod h1:UVdnD1Gm6xHRNCYTkRU2/jEulfH38KcIWyp/GAMgvoE=
+golang.org/x/lint v0.0.0-20190313153728-d0100b6bd8b3/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
+golang.org/x/lint v0.0.0-20190409202823-959b441ac422/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
+golang.org/x/lint v0.0.0-20190909230951-414d861bb4ac/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
+golang.org/x/lint v0.0.0-20190930215403-16217165b5de h1:5hukYrvBGR8/eNkX5mdUezrA6JiaEZDtJb9Ei+1LlBs=
+golang.org/x/lint v0.0.0-20190930215403-16217165b5de/go.mod h1:6SW0HCj/g11FgYtHlgUYUwCkIfeOF89ocIRzGO/8vkc=
+golang.org/x/lint v0.0.0-20191125180803-fdd1cda4f05f h1:J5lckAjkw6qYlOZNj90mLYNTEKDvWeuc1yieZ8qUzUE=
+golang.org/x/lint v0.0.0-20191125180803-fdd1cda4f05f/go.mod h1:5qLYkcX4OjUUV8bRuDixDT3tpyyb+LUpUlRWLxfhWrs=
+golang.org/x/mobile v0.0.0-20190312151609-d3739f865fa6/go.mod h1:z+o9i4GpDbdi3rU15maQ/Ox0txvL9dWGYEHz965HBQE=
+golang.org/x/mobile v0.0.0-20190719004257-d2bd2a29d028/go.mod h1:E/iHnbuqvinMTCcRqshq8CkpyQDoeVncDDYHnLhea+o=
+golang.org/x/mod v0.0.0-20190513183733-4bf6d317e70e/go.mod h1:mXi4GBBbnImb6dmsKGUJ2LatrhH/nqhxcFungHvyanc=
+golang.org/x/mod v0.1.0/go.mod h1:0QHyrYULN0/3qlju5TqG8bIK38QM8yzMo5ekMj3DlcY=
+golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20180826012351-8a410e7b638d/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20180911220305-26e67e76b6c3/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
-golang.org/x/sys v0.0.0-20190813064441-fde4db37ae7a h1:aYOabOQFp6Vj6W1F80affTUvO9UxmJRx8K0gsfABByQ=
-golang.org/x/sys v0.0.0-20190813064441-fde4db37ae7a/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/net v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20190213061140-3a22650c66bd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20190311183353-d8887717615a/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190501004415-9ce7a6920f09/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190503192946-f4e77d36d62c/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
+golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20191126235420-ef20fe5d7933 h1:e6HwijUxhDe+hPNjZQQn9bA5PW3vNmnN64U2ZW759Lk=
+golang.org/x/net v0.0.0-20191126235420-ef20fe5d7933/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/oauth2 v0.0.0-20180821212333-d2e6202438be/go.mod h1:N/0e6XlmueqKjAGxoOufVs8QHGRruUQn6yWY3a++T0U=
+golang.org/x/oauth2 v0.0.0-20190226205417-e64efc72b421/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
+golang.org/x/oauth2 v0.0.0-20190604053449-0f29369cfe45 h1:SVwTIAaPC2U/AvvLNZ2a7OVsmBpC8L5BlwK1whH3hm0=
+golang.org/x/oauth2 v0.0.0-20190604053449-0f29369cfe45/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
+golang.org/x/oauth2 v0.0.0-20191122200657-5d9234df094c h1:HjRaKPaiWks0f5tA6ELVF7ZfqSppfPwOEEAvsrKUTO4=
+golang.org/x/oauth2 v0.0.0-20191122200657-5d9234df094c/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
+golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20181221193216-37e7f081c4d4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20190227155943-e225da77a7e6/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20190423024810-112230192c58 h1:8gQV6CLnAEikrhgkHFbMAEhagSSnXWGV915qUMm9mrU=
+golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sys v0.0.0-20180830151530-49385e6e1522/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20190312061237-fead79001313/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190502145724-3ef323f4f1fd/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190507160741-ecd444e8653b/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190606165138-5da285871e9c/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190624142023-c5567b49c5d0/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20191127021746-63cb32ae39b2 h1:/J2nHFg1MTqaRLFO7M+J78ASNsJoz3r0cvHBPQ77fsE=
+golang.org/x/sys v0.0.0-20191127021746-63cb32ae39b2/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
+golang.org/x/text v0.3.1-0.20180807135948-17ff2d5776d2/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
+golang.org/x/text v0.3.2 h1:tW2bmiBqwgJj/UpqtC8EpXEZVYOwU0yG4iWbprSVAcs=
+golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
+golang.org/x/time v0.0.0-20181108054448-85acf8d2951c/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ=
+golang.org/x/time v0.0.0-20190308202827-9d24e82272b4/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20190114222345-bf090417da8b/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20190226205152-f727befe758c/go.mod h1:9Yl7xja0Znq3iFh3HoIrodX9oNMXvdceNzlUR8zjMvY=
+golang.org/x/tools v0.0.0-20190311212946-11955173bddd/go.mod h1:LCzVGOaR6xXOjkQ3onu1FJEFr0SW1gC7cKk1uF8kGRs=
+golang.org/x/tools v0.0.0-20190312151545-0bb0c0a6e846/go.mod h1:LCzVGOaR6xXOjkQ3onu1FJEFr0SW1gC7cKk1uF8kGRs=
+golang.org/x/tools v0.0.0-20190312170243-e65039ee4138/go.mod h1:LCzVGOaR6xXOjkQ3onu1FJEFr0SW1gC7cKk1uF8kGRs=
+golang.org/x/tools v0.0.0-20190425150028-36563e24a262/go.mod h1:RgjU9mgBXZiqYHBnxXauZ1Gv1EHHAz9KjViQ78xBX0Q=
+golang.org/x/tools v0.0.0-20190506145303-2d16b83fe98c/go.mod h1:RgjU9mgBXZiqYHBnxXauZ1Gv1EHHAz9KjViQ78xBX0Q=
+golang.org/x/tools v0.0.0-20190524140312-2c0ae7006135/go.mod h1:RgjU9mgBXZiqYHBnxXauZ1Gv1EHHAz9KjViQ78xBX0Q=
+golang.org/x/tools v0.0.0-20190606124116-d0a3d012864b/go.mod h1:/rFqwRUd4F7ZHNgwSSTFct+R/Kf4OFW1sUzUTQQTgfc=
+golang.org/x/tools v0.0.0-20190621195816-6e04913cbbac/go.mod h1:/rFqwRUd4F7ZHNgwSSTFct+R/Kf4OFW1sUzUTQQTgfc=
+golang.org/x/tools v0.0.0-20190628153133-6cdbf07be9d0/go.mod h1:/rFqwRUd4F7ZHNgwSSTFct+R/Kf4OFW1sUzUTQQTgfc=
+golang.org/x/tools v0.0.0-20190816200558-6889da9d5479/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20190911174233-4f2ddba30aff/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191012152004-8de300cfc20a/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191115202509-3a792d9c32b2/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191125144606-a911d9008d1f/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191127064951-724660f1afeb h1:K4JMHRJSgd1q/yXZNrKKyneQJcLm1rn7JsokEs/xE9I=
+golang.org/x/tools v0.0.0-20191127064951-724660f1afeb/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+google.golang.org/api v0.4.0/go.mod h1:8k5glujaEP+g9n7WNsDg8QP6cUVNI86fCNMcbazEtwE=
+google.golang.org/api v0.7.0/go.mod h1:WtwebWUNSVBH/HAw79HIFXZNqEvBhG+Ra+ax0hx3E3M=
+google.golang.org/api v0.8.0/go.mod h1:o4eAsZoiT+ibD93RtjEohWalFOjRDx6CVaqeizhEnKg=
+google.golang.org/api v0.9.0/go.mod h1:o4eAsZoiT+ibD93RtjEohWalFOjRDx6CVaqeizhEnKg=
+google.golang.org/api v0.14.0 h1:uMf5uLi4eQMRrMKhCplNik4U4H8Z6C1br3zOtAa/aDE=
+google.golang.org/api v0.14.0/go.mod h1:iLdEw5Ide6rF15KTC1Kkl0iskquN2gFfn9o9XIsbkAI=
+google.golang.org/appengine v1.1.0/go.mod h1:EbEs0AVv82hx2wNQdGPgUI5lhzA/G0D9YwlJXL52JkM=
+google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
+google.golang.org/appengine v1.5.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
+google.golang.org/appengine v1.6.1/go.mod h1:i06prIuMbXzDqacNJfV5OdTW448YApPu5ww/cMBSeb0=
+google.golang.org/appengine v1.6.5 h1:tycE03LOZYQNhDpS27tcQdAzLCVMaj7QT2SXxebnpCM=
+google.golang.org/appengine v1.6.5/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
+google.golang.org/genproto v0.0.0-20180817151627-c66870c02cf8/go.mod h1:JiN7NxoALGmiZfu7CAH4rXhgtRTLTxftemlI0sWmxmc=
+google.golang.org/genproto v0.0.0-20190307195333-5fe7a883aa19/go.mod h1:VzzqZJRnGkLBvHegQrXjBqPurQTc5/KpmUdxsrq26oE=
+google.golang.org/genproto v0.0.0-20190418145605-e7d98fc518a7/go.mod h1:VzzqZJRnGkLBvHegQrXjBqPurQTc5/KpmUdxsrq26oE=
+google.golang.org/genproto v0.0.0-20190425155659-357c62f0e4bb/go.mod h1:VzzqZJRnGkLBvHegQrXjBqPurQTc5/KpmUdxsrq26oE=
+google.golang.org/genproto v0.0.0-20190502173448-54afdca5d873/go.mod h1:VzzqZJRnGkLBvHegQrXjBqPurQTc5/KpmUdxsrq26oE=
+google.golang.org/genproto v0.0.0-20190801165951-fa694d86fc64/go.mod h1:DMBHOl98Agz4BDEuKkezgsaosCRResVns1a3J2ZsMNc=
+google.golang.org/genproto v0.0.0-20190819201941-24fa4b261c55/go.mod h1:DMBHOl98Agz4BDEuKkezgsaosCRResVns1a3J2ZsMNc=
+google.golang.org/genproto v0.0.0-20190911173649-1774047e7e51/go.mod h1:IbNlFCBrqXvoKpeg0TB2l7cyZUmoaFKYIwrEpbDKLA8=
+google.golang.org/genproto v0.0.0-20191115194625-c23dd37a84c9/go.mod h1:n3cpQtvxv34hfy77yVDNjmbRyujviMdxYliBSkLhpCc=
+google.golang.org/genproto v0.0.0-20191115221424-83cc0476cb11 h1:51D++eCgOHufw5VfDE9Uzqyyc+OyQIjb9hkYy9LN5Fk=
+google.golang.org/genproto v0.0.0-20191115221424-83cc0476cb11/go.mod h1:n3cpQtvxv34hfy77yVDNjmbRyujviMdxYliBSkLhpCc=
+google.golang.org/grpc v1.19.0/go.mod h1:mqu4LbDTu4XGKhr4mRzUsmM4RtVoemTSY81AxZiDr8c=
+google.golang.org/grpc v1.20.1/go.mod h1:10oTOabMzJvdu6/UiuZezV6QK5dSlG84ov/aaiqXj38=
+google.golang.org/grpc v1.21.1/go.mod h1:oYelfM1adQP15Ek0mdvEgi9Df8B9CZIaU1084ijfRaM=
+google.golang.org/grpc v1.23.0/go.mod h1:Y5yQAOtifL1yxbo5wqy6BxZv8vAUGQwXBOALyacEbxg=
+google.golang.org/grpc v1.25.1 h1:wdKvqQk7IttEw92GoRyKG2IDrUIpgpj6H6m81yfeMW0=
+google.golang.org/grpc v1.25.1/go.mod h1:c3i+UQWmh7LiEpx4sFZnkU36qjEYZ0imhYfXVyQciAY=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127 h1:qIbj1fsPNlZgppZ+VLlY7N33q108Sa+fhmuc+sWQYwY=
+gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/errgo.v2 v2.1.0/go.mod h1:hNsd1EY+bozCKY1Ytp96fpM3vjJbqLJn88ws8XvfDNI=
+gopkg.in/yaml.v2 v2.2.2 h1:ZCJp+EgiOT7lHqUV2J862kp8Qj64Jo6az82+3Td9dZw=
+gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
+honnef.co/go/tools v0.0.0-20190102054323-c2f93a96b099/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
+honnef.co/go/tools v0.0.0-20190106161140-3f1c8253044a/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
+honnef.co/go/tools v0.0.0-20190418001031-e561f6794a2a/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
+honnef.co/go/tools v0.0.0-20190523083050-ea95bdfd59fc/go.mod h1:rf3lG4BRIbNafJWhAfAdb/ePZxsR/4RtNHQocxwk9r4=
+honnef.co/go/tools v0.0.1-2019.2.3 h1:3JgtbtFHMiCmsznwGVTUWbgGov+pVqnlf1dEJTNAXeM=
+honnef.co/go/tools v0.0.1-2019.2.3/go.mod h1:a3bituU0lyd329TUQxRnasdCoJDkEUEAqEt0JzvZhAg=
+rsc.io/binaryregexp v0.2.0/go.mod h1:qTv7/COck+e2FymRvadv62gMdZztPaShugOCi3I+8D8=

lib/backup/actions/backup.go

@@ -0,0 +1,167 @@
+package actions
+
+import (
+	"fmt"
+	"io"
+	"sync/atomic"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fslocal"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fsnil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// Backup performs backup according to the provided settings.
+//
+// Note that the backup works only for VictoriaMetrics snapshots
+// made via `/snapshot/create`. It works improperly on mutable files.
+type Backup struct {
+	// Concurrency is the number of concurrent workers during the backup.
+	// Concurrency=1 by default.
+	Concurrency int
+
+	// Src is backup source
+	Src *fslocal.FS
+
+	// Dst is backup destination.
+	//
+	// If dst contains the previous backup data, then incremental backup
+	// is made, i.e. only the changed data is uploaded.
+	//
+	// If dst points to empty dir, then full backup is made.
+	// Origin can be set to the previous backup in order to reduce backup duration
+	// and reduce network bandwidth usage.
+	Dst common.RemoteFS
+
+	// Origin is optional origin for speeding up full backup if Dst points
+	// to empty dir.
+	Origin common.OriginFS
+}
+
+// Run runs b with the provided settings.
+func (b *Backup) Run() error {
+	startTime := time.Now()
+
+	concurrency := b.Concurrency
+	src := b.Src
+	dst := b.Dst
+	origin := b.Origin
+
+	if origin == nil {
+		origin = &fsnil.FS{}
+	}
+
+	logger.Infof("starting backup from %s to %s using origin %s", src, dst, origin)
+
+	logger.Infof("obtaining list of parts at %s", src)
+	srcParts, err := src.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list src parts: %s", err)
+	}
+	logger.Infof("obtaining list of parts at %s", dst)
+	dstParts, err := dst.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list dst parts: %s", err)
+	}
+	logger.Infof("obtaining list of parts at %s", origin)
+	originParts, err := origin.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list origin parts: %s", err)
+	}
+
+	backupSize := getPartsSize(srcParts)
+
+	partsToDelete := common.PartsDifference(dstParts, srcParts)
+	deleteSize := getPartsSize(partsToDelete)
+	if len(partsToDelete) > 0 {
+		logger.Infof("deleting %d parts from %s", len(partsToDelete), dst)
+		deletedParts := uint64(0)
+		err = runParallel(concurrency, partsToDelete, func(p common.Part) error {
+			logger.Infof("deleting %s from %s", &p, dst)
+			if err := dst.DeletePart(p); err != nil {
+				return fmt.Errorf("cannot delete %s from %s: %s", &p, dst, err)
+			}
+			atomic.AddUint64(&deletedParts, 1)
+			return nil
+		}, func(elapsed time.Duration) {
+			n := atomic.LoadUint64(&deletedParts)
+			logger.Infof("deleted %d out of %d parts from %s in %s", n, len(partsToDelete), dst, elapsed)
+		})
+		if err != nil {
+			return err
+		}
+		if err := dst.RemoveEmptyDirs(); err != nil {
+			return fmt.Errorf("cannot remove empty directories at %s: %s", dst, err)
+		}
+	}
+
+	partsToCopy := common.PartsDifference(srcParts, dstParts)
+	originCopyParts := common.PartsIntersect(originParts, partsToCopy)
+	copySize := getPartsSize(originCopyParts)
+	if len(originCopyParts) > 0 {
+		logger.Infof("server-side copying %d parts from %s to %s", len(originCopyParts), origin, dst)
+		copiedParts := uint64(0)
+		err = runParallel(concurrency, originCopyParts, func(p common.Part) error {
+			logger.Infof("server-side copying %s from %s to %s", &p, origin, dst)
+			if err := dst.CopyPart(origin, p); err != nil {
+				return fmt.Errorf("cannot copy %s from %s to %s: %s", &p, origin, dst, err)
+			}
+			atomic.AddUint64(&copiedParts, 1)
+			return nil
+		}, func(elapsed time.Duration) {
+			n := atomic.LoadUint64(&copiedParts)
+			logger.Infof("server-side copied %d out of %d parts from %s to %s in %s", n, len(originCopyParts), origin, dst, elapsed)
+		})
+		if err != nil {
+			return err
+		}
+	}
+
+	srcCopyParts := common.PartsDifference(partsToCopy, originParts)
+	uploadSize := getPartsSize(srcCopyParts)
+	if len(srcCopyParts) > 0 {
+		logger.Infof("uploading %d parts from %s to %s", len(srcCopyParts), src, dst)
+		bytesUploaded := uint64(0)
+		err = runParallel(concurrency, srcCopyParts, func(p common.Part) error {
+			logger.Infof("uploading %s from %s to %s", &p, src, dst)
+			rc, err := src.NewReadCloser(p)
+			if err != nil {
+				return fmt.Errorf("cannot create reader for %s from %s: %s", &p, src, err)
+			}
+			sr := &statReader{
+				r:         rc,
+				bytesRead: &bytesUploaded,
+			}
+			if err := dst.UploadPart(p, sr); err != nil {
+				return fmt.Errorf("cannot upload %s to %s: %s", &p, dst, err)
+			}
+			if err = rc.Close(); err != nil {
+				return fmt.Errorf("cannot close reader for %s from %s: %s", &p, src, err)
+			}
+			return nil
+		}, func(elapsed time.Duration) {
+			n := atomic.LoadUint64(&bytesUploaded)
+			logger.Infof("uploaded %d out of %d bytes from %s to %s in %s", n, uploadSize, src, dst, elapsed)
+		})
+		if err != nil {
+			return err
+		}
+	}
+
+	logger.Infof("backed up %d bytes in %s; deleted %d bytes; server-side copied %d bytes; uploaded %d bytes",
+		backupSize, time.Since(startTime), deleteSize, copySize, uploadSize)
+
+	return nil
+}
+
+type statReader struct {
+	r         io.Reader
+	bytesRead *uint64
+}
+
+func (sr *statReader) Read(p []byte) (int, error) {
+	n, err := sr.r.Read(p)
+	atomic.AddUint64(sr.bytesRead, uint64(n))
+	return n, err
+}

lib/backup/actions/restore.go

@@ -0,0 +1,181 @@
+package actions
+
+import (
+	"fmt"
+	"io"
+	"sync/atomic"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fslocal"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// Restore restores data according to the provided settings.
+//
+// Note that the restore works only for VictoriaMetrics backups made from snapshots.
+// It works improperly on mutable files.
+type Restore struct {
+	// Concurrency is the number of concurrent workers to run during restore.
+	// Concurrency=1 is used by default.
+	Concurrency int
+
+	// Src is the source containing backed up data.
+	Src common.RemoteFS
+
+	// Dst is destination to restore the data.
+	//
+	// If dst points to existing directory, then incremental restore is performed,
+	// i.e. only new data is downloaded from src.
+	Dst *fslocal.FS
+}
+
+// Run runs r with the provided settings.
+func (r *Restore) Run() error {
+	startTime := time.Now()
+
+	// Make sure VictoriaMetrics doesn't run during the restore process.
+	if err := fs.MkdirAllIfNotExist(r.Dst.Dir); err != nil {
+		return fmt.Errorf("cannot create dir %q: %s", r.Dst.Dir, err)
+	}
+	flockF, err := fs.CreateFlockFile(r.Dst.Dir)
+	if err != nil {
+		return fmt.Errorf("cannot create lock file in %q; make sure VictoriaMetrics doesn't use the dir; error: %s", r.Dst.Dir, err)
+	}
+	defer fs.MustClose(flockF)
+
+	concurrency := r.Concurrency
+	src := r.Src
+	dst := r.Dst
+	logger.Infof("starting restore from %s to %s", src, dst)
+
+	logger.Infof("obtaining list of parts at %s", src)
+	srcParts, err := src.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list src parts: %s", err)
+	}
+	logger.Infof("obtaining list of parts at %s", dst)
+	dstParts, err := dst.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list dst parts: %s", err)
+	}
+
+	backupSize := getPartsSize(srcParts)
+
+	// Validate srcParts. They must cover the whole files.
+	common.SortParts(srcParts)
+	offset := uint64(0)
+	var pOld common.Part
+	var path string
+	for _, p := range srcParts {
+		if p.Path != path {
+			if offset != pOld.FileSize {
+				return fmt.Errorf("invalid size for %q; got %d; want %d", path, offset, pOld.FileSize)
+			}
+			pOld = p
+			path = p.Path
+			offset = 0
+		}
+		if p.Offset < offset {
+			return fmt.Errorf("there is an overlap in %d bytes between %s and %s", offset-p.Offset, &pOld, &p)
+		}
+		if p.Offset > offset {
+			if offset == 0 {
+				return fmt.Errorf("there is a gap in %d bytes from file start to %s", p.Offset, &p)
+			}
+			return fmt.Errorf("there is a gap in %d bytes between %s and %s", p.Offset-offset, &pOld, &p)
+		}
+		if p.Size != p.ActualSize {
+			return fmt.Errorf("invalid size for %s; got %d; want %d", &p, p.ActualSize, p.Size)
+		}
+		offset += p.Size
+	}
+
+	partsToDelete := common.PartsDifference(dstParts, srcParts)
+	deleteSize := uint64(0)
+	if len(partsToDelete) > 0 {
+		// Fully remove local file if certain parts from the remote part are missing.
+		pathsToDelete := make(map[string]bool)
+		for _, p := range partsToDelete {
+			pathsToDelete[p.Path] = true
+		}
+		logger.Infof("deleting %d files from %s", len(pathsToDelete), dst)
+		for path := range pathsToDelete {
+			logger.Infof("deleting %s from %s", path, dst)
+			size, err := dst.DeletePath(path)
+			if err != nil {
+				return fmt.Errorf("cannot delete %s from %s: %s", path, dst, err)
+			}
+			deleteSize += size
+		}
+		if err != nil {
+			return err
+		}
+		if err := dst.RemoveEmptyDirs(); err != nil {
+			return fmt.Errorf("cannot remove empty directories at %s: %s", dst, err)
+		}
+	}
+
+	// Re-read dstParts, since additional parts may be removed on the previous step.
+	dstParts, err = dst.ListParts()
+	if err != nil {
+		return fmt.Errorf("cannot list dst parts after the deletion: %s", err)
+	}
+
+	partsToCopy := common.PartsDifference(srcParts, dstParts)
+	downloadSize := getPartsSize(partsToCopy)
+	if len(partsToCopy) > 0 {
+		perPath := make(map[string][]common.Part)
+		for _, p := range partsToCopy {
+			parts := perPath[p.Path]
+			parts = append(parts, p)
+			perPath[p.Path] = parts
+		}
+		logger.Infof("downloading %d parts from %s to %s", len(partsToCopy), src, dst)
+		bytesDownloaded := uint64(0)
+		err = runParallelPerPath(concurrency, perPath, func(parts []common.Part) error {
+			// Sort partsToCopy in order to properly grow file size during downloading.
+			common.SortParts(parts)
+			for _, p := range parts {
+				logger.Infof("downloading %s from %s to %s", &p, src, dst)
+				wc, err := dst.NewWriteCloser(p)
+				if err != nil {
+					return fmt.Errorf("cannot create writer for %q to %s: %s", &p, dst, err)
+				}
+				sw := &statWriter{
+					w:            wc,
+					bytesWritten: &bytesDownloaded,
+				}
+				if err := src.DownloadPart(p, sw); err != nil {
+					return fmt.Errorf("cannot download %s to %s: %s", &p, dst, err)
+				}
+				if err := wc.Close(); err != nil {
+					return fmt.Errorf("cannot close reader fro %s from %s: %s", &p, src, err)
+				}
+			}
+			return nil
+		}, func(elapsed time.Duration) {
+			n := atomic.LoadUint64(&bytesDownloaded)
+			logger.Infof("downloaded %d out of %d bytes from %s to %s in %s", n, downloadSize, src, dst, elapsed)
+		})
+		if err != nil {
+			return err
+		}
+	}
+
+	logger.Infof("restored %d bytes from backup in %s; deleted %d bytes; downloaded %d bytes", backupSize, time.Since(startTime), deleteSize, downloadSize)
+
+	return nil
+}
+
+type statWriter struct {
+	w            io.Writer
+	bytesWritten *uint64
+}
+
+func (sw *statWriter) Write(p []byte) (int, error) {
+	n, err := sw.w.Write(p)
+	atomic.AddUint64(sw.bytesWritten, uint64(n))
+	return n, err
+}

lib/backup/actions/util.go

@@ -0,0 +1,235 @@
+package actions
+
+import (
+	"flag"
+	"fmt"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fsremote"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/gcsremote"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/s3remote"
+)
+
+var (
+	credsFilePath = flag.String("credsFilePath", "", "Path to file with GCS or S3 credentials. Credentials are loaded from default locations if not set.\n"+
+		"See https://cloud.google.com/iam/docs/creating-managing-service-account-keys and https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html")
+	configFilePath = flag.String("configFilePath", "", "Path to file with S3 configs. Configs are loaded from default location if not set.\n"+
+		"See https://docs.aws.amazon.com/general/latest/gr/aws-security-credentials.html")
+	configProfile    = flag.String("configProfile", "default", "Profile name for S3 configs")
+	customS3Endpoint = flag.String("customS3Endpoint", "", "Custom S3 endpoint for use with S3-compatible storages (e.g. MinIO). S3 is used if not set")
+)
+
+func runParallel(concurrency int, parts []common.Part, f func(p common.Part) error, progress func(elapsed time.Duration)) error {
+	var err error
+	runWithProgress(progress, func() {
+		err = runParallelInternal(concurrency, parts, f)
+	})
+	return err
+}
+
+func runParallelPerPath(concurrency int, perPath map[string][]common.Part, f func(parts []common.Part) error, progress func(elapsed time.Duration)) error {
+	var err error
+	runWithProgress(progress, func() {
+		err = runParallelPerPathInternal(concurrency, perPath, f)
+	})
+	return err
+}
+
+func runParallelPerPathInternal(concurrency int, perPath map[string][]common.Part, f func(parts []common.Part) error) error {
+	if concurrency <= 0 {
+		concurrency = 1
+	}
+	if len(perPath) == 0 {
+		return nil
+	}
+
+	// len(perPath) capacity guarantees non-blocking behavior below.
+	resultCh := make(chan error, len(perPath))
+	workCh := make(chan []common.Part, len(perPath))
+	stopCh := make(chan struct{})
+
+	// Start workers
+	var wg sync.WaitGroup
+	wg.Add(concurrency)
+	for i := 0; i < concurrency; i++ {
+		go func() {
+			defer wg.Done()
+			for parts := range workCh {
+				select {
+				case <-stopCh:
+					return
+				default:
+				}
+				resultCh <- f(parts)
+			}
+		}()
+	}
+
+	// Feed workers with work.
+	for _, parts := range perPath {
+		workCh <- parts
+	}
+	close(workCh)
+
+	// Read results.
+	var err error
+	for i := 0; i < len(perPath); i++ {
+		err = <-resultCh
+		if err != nil {
+			// Stop the work.
+			close(stopCh)
+			break
+		}
+	}
+
+	// Wait for all the workers to stop.
+	wg.Wait()
+
+	return err
+}
+
+func runParallelInternal(concurrency int, parts []common.Part, f func(p common.Part) error) error {
+	if concurrency <= 0 {
+		concurrency = 1
+	}
+	if len(parts) == 0 {
+		return nil
+	}
+
+	// len(parts) capacity guarantees non-blocking behavior below.
+	resultCh := make(chan error, len(parts))
+	workCh := make(chan common.Part, len(parts))
+	stopCh := make(chan struct{})
+
+	// Start workers
+	var wg sync.WaitGroup
+	wg.Add(concurrency)
+	for i := 0; i < concurrency; i++ {
+		go func() {
+			defer wg.Done()
+			for p := range workCh {
+				select {
+				case <-stopCh:
+					return
+				default:
+				}
+				resultCh <- f(p)
+			}
+		}()
+	}
+
+	// Feed workers with work.
+	for _, p := range parts {
+		workCh <- p
+	}
+	close(workCh)
+
+	// Read results.
+	var err error
+	for i := 0; i < len(parts); i++ {
+		err = <-resultCh
+		if err != nil {
+			// Stop the work.
+			close(stopCh)
+			break
+		}
+	}
+
+	// Wait for all the workers to stop.
+	wg.Wait()
+
+	return err
+}
+
+func runWithProgress(progress func(elapsed time.Duration), f func()) {
+	startTime := time.Now()
+	doneCh := make(chan struct{})
+	go func() {
+		f()
+		close(doneCh)
+	}()
+
+	tc := time.NewTicker(10 * time.Second)
+	for {
+		select {
+		case <-doneCh:
+			tc.Stop()
+			// The last progress call.
+			progress(time.Since(startTime))
+			return
+		case <-tc.C:
+			progress(time.Since(startTime))
+		}
+	}
+}
+
+func getPartsSize(parts []common.Part) uint64 {
+	n := uint64(0)
+	for _, p := range parts {
+		n += p.Size
+	}
+	return n
+}
+
+// NewRemoteFS returns new remote fs from the given path.
+func NewRemoteFS(path string) (common.RemoteFS, error) {
+	if len(path) == 0 {
+		return nil, fmt.Errorf("path cannot be empty")
+	}
+	n := strings.Index(path, "://")
+	if n < 0 {
+		return nil, fmt.Errorf("Missing scheme in path %q. Supported schemes: `gcs://`, `s3://`, `fs://`", path)
+	}
+	scheme := path[:n]
+	dir := path[n+len("://"):]
+	switch scheme {
+	case "fs":
+		if !strings.HasPrefix(dir, "/") {
+			return nil, fmt.Errorf("dir must be absolute; got %q", dir)
+		}
+		fs := &fsremote.FS{
+			Dir: dir,
+		}
+		return fs, nil
+	case "gcs":
+		n := strings.Index(dir, "/")
+		if n < 0 {
+			return nil, fmt.Errorf("missing directory on the gcs bucket %q", dir)
+		}
+		bucket := dir[:n]
+		dir = dir[n:]
+		fs := &gcsremote.FS{
+			CredsFilePath: *credsFilePath,
+			Bucket:        bucket,
+			Dir:           dir,
+		}
+		if err := fs.Init(); err != nil {
+			return nil, fmt.Errorf("cannot initialize connection to gcs: %s", err)
+		}
+		return fs, nil
+	case "s3":
+		n := strings.Index(dir, "/")
+		if n < 0 {
+			return nil, fmt.Errorf("missing directory on the s3 bucket %q", dir)
+		}
+		bucket := dir[:n]
+		dir = dir[n:]
+		fs := &s3remote.FS{
+			CredsFilePath:  *credsFilePath,
+			ConfigFilePath: *configFilePath,
+			CustomEndpoint: *customS3Endpoint,
+			ProfileName:    *configProfile,
+			Bucket:         bucket,
+			Dir:            dir,
+		}
+		if err := fs.Init(); err != nil {
+			return nil, fmt.Errorf("cannot initialize connection to s3: %s", err)
+		}
+		return fs, nil
+	default:
+		return nil, fmt.Errorf("unsupported scheme %q in `-dst`", scheme)
+	}
+}

lib/backup/common/fs.go

@@ -0,0 +1,41 @@
+package common
+
+import (
+	"io"
+)
+
+// OriginFS is an interface for remote origin filesystem.
+//
+// This filesystem is used for performing server-side file copies
+// instead of uploading data from local filesystem.
+type OriginFS interface {
+	// String must return human-readable representation of OriginFS.
+	String() string
+
+	// ListParts must return all the parts for the OriginFS.
+	ListParts() ([]Part, error)
+}
+
+// RemoteFS is a filesystem where backups are stored.
+type RemoteFS interface {
+	// String must return human-readable representation of RemoteFS.
+	String() string
+
+	// ListParts must return all the parts for the RemoteFS.
+	ListParts() ([]Part, error)
+
+	// DeletePart must delete part p from RemoteFS.
+	DeletePart(p Part) error
+
+	// RemoveEmptyDirs must recursively remove empty directories in RemoteFS.
+	RemoveEmptyDirs() error
+
+	// CopyPart must copy part p from dstFS to RemoteFS.
+	CopyPart(dstFS OriginFS, p Part) error
+
+	// DownloadPart must download part p from RemoteFS to w.
+	DownloadPart(p Part, w io.Writer) error
+
+	// UploadPart must upload part p from r to RemoteFS.
+	UploadPart(p Part, r io.Reader) error
+}

lib/backup/common/part.go

@@ -0,0 +1,136 @@
+package common
+
+import (
+	"fmt"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// Part is an atomic unit for transfer during backup / restore.
+//
+// Each source file can be split into parts with up to MaxPartSize sizes.
+type Part struct {
+	// Path is the path to file for backup.
+	Path string
+
+	// FileSize is the size of the whole file for the given part.
+	FileSize uint64
+
+	// Offset is offset in the file to backup.
+	Offset uint64
+
+	// Size is the size of the part to backup starting from Offset.
+	Size uint64
+
+	// ActualSize is the actual size of the part.
+	//
+	// The part is considered broken if it isn't equal to Size.
+	// Such a part must be removed from remote storage.
+	ActualSize uint64
+}
+
+func (p *Part) key() string {
+	return fmt.Sprintf("%s%016X%016X%016X%016X", p.Path, p.FileSize, p.Offset, p.Size, p.ActualSize)
+}
+
+// String returns human-readable representation of the part.
+func (p *Part) String() string {
+	return fmt.Sprintf("part{path: %q, file_size: %d, offset: %d, size: %d}", p.Path, p.FileSize, p.Offset, p.Size)
+}
+
+// RemotePath returns remote path for the part p and the given prefix.
+func (p *Part) RemotePath(prefix string) string {
+	for strings.HasSuffix(prefix, "/") {
+		prefix = prefix[:len(prefix)-1]
+	}
+	return fmt.Sprintf("%s/%s/%016X_%016X_%016X", prefix, p.Path, p.FileSize, p.Offset, p.Size)
+}
+
+var partNameRegexp = regexp.MustCompile(`^(.+)/([0-9A-F]{16})_([0-9A-F]{16})_([0-9A-F]{16})$`)
+
+// ParseFromRemotePath parses p from remotePath.
+//
+// Returns true on success.
+func (p *Part) ParseFromRemotePath(remotePath string) bool {
+	tmp := partNameRegexp.FindStringSubmatch(remotePath)
+	if len(tmp) != 5 {
+		return false
+	}
+	path := tmp[1]
+	for strings.HasPrefix(path, "/") {
+		path = path[1:]
+	}
+	fileSize, err := strconv.ParseUint(tmp[2], 16, 64)
+	if err != nil {
+		logger.Panicf("BUG: cannot parse fileSize from %q: %s", tmp[2], err)
+	}
+	offset, err := strconv.ParseUint(tmp[3], 16, 64)
+	if err != nil {
+		logger.Panicf("BUG: cannot parse offset from %q: %s", tmp[3], err)
+	}
+	size, err := strconv.ParseUint(tmp[4], 16, 64)
+	if err != nil {
+		logger.Panicf("BUG: cannot parse size from %q: %s", tmp[4], err)
+	}
+	p.Path = path
+	p.FileSize = fileSize
+	p.Offset = offset
+	p.Size = size
+	return true
+}
+
+// MaxPartSize is the maximum size for each part.
+//
+// The MaxPartSize reduces bandwidth usage during retires on network errors
+// when transferring multi-TB files.
+const MaxPartSize = 128 * 1024 * 1024
+
+// SortParts sorts parts by (Path, Offset)
+func SortParts(parts []Part) {
+	sort.Slice(parts, func(i, j int) bool {
+		a := parts[i]
+		b := parts[j]
+		if a.Path != b.Path {
+			return a.Path < b.Path
+		}
+		return a.Offset < b.Offset
+	})
+}
+
+// PartsDifference returns a - b
+func PartsDifference(a, b []Part) []Part {
+	m := make(map[string]bool, len(b))
+	for _, p := range b {
+		k := p.key()
+		m[k] = true
+	}
+	var d []Part
+	for _, p := range a {
+		k := p.key()
+		if !m[k] {
+			d = append(d, p)
+		}
+	}
+	return d
+}
+
+// PartsIntersect returns the intersection of a and b
+func PartsIntersect(a, b []Part) []Part {
+	m := make(map[string]bool, len(a))
+	for _, p := range a {
+		k := p.key()
+		m[k] = true
+	}
+	var d []Part
+	for _, p := range b {
+		k := p.key()
+		if m[k] {
+			d = append(d, p)
+		}
+	}
+	return d
+}

lib/backup/fscommon/fscommon.go

@@ -0,0 +1,262 @@
+package fscommon
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// FsyncFile fsyncs path contents and the parent directory contents.
+func FsyncFile(path string) error {
+	if err := fsync(path); err != nil {
+		_ = os.RemoveAll(path)
+		return fmt.Errorf("cannot fsync file %q: %s", path, err)
+	}
+	dir := filepath.Dir(path)
+	if err := fsync(dir); err != nil {
+		return fmt.Errorf("cannot fsync dir %q: %s", dir, err)
+	}
+	return nil
+}
+
+// FsyncDir fsyncs dir contents.
+func FsyncDir(dir string) error {
+	return fsync(dir)
+}
+
+func fsync(path string) error {
+	f, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	if err := f.Sync(); err != nil {
+		_ = f.Close()
+		return err
+	}
+	return f.Close()
+}
+
+// AppendFiles appends all the files from dir to dst.
+//
+// All the appended files will have dir prefix.
+func AppendFiles(dst []string, dir string) ([]string, error) {
+	d, err := os.Open(dir)
+	if err != nil {
+		return nil, fmt.Errorf("cannot open %q: %s", dir, err)
+	}
+	dst, err = appendFilesInternal(dst, d)
+	if err1 := d.Close(); err1 != nil {
+		err = err1
+	}
+	return dst, err
+}
+
+func appendFilesInternal(dst []string, d *os.File) ([]string, error) {
+	dir := d.Name()
+	dfi, err := d.Stat()
+	if err != nil {
+		return nil, fmt.Errorf("cannot stat %q: %s", dir, err)
+	}
+	if !dfi.IsDir() {
+		return nil, fmt.Errorf("%q isn't a directory", dir)
+	}
+	fis, err := d.Readdir(-1)
+	if err != nil {
+		return nil, fmt.Errorf("cannot read directory contents in %q: %s", dir, err)
+	}
+	for _, fi := range fis {
+		name := fi.Name()
+		if name == "." || name == ".." {
+			continue
+		}
+		if name == "flock.lock" {
+			// Do not take into account flock.lock files, since they are used
+			// for preventing from concurrent access.
+			continue
+		}
+		path := dir + "/" + name
+		if fi.IsDir() {
+			// Process directory
+			dst, err = AppendFiles(dst, path)
+			if err != nil {
+				return nil, fmt.Errorf("cannot list %q: %s", path, err)
+			}
+			continue
+		}
+		if fi.Mode()&os.ModeSymlink != os.ModeSymlink {
+			// Process file
+			dst = append(dst, path)
+			continue
+		}
+		pathOrig := path
+	again:
+		// Process symlink
+		pathReal, err := filepath.EvalSymlinks(pathOrig)
+		if err != nil {
+			if os.IsNotExist(err) || strings.Contains(err.Error(), "no such file or directory") {
+				// Skip symlink that points to nowhere.
+				continue
+			}
+			return nil, fmt.Errorf("cannot resolve symlink %q: %s", pathOrig, err)
+		}
+		sfi, err := os.Stat(pathReal)
+		if err != nil {
+			return nil, fmt.Errorf("cannot stat %q from symlink %q: %s", pathReal, path, err)
+		}
+		if sfi.IsDir() {
+			// Symlink points to directory
+			dstNew, err := AppendFiles(dst, pathReal)
+			if err != nil {
+				return nil, fmt.Errorf("cannot list files at %q from symlink %q: %s", pathReal, path, err)
+			}
+			pathReal += "/"
+			for i := len(dst); i < len(dstNew); i++ {
+				x := dstNew[i]
+				if !strings.HasPrefix(x, pathReal) {
+					return nil, fmt.Errorf("unexpected prefix for path %q; want %q", x, pathReal)
+				}
+				dstNew[i] = path + "/" + x[len(pathReal):]
+			}
+			dst = dstNew
+			continue
+		}
+		if sfi.Mode()&os.ModeSymlink != os.ModeSymlink {
+			// Symlink points to file
+			dst = append(dst, path)
+			continue
+		}
+		// Symlink points to symlink. Process it again.
+		pathOrig = pathReal
+		goto again
+	}
+	return dst, nil
+}
+
+// RemoveEmptyDirs recursively removes empty directories under the given dir.
+func RemoveEmptyDirs(dir string) error {
+	_, err := removeEmptyDirs(dir)
+	return err
+}
+
+func removeEmptyDirs(dir string) (bool, error) {
+	d, err := os.Open(dir)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return true, nil
+		}
+		return false, err
+	}
+	ok, err := removeEmptyDirsInternal(d)
+	if err1 := d.Close(); err1 != nil {
+		err = err1
+	}
+	if err != nil {
+		return false, err
+	}
+	return ok, nil
+}
+
+func removeEmptyDirsInternal(d *os.File) (bool, error) {
+	dir := d.Name()
+	dfi, err := d.Stat()
+	if err != nil {
+		return false, fmt.Errorf("cannot stat %q: %s", dir, err)
+	}
+	if !dfi.IsDir() {
+		return false, fmt.Errorf("%q isn't a directory", dir)
+	}
+	fis, err := d.Readdir(-1)
+	if err != nil {
+		return false, fmt.Errorf("cannot read directory contents in %q: %s", dir, err)
+	}
+	dirEntries := 0
+	hasFlock := false
+	for _, fi := range fis {
+		name := fi.Name()
+		if name == "." || name == ".." {
+			continue
+		}
+		path := dir + "/" + name
+		if fi.IsDir() {
+			// Process directory
+			ok, err := removeEmptyDirs(path)
+			if err != nil {
+				return false, fmt.Errorf("cannot list %q: %s", path, err)
+			}
+			if !ok {
+				dirEntries++
+			}
+			continue
+		}
+		if fi.Mode()&os.ModeSymlink != os.ModeSymlink {
+			if name == "flock.lock" {
+				hasFlock = true
+				continue
+			}
+			// Skip plain files.
+			dirEntries++
+			continue
+		}
+		pathOrig := path
+	again:
+		// Process symlink
+		pathReal, err := filepath.EvalSymlinks(pathOrig)
+		if err != nil {
+			if os.IsNotExist(err) || strings.Contains(err.Error(), "no such file or directory") {
+				// Remove symlink that points to nowere.
+				logger.Infof("removing broken symlink %q", pathOrig)
+				if err := os.Remove(pathOrig); err != nil {
+					return false, fmt.Errorf("cannot remove %q: %s", pathOrig, err)
+				}
+				continue
+			}
+			return false, fmt.Errorf("cannot resolve symlink %q: %s", pathOrig, err)
+		}
+		sfi, err := os.Stat(pathReal)
+		if err != nil {
+			return false, fmt.Errorf("cannot stat %q from symlink %q: %s", pathReal, path, err)
+		}
+		if sfi.IsDir() {
+			// Symlink points to directory
+			ok, err := removeEmptyDirs(pathReal)
+			if err != nil {
+				return false, fmt.Errorf("cannot list files at %q from symlink %q: %s", pathReal, path, err)
+			}
+			if !ok {
+				dirEntries++
+			} else {
+				// Remove the symlink
+				logger.Infof("removing symlink that points to empty dir %q", pathOrig)
+				if err := os.Remove(pathOrig); err != nil {
+					return false, fmt.Errorf("cannot remove %q: %s", pathOrig, err)
+				}
+			}
+			continue
+		}
+		if sfi.Mode()&os.ModeSymlink != os.ModeSymlink {
+			// Symlink points to file. Skip it.
+			dirEntries++
+			continue
+		}
+		// Symlink points to symlink. Process it again.
+		pathOrig = pathReal
+		goto again
+	}
+	if dirEntries > 0 {
+		return false, nil
+	}
+	logger.Infof("removing empty dir %q", dir)
+	if hasFlock {
+		flockFilepath := dir + "/flock.lock"
+		if err := os.Remove(flockFilepath); err != nil {
+			return false, fmt.Errorf("cannot remove %q: %s", flockFilepath, err)
+		}
+	}
+	if err := os.Remove(dir); err != nil {
+		return false, fmt.Errorf("cannot remove %q: %s", dir, err)
+	}
+	return true, nil
+}

lib/backup/fslocal/bandwidth_limiter.go

@@ -0,0 +1,116 @@
+package fslocal
+
+import (
+	"io"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+type bandwidthLimiter struct {
+	perSecondLimit int
+
+	c *sync.Cond
+
+	// quota for the current second
+	quota int
+}
+
+func newBandwidthLimiter(perSecondLimit int) *bandwidthLimiter {
+	if perSecondLimit <= 0 {
+		logger.Panicf("BUG: perSecondLimit must be positive; got %d", perSecondLimit)
+	}
+	var bl bandwidthLimiter
+	bl.perSecondLimit = perSecondLimit
+	var mu sync.Mutex
+	bl.c = sync.NewCond(&mu)
+	go bl.perSecondUpdater()
+	return &bl
+}
+
+func (bl *bandwidthLimiter) NewReadCloser(rc io.ReadCloser) *bandwidthLimitedReader {
+	return &bandwidthLimitedReader{
+		rc: rc,
+		bl: bl,
+	}
+}
+
+func (bl *bandwidthLimiter) NewWriteCloser(wc io.WriteCloser) *bandwidthLimitedWriter {
+	return &bandwidthLimitedWriter{
+		wc: wc,
+		bl: bl,
+	}
+}
+
+type bandwidthLimitedReader struct {
+	rc io.ReadCloser
+	bl *bandwidthLimiter
+}
+
+func (blr *bandwidthLimitedReader) Read(p []byte) (int, error) {
+	quota := blr.bl.GetQuota(len(p))
+	return blr.rc.Read(p[:quota])
+}
+
+func (blr *bandwidthLimitedReader) Close() error {
+	return blr.rc.Close()
+}
+
+type bandwidthLimitedWriter struct {
+	wc io.WriteCloser
+	bl *bandwidthLimiter
+}
+
+func (blw *bandwidthLimitedWriter) Write(p []byte) (int, error) {
+	nn := 0
+	for len(p) > 0 {
+		quota := blw.bl.GetQuota(len(p))
+		n, err := blw.wc.Write(p[:quota])
+		nn += n
+		if err != nil {
+			return nn, err
+		}
+		p = p[quota:]
+	}
+	return nn, nil
+}
+
+func (blw *bandwidthLimitedWriter) Close() error {
+	return blw.wc.Close()
+}
+
+func (bl *bandwidthLimiter) perSecondUpdater() {
+	tc := time.NewTicker(time.Second)
+	c := bl.c
+	for range tc.C {
+		c.L.Lock()
+		bl.quota = bl.perSecondLimit
+		c.Signal()
+		c.L.Unlock()
+	}
+}
+
+// GetQuota returns the number in the range [1..n] - the allowed quota for now.
+//
+// The function blocks until at least 1 can be returned from it.
+func (bl *bandwidthLimiter) GetQuota(n int) int {
+	if n <= 0 {
+		logger.Panicf("BUG: n must be positive; got %d", n)
+	}
+	c := bl.c
+	c.L.Lock()
+	for bl.quota <= 0 {
+		c.Wait()
+	}
+	quota := bl.quota
+	if quota > n {
+		quota = n
+	}
+	bl.quota -= quota
+	if bl.quota > 0 {
+		c.Signal()
+	}
+	c.L.Unlock()
+	return quota
+}

lib/backup/fslocal/fslocal.go

@@ -0,0 +1,232 @@
+package fslocal
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fscommon"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/filestream"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// FS represents local filesystem.
+//
+// Backups are made from local fs.
+// Data is restored from backups to local fs.
+type FS struct {
+	// Dir is a path to local directory to work with.
+	Dir string
+
+	// MaxBytesPerSecond is the maximum bandwidth usage during backups or restores.
+	MaxBytesPerSecond int
+
+	bl *bandwidthLimiter
+}
+
+// Init initializes fs
+func (fs *FS) Init() error {
+	if fs.MaxBytesPerSecond > 0 {
+		fs.bl = newBandwidthLimiter(fs.MaxBytesPerSecond)
+	}
+	return nil
+}
+
+// String returns user-readable representation for the fs.
+func (fs *FS) String() string {
+	return fmt.Sprintf("fslocal %q", fs.Dir)
+}
+
+// ListParts returns all the parts for fs.
+func (fs *FS) ListParts() ([]common.Part, error) {
+	dir := fs.Dir
+	if _, err := os.Stat(dir); err != nil {
+		if os.IsNotExist(err) {
+			// Return empty part list for non-existing directory.
+			// The directory will be created later.
+			return nil, nil
+		}
+		return nil, err
+	}
+	files, err := fscommon.AppendFiles(nil, dir)
+	if err != nil {
+		return nil, err
+	}
+
+	var parts []common.Part
+	dir += "/"
+	for _, file := range files {
+		if !strings.HasPrefix(file, dir) {
+			logger.Panicf("BUG: unexpected prefix for file %q; want %q", file, dir)
+		}
+		fi, err := os.Stat(file)
+		if err != nil {
+			return nil, fmt.Errorf("cannot stat %q: %s", file, err)
+		}
+		path := file[len(dir):]
+		size := uint64(fi.Size())
+		if size == 0 {
+			parts = append(parts, common.Part{
+				Path:   path,
+				Offset: 0,
+				Size:   0,
+			})
+			continue
+		}
+		offset := uint64(0)
+		for offset < size {
+			n := size - offset
+			if n > common.MaxPartSize {
+				n = common.MaxPartSize
+			}
+			parts = append(parts, common.Part{
+				Path:       path,
+				FileSize:   size,
+				Offset:     offset,
+				Size:       n,
+				ActualSize: n,
+			})
+			offset += n
+		}
+	}
+	return parts, nil
+}
+
+// NewReadCloser returns io.ReadCloser for the given part p located in fs.
+func (fs *FS) NewReadCloser(p common.Part) (io.ReadCloser, error) {
+	path := fs.path(p)
+	r, err := filestream.OpenReaderAt(path, int64(p.Offset), true)
+	if err != nil {
+		return nil, fmt.Errorf("cannot open %q at %q: %s", p.Path, fs.Dir, err)
+	}
+	lrc := &limitedReadCloser{
+		r: r,
+		n: p.Size,
+	}
+	if fs.bl == nil {
+		return lrc, nil
+	}
+	blrc := fs.bl.NewReadCloser(lrc)
+	return blrc, nil
+}
+
+// NewWriteCloser returns io.WriteCloser for the given part p located in fs.
+func (fs *FS) NewWriteCloser(p common.Part) (io.WriteCloser, error) {
+	path := fs.path(p)
+	if err := fs.mkdirAll(path); err != nil {
+		return nil, err
+	}
+	w, err := filestream.OpenWriterAt(path, int64(p.Offset), true)
+	if err != nil {
+		return nil, fmt.Errorf("cannot open writer for %q at offset %d: %s", path, p.Offset, err)
+	}
+	wc := &writeCloser{
+		w:    w,
+		n:    p.Size,
+		path: path,
+	}
+	if fs.bl == nil {
+		return wc, nil
+	}
+	blwc := fs.bl.NewWriteCloser(wc)
+	return blwc, nil
+}
+
+// DeletePath deletes the given path from fs and returns the size
+// for the deleted file.
+func (fs *FS) DeletePath(path string) (uint64, error) {
+	p := common.Part{
+		Path: path,
+	}
+	fullPath := fs.path(p)
+	f, err := os.Open(fullPath)
+	if err != nil {
+		if os.IsNotExist(err) {
+			// The file could be deleted earlier via symlink.
+			return 0, nil
+		}
+		return 0, fmt.Errorf("cannot open %q at %q: %s", path, fullPath, err)
+	}
+	fi, err := f.Stat()
+	_ = f.Close()
+	if err != nil {
+		return 0, fmt.Errorf("cannot stat %q at %q: %s", path, fullPath, err)
+	}
+	size := uint64(fi.Size())
+	if err := os.Remove(fullPath); err != nil {
+		return 0, fmt.Errorf("cannot remove %q: %s", fullPath, err)
+	}
+	return size, nil
+}
+
+// RemoveEmptyDirs recursively removes all the empty directories in fs.
+func (fs *FS) RemoveEmptyDirs() error {
+	return fscommon.RemoveEmptyDirs(fs.Dir)
+}
+
+func (fs *FS) mkdirAll(filePath string) error {
+	dir := filepath.Dir(filePath)
+	if err := os.MkdirAll(dir, 0700); err != nil {
+		return fmt.Errorf("cannot create directory %q: %s", dir, err)
+	}
+	return nil
+}
+
+func (fs *FS) path(p common.Part) string {
+	dir := fs.Dir
+	for strings.HasSuffix(dir, "/") {
+		dir = dir[:len(dir)-1]
+	}
+	return fs.Dir + "/" + p.Path
+}
+
+type limitedReadCloser struct {
+	r *filestream.Reader
+	n uint64
+}
+
+func (lrc *limitedReadCloser) Read(p []byte) (int, error) {
+	if lrc.n == 0 {
+		return 0, io.EOF
+	}
+	if uint64(len(p)) > lrc.n {
+		p = p[:lrc.n]
+	}
+	n, err := lrc.r.Read(p)
+	if n > len(p) {
+		return n, fmt.Errorf("too much data read; got %d bytes; want %d bytes", n, len(p))
+	}
+	lrc.n -= uint64(n)
+	return n, err
+}
+
+func (lrc *limitedReadCloser) Close() error {
+	lrc.r.MustClose()
+	return nil
+}
+
+type writeCloser struct {
+	w    *filestream.Writer
+	n    uint64
+	path string
+}
+
+func (wc *writeCloser) Write(p []byte) (int, error) {
+	n, err := wc.w.Write(p)
+	if uint64(n) > wc.n {
+		return n, fmt.Errorf("too much data written; got %d bytes; want %d bytes", n, wc.n)
+	}
+	wc.n -= uint64(n)
+	return n, err
+}
+
+func (wc *writeCloser) Close() error {
+	wc.w.MustClose()
+	if wc.n != 0 {
+		return fmt.Errorf("missing data writes for %d bytes", wc.n)
+	}
+	return fscommon.FsyncFile(wc.path)
+}

lib/backup/fsnil/fsnil.go

@@ -0,0 +1,20 @@
+package fsnil
+
+import (
+	"fmt"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+)
+
+// FS represents nil remote filesystem.
+type FS struct{}
+
+// String returns human-readable string representation for fs.
+func (fs *FS) String() string {
+	return fmt.Sprintf("fsnil")
+}
+
+// ListParts returns all the parts from fs.
+func (fs *FS) ListParts() ([]common.Part, error) {
+	return nil, nil
+}

lib/backup/fsremote/fsremote.go

@@ -0,0 +1,190 @@
+package fsremote
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/fscommon"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+)
+
+// FS represents remote filesystem.
+//
+// Backups are uploaded there.
+// Data is downloaded from there during restore.
+type FS struct {
+	// Dir is a path to remote directory with backup data.
+	Dir string
+}
+
+// String returns human-readable string representation for fs.
+func (fs *FS) String() string {
+	return fmt.Sprintf("fsremote %q", fs.Dir)
+}
+
+// ListParts returns all the parts from fs.
+func (fs *FS) ListParts() ([]common.Part, error) {
+	dir := fs.Dir
+	if _, err := os.Stat(dir); err != nil {
+		if os.IsNotExist(err) {
+			// Return empty part list for non-existing directory.
+			// The directory will be created later.
+			return nil, nil
+		}
+		return nil, err
+	}
+	files, err := fscommon.AppendFiles(nil, dir)
+	if err != nil {
+		return nil, err
+	}
+
+	var parts []common.Part
+	dir += "/"
+	for _, file := range files {
+		if !strings.HasPrefix(file, dir) {
+			logger.Panicf("BUG: unexpected prefix for file %q; want %q", file, dir)
+		}
+		var p common.Part
+		if !p.ParseFromRemotePath(file[len(dir):]) {
+			logger.Infof("skipping unknown file %s", file)
+			continue
+		}
+		// Check for correct part size.
+		fi, err := os.Stat(file)
+		if err != nil {
+			return nil, fmt.Errorf("cannot stat file %q for part %q: %s", file, p.Path, err)
+		}
+		p.ActualSize = uint64(fi.Size())
+		parts = append(parts, p)
+	}
+	return parts, nil
+}
+
+// DeletePart deletes the given part p from fs.
+func (fs *FS) DeletePart(p common.Part) error {
+	path := fs.path(p)
+	if err := os.Remove(path); err != nil {
+		return fmt.Errorf("cannot remove %q: %s", path, err)
+	}
+	return nil
+}
+
+// RemoveEmptyDirs recursively removes all the empty directories in fs.
+func (fs *FS) RemoveEmptyDirs() error {
+	return fscommon.RemoveEmptyDirs(fs.Dir)
+}
+
+// CopyPart copies the part p from srcFS to fs.
+//
+// srcFS must have *FS type.
+func (fs *FS) CopyPart(srcFS common.OriginFS, p common.Part) error {
+	src, ok := srcFS.(*FS)
+	if !ok {
+		return fmt.Errorf("cannot perform server-side copying from %s to %s: both of them must be fsremote", srcFS, fs)
+	}
+	srcPath := src.path(p)
+	dstPath := fs.path(p)
+	if err := fs.mkdirAll(dstPath); err != nil {
+		return err
+	}
+	// Attempt to create hardlink from srcPath to dstPath.
+	if err := os.Link(srcPath, dstPath); err == nil {
+		return fscommon.FsyncFile(dstPath)
+	}
+
+	// Cannot create hardlink. Just copy file contents
+	srcFile, err := os.Open(srcPath)
+	if err != nil {
+		return fmt.Errorf("cannot open file %q: %s", srcPath, err)
+	}
+	dstFile, err := os.Create(dstPath)
+	if err != nil {
+		_ = srcFile.Close()
+		return fmt.Errorf("cannot create file %q: %s", dstPath, err)
+	}
+	n, err := io.Copy(dstFile, srcFile)
+	if err1 := dstFile.Close(); err1 != nil {
+		err = err1
+	}
+	if err1 := srcFile.Close(); err1 != nil {
+		err = err1
+	}
+	if err != nil {
+		_ = os.RemoveAll(dstPath)
+		return err
+	}
+	if uint64(n) != p.Size {
+		_ = os.RemoveAll(dstPath)
+		return fmt.Errorf("unexpected number of bytes copied from %q to %q; got %d bytes; want %d bytes", srcPath, dstPath, n, p.Size)
+	}
+	if err := fscommon.FsyncFile(dstPath); err != nil {
+		_ = os.RemoveAll(dstPath)
+		return err
+	}
+	return nil
+}
+
+// DownloadPart download part p from fs to w.
+func (fs *FS) DownloadPart(p common.Part, w io.Writer) error {
+	path := fs.path(p)
+	r, err := os.Open(path)
+	if err != nil {
+		return fmt.Errorf("cannot open %q: %s", path, err)
+	}
+	n, err := io.Copy(w, r)
+	if err1 := r.Close(); err1 != nil && err == nil {
+		err = err1
+	}
+	if err != nil {
+		return fmt.Errorf("cannot download data from %q: %s", path, err)
+	}
+	if uint64(n) != p.Size {
+		return fmt.Errorf("wrong data size downloaded from %q; got %d bytes; want %d bytes", path, n, p.Size)
+	}
+	return nil
+}
+
+// UploadPart uploads p from r to fs.
+func (fs *FS) UploadPart(p common.Part, r io.Reader) error {
+	path := fs.path(p)
+	if err := fs.mkdirAll(path); err != nil {
+		return err
+	}
+	w, err := os.Create(path)
+	if err != nil {
+		return fmt.Errorf("cannot create file %q: %s", path, err)
+	}
+	n, err := io.Copy(w, r)
+	if err1 := w.Close(); err1 != nil && err == nil {
+		err = err1
+	}
+	if err != nil {
+		_ = os.RemoveAll(path)
+		return fmt.Errorf("cannot upload data to %q: %s", path, err)
+	}
+	if uint64(n) != p.Size {
+		_ = os.RemoveAll(path)
+		return fmt.Errorf("wrong data size uploaded to %q; got %d bytes; want %d bytes", path, n, p.Size)
+	}
+	if err := fscommon.FsyncFile(path); err != nil {
+		_ = os.RemoveAll(path)
+		return err
+	}
+	return nil
+}
+
+func (fs *FS) mkdirAll(filePath string) error {
+	dir := filepath.Dir(filePath)
+	if err := os.MkdirAll(dir, 0700); err != nil {
+		return fmt.Errorf("cannot create directory %q: %s", dir, err)
+	}
+	return nil
+}
+
+func (fs *FS) path(p common.Part) string {
+	return p.RemotePath(fs.Dir)
+}

lib/backup/gcsremote/gcs.go

@@ -0,0 +1,189 @@
+package gcsremote
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"strings"
+
+	"cloud.google.com/go/storage"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"google.golang.org/api/iterator"
+	"google.golang.org/api/option"
+)
+
+// FS represents filesystem for backups in GCS.
+//
+// Init must be called before calling other FS methods.
+type FS struct {
+	// Path to GCP credentials file.
+	//
+	// Default credentials are used if empty.
+	CredsFilePath string
+
+	// GCS bucket to use.
+	Bucket string
+
+	// Directory in the bucket to write to.
+	Dir string
+
+	bkt *storage.BucketHandle
+}
+
+// Init initializes fs.
+func (fs *FS) Init() error {
+	if fs.bkt != nil {
+		logger.Panicf("BUG: fs.Init has been already called")
+	}
+	for strings.HasPrefix(fs.Dir, "/") {
+		fs.Dir = fs.Dir[1:]
+	}
+	if !strings.HasSuffix(fs.Dir, "/") {
+		fs.Dir += "/"
+	}
+	ctx := context.Background()
+	var client *storage.Client
+	if len(fs.CredsFilePath) > 0 {
+		creds := option.WithCredentialsFile(fs.CredsFilePath)
+		c, err := storage.NewClient(ctx, creds)
+		if err != nil {
+			return fmt.Errorf("cannot create gcs client with credsFile %q: %s", fs.CredsFilePath, err)
+		}
+		client = c
+	} else {
+		c, err := storage.NewClient(ctx)
+		if err != nil {
+			return fmt.Errorf("cannot create default gcs client: %q", err)
+		}
+		client = c
+	}
+	fs.bkt = client.Bucket(fs.Bucket)
+	return nil
+}
+
+// String returns human-readable description for fs.
+func (fs *FS) String() string {
+	return fmt.Sprintf("GCS{bucket: %q, dir: %q}", fs.Bucket, fs.Dir)
+}
+
+// selectAttrs contains object attributes to select in ListParts.
+var selectAttrs = []string{
+	"Name",
+	"Size",
+}
+
+// ListParts returns all the parts for fs.
+func (fs *FS) ListParts() ([]common.Part, error) {
+	dir := fs.Dir
+	ctx := context.Background()
+	q := &storage.Query{
+		Prefix: dir,
+	}
+	if err := q.SetAttrSelection(selectAttrs); err != nil {
+		return nil, fmt.Errorf("error in SetAttrSelection: %s", err)
+	}
+	it := fs.bkt.Objects(ctx, q)
+	var parts []common.Part
+	for {
+		attr, err := it.Next()
+		if err == iterator.Done {
+			return parts, nil
+		}
+		if err != nil {
+			return nil, fmt.Errorf("error when iterating objects at %q: %s", dir, err)
+		}
+		file := attr.Name
+		if !strings.HasPrefix(file, dir) {
+			return nil, fmt.Errorf("unexpected prefix for gcs key %q; want %q", file, dir)
+		}
+		var p common.Part
+		if !p.ParseFromRemotePath(file[len(dir):]) {
+			logger.Infof("skipping unknown object %q", file)
+			continue
+		}
+		p.ActualSize = uint64(attr.Size)
+		parts = append(parts, p)
+	}
+}
+
+// DeletePart deletes part p from fs.
+func (fs *FS) DeletePart(p common.Part) error {
+	o := fs.object(p)
+	ctx := context.Background()
+	if err := o.Delete(ctx); err != nil {
+		return fmt.Errorf("cannot delete %q at %s (remote path %q): %s", p.Path, fs, o.ObjectName(), err)
+	}
+	return nil
+}
+
+// RemoveEmptyDirs recursively removes empty dirs in fs.
+func (fs *FS) RemoveEmptyDirs() error {
+	// GCS has no directories, so nothing to remove.
+	return nil
+}
+
+// CopyPart copies p from srcFS to fs.
+func (fs *FS) CopyPart(srcFS common.OriginFS, p common.Part) error {
+	src, ok := srcFS.(*FS)
+	if !ok {
+		return fmt.Errorf("cannot perform server-side copying from %s to %s: both of them must be GCS", srcFS, fs)
+	}
+	srcObj := src.object(p)
+	dstObj := fs.object(p)
+
+	copier := dstObj.CopierFrom(srcObj)
+	ctx := context.Background()
+	attr, err := copier.Run(ctx)
+	if err != nil {
+		return fmt.Errorf("cannot copy %q from %s to %s: %s", p.Path, src, fs, err)
+	}
+	if uint64(attr.Size) != p.Size {
+		return fmt.Errorf("unexpected %q size after copying from %s to %s; got %d bytes; want %d bytes", p.Path, src, fs, attr.Size, p.Size)
+	}
+	return nil
+}
+
+// DownloadPart downloads part p from fs to w.
+func (fs *FS) DownloadPart(p common.Part, w io.Writer) error {
+	o := fs.object(p)
+	ctx := context.Background()
+	r, err := o.NewReader(ctx)
+	if err != nil {
+		return fmt.Errorf("cannot open reader for %q at %s (remote path %q): %s", p.Path, fs, o.ObjectName(), err)
+	}
+	n, err := io.Copy(w, r)
+	if err1 := r.Close(); err1 != nil && err == nil {
+		err = err1
+	}
+	if err != nil {
+		return fmt.Errorf("cannot download %q from at %s (remote path %q): %s", p.Path, fs, o.ObjectName(), err)
+	}
+	if uint64(n) != p.Size {
+		return fmt.Errorf("wrong data size downloaded from %q at %s; got %d bytes; want %d bytes", p.Path, fs, n, p.Size)
+	}
+	return nil
+}
+
+// UploadPart uploads part p from r to fs.
+func (fs *FS) UploadPart(p common.Part, r io.Reader) error {
+	o := fs.object(p)
+	ctx := context.Background()
+	w := o.NewWriter(ctx)
+	n, err := io.Copy(w, r)
+	if err1 := w.Close(); err1 != nil && err == nil {
+		err = err1
+	}
+	if err != nil {
+		return fmt.Errorf("cannot upload data to %q at %s (remote path %q): %s", p.Path, fs, o.ObjectName(), err)
+	}
+	if uint64(n) != p.Size {
+		return fmt.Errorf("wrong data size uploaded to %q at %s; got %d bytes; want %d bytes", p.Path, fs, n, p.Size)
+	}
+	return nil
+}
+
+func (fs *FS) object(p common.Part) *storage.ObjectHandle {
+	path := p.RemotePath(fs.Dir)
+	return fs.bkt.Object(path)
+}

lib/backup/s3remote/s3.go

@@ -0,0 +1,236 @@
+package s3remote
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"strings"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/aws/session"
+	"github.com/aws/aws-sdk-go/service/s3"
+	"github.com/aws/aws-sdk-go/service/s3/s3manager"
+)
+
+// FS represents filesystem for backups in S3.
+//
+// Init must be called before calling other FS methods.
+type FS struct {
+	// Path to S3 credentials file.
+	CredsFilePath string
+
+	// Path to S3 configs file.
+	ConfigFilePath string
+
+	// GCS bucket to use.
+	Bucket string
+
+	// Directory in the bucket to write to.
+	Dir string
+
+	// Set for using S3-compatible enpoint such as MinIO etc.
+	CustomEndpoint string
+
+	// The name of S3 config profile to use.
+	ProfileName string
+
+	s3       *s3.S3
+	uploader *s3manager.Uploader
+}
+
+// Init initializes fs.
+func (fs *FS) Init() error {
+	if fs.s3 != nil {
+		logger.Panicf("BUG: Init is already called")
+	}
+	for strings.HasPrefix(fs.Dir, "/") {
+		fs.Dir = fs.Dir[1:]
+	}
+	if !strings.HasSuffix(fs.Dir, "/") {
+		fs.Dir += "/"
+	}
+	opts := session.Options{
+		SharedConfigState: session.SharedConfigEnable,
+		Profile:           fs.ProfileName,
+	}
+	if len(fs.CredsFilePath) > 0 {
+		opts.SharedConfigFiles = []string{
+			fs.ConfigFilePath,
+			fs.CredsFilePath,
+		}
+	}
+	sess, err := session.NewSessionWithOptions(opts)
+	if err != nil {
+		return fmt.Errorf("cannot create S3 session: %s", err)
+	}
+
+	if len(fs.CustomEndpoint) > 0 {
+		// Use provided custom endpoint for S3
+		logger.Infof("Using provided custom S3 endpoint: %q", fs.CustomEndpoint)
+		sess.Config.WithEndpoint(fs.CustomEndpoint)
+
+		// Disable prefixing endpoint with bucket name
+		sess.Config.WithS3ForcePathStyle(true)
+	} else {
+		// Determine bucket region.
+		ctx := context.Background()
+		region, err := s3manager.GetBucketRegion(ctx, sess, fs.Bucket, "")
+		if err != nil {
+			return fmt.Errorf("cannot determine region for bucket %q: %s", fs.Bucket, err)
+		}
+		sess.Config.WithRegion(region)
+		logger.Infof("bucket %q is stored at region %q; switching to this region", fs.Bucket, region)
+	}
+
+	fs.s3 = s3.New(sess)
+	fs.uploader = s3manager.NewUploader(sess, func(u *s3manager.Uploader) {
+		// We manage upload concurrency by ourselves.
+		u.Concurrency = 1
+	})
+	return nil
+}
+
+// String returns human-readable description for fs.
+func (fs *FS) String() string {
+	return fmt.Sprintf("S3{bucket: %q, dir: %q}", fs.Bucket, fs.Dir)
+}
+
+// ListParts returns all the parts for fs.
+func (fs *FS) ListParts() ([]common.Part, error) {
+	dir := fs.Dir
+	input := &s3.ListObjectsV2Input{
+		Bucket: aws.String(fs.Bucket),
+		Prefix: aws.String(dir),
+	}
+	var errOuter error
+	var parts []common.Part
+	err := fs.s3.ListObjectsV2Pages(input, func(page *s3.ListObjectsV2Output, lastPage bool) bool {
+		for _, o := range page.Contents {
+			file := *o.Key
+			if !strings.HasPrefix(file, dir) {
+				errOuter = fmt.Errorf("unexpected prefix for s3 key %q; want %q", file, dir)
+				return false
+			}
+			var p common.Part
+			if !p.ParseFromRemotePath(file[len(dir):]) {
+				logger.Infof("skipping unknown object %q", file)
+				continue
+			}
+			p.ActualSize = uint64(*o.Size)
+			parts = append(parts, p)
+		}
+		return !lastPage
+	})
+	if errOuter != nil && err == nil {
+		err = errOuter
+	}
+	if err != nil {
+		return nil, fmt.Errorf("error when listing s3 objects inside dir %q: %s", dir, err)
+	}
+	return parts, nil
+}
+
+// DeletePart deletes part p from fs.
+func (fs *FS) DeletePart(p common.Part) error {
+	path := fs.path(p)
+	input := &s3.DeleteObjectInput{
+		Bucket: aws.String(fs.Bucket),
+		Key:    aws.String(path),
+	}
+	_, err := fs.s3.DeleteObject(input)
+	if err != nil {
+		return fmt.Errorf("cannot delete %q at %s (remote path %q): %s", p.Path, fs, path, err)
+	}
+	return nil
+}
+
+// RemoveEmptyDirs recursively removes empty dirs in fs.
+func (fs *FS) RemoveEmptyDirs() error {
+	// S3 has no directories, so nothing to remove.
+	return nil
+}
+
+// CopyPart copies p from srcFS to fs.
+func (fs *FS) CopyPart(srcFS common.OriginFS, p common.Part) error {
+	src, ok := srcFS.(*FS)
+	if !ok {
+		return fmt.Errorf("cannot perform server-side copying from %s to %s: both of them must be S3", srcFS, fs)
+	}
+	srcPath := src.path(p)
+	dstPath := fs.path(p)
+	copySource := fmt.Sprintf("/%s/%s", src.Bucket, srcPath)
+
+	input := &s3.CopyObjectInput{
+		Bucket:     aws.String(fs.Bucket),
+		CopySource: aws.String(copySource),
+		Key:        aws.String(dstPath),
+	}
+	_, err := fs.s3.CopyObject(input)
+	if err != nil {
+		return fmt.Errorf("cannot copy %q from %s to %s (copySource %q): %s", p.Path, src, fs, copySource, err)
+	}
+	return nil
+}
+
+// DownloadPart downloads part p from fs to w.
+func (fs *FS) DownloadPart(p common.Part, w io.Writer) error {
+	path := fs.path(p)
+	input := &s3.GetObjectInput{
+		Bucket: aws.String(fs.Bucket),
+		Key:    aws.String(path),
+	}
+	o, err := fs.s3.GetObject(input)
+	if err != nil {
+		return fmt.Errorf("cannot open %q at %s (remote path %q): %s", p.Path, fs, path, err)
+	}
+	r := o.Body
+	n, err := io.Copy(w, r)
+	if err1 := r.Close(); err1 != nil && err == nil {
+		err = err1
+	}
+	if err != nil {
+		return fmt.Errorf("cannot download %q from at %s (remote path %q): %s", p.Path, fs, path, err)
+	}
+	if uint64(n) != p.Size {
+		return fmt.Errorf("wrong data size downloaded from %q at %s; got %d bytes; want %d bytes", p.Path, fs, n, p.Size)
+	}
+	return nil
+}
+
+// UploadPart uploads part p from r to fs.
+func (fs *FS) UploadPart(p common.Part, r io.Reader) error {
+	path := fs.path(p)
+	sr := &statReader{
+		r: r,
+	}
+	input := &s3manager.UploadInput{
+		Bucket: aws.String(fs.Bucket),
+		Key:    aws.String(path),
+		Body:   sr,
+	}
+	_, err := fs.uploader.Upload(input)
+	if err != nil {
+		return fmt.Errorf("cannot upoad data to %q at %s (remote path %q): %s", p.Path, fs, path, err)
+	}
+	if uint64(sr.size) != p.Size {
+		return fmt.Errorf("wrong data size uploaded to %q at %s; got %d bytes; want %d bytes", p.Path, fs, sr.size, p.Size)
+	}
+	return nil
+}
+
+func (fs *FS) path(p common.Part) string {
+	return p.RemotePath(fs.Dir)
+}
+
+type statReader struct {
+	r    io.Reader
+	size int64
+}
+
+func (sr *statReader) Read(p []byte) (int, error) {
+	n, err := sr.r.Read(p)
+	sr.size += int64(n)
+	return n, err
+}