lib/httpserver: do not flush and do not close gzip writer if response compression is disabled

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/535

.github/workflows/main.yml

@@ -24,7 +24,7 @@ jobs:
         run: |
           go get -u golang.org/x/lint/golint
           go get -u github.com/kisielk/errcheck
-          go get -u github.com/golangci/golangci-lint/cmd/golangci-lint
+          curl -sSfL https://raw.githubusercontent.com/golangci/golangci-lint/master/install.sh | sh -s -- -b $(go env GOPATH)/bin v1.27.0
       - name: Code checkout
         uses: actions/checkout@master
       - name: Build

Makefile

@@ -145,3 +145,11 @@ golangci-lint: install-golangci-lint
 
 install-golangci-lint:
 	which golangci-lint || GO111MODULE=off go get -u github.com/golangci/golangci-lint/cmd/golangci-lint
+
+docs-sync:
+	cp app/vmagent/README.md docs/vmagent.md
+	cp app/vmalert/README.md docs/vmalert.md
+	cp app/vmauth/README.md docs/vmauth.md
+	cp app/vmbackup/README.md docs/vmbackup.md
+	cp app/vmrestore/README.md docs/vmrestore.md
+	cp README.md docs/Single-server-VictoriaMetrics.md

README.md

@@ -10,18 +10,25 @@
 
 ## VictoriaMetrics
 
-VictoriaMetrics is fast, cost-effective and scalable time-series database. It can be used as long-term remote storage for Prometheus.
+VictoriaMetrics is fast, cost-effective and scalable time-series database.
+
 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). Just download VictoriaMetrics and see [how to start it](#how-to-start-victoriametrics).
 
 Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
 
+See our [Wiki](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki) for additional documentation.
+
+[Contact us](mailto:info@victoriametrics.com) if you need paid enterprise support for VictoriaMetrics.
+
+
 ## Case studies and talks
 
 * [Adidas](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#adidas)
 * [CERN](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#cern)
 * [COLOPL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#colopl)
+* [Zerodha](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#zerodha)
 * [Wix.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wixcom)
 * [Wedos.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wedoscom)
 * [Synthesio](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#synthesio)
@@ -34,6 +41,8 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 
 ## Prominent features
 
+* VictoriaMetrics can be used as long-term storage for Prometheus or for [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md).
+  See [these docs](#prometheus-setup) for details.
 * Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
   VictoriaMetrics implements [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/MetricsQL) query language, which is inspired by PromQL.
 * Supports global query view. Multiple Prometheus instances may write data into VictoriaMetrics. Later this data may be used in a single query.
@@ -116,6 +125,8 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 * [Monitoring](#monitoring)
 * [Troubleshooting](#troubleshooting)
 * [Backfilling](#backfilling)
+* [Replication](#replication)
+* [Backups](#backups)
 * [Profiling](#profiling)
 * [Integrations](#integrations)
 * [Third-party contributions](#third-party-contributions)
@@ -571,11 +582,11 @@ Run `make package-victoria-metrics`. It builds `victoriametrics/victoria-metrics
 `<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`.
 
-By default the image is built on top of `scratch` image. It is possible to build the package on top of any other base image
-by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds the image on top of `alpine:3.11` image:
+By default the image is built on top of `alpine` image for improved debuggability. It is possible to build the package on top of any other base image
+by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds the image on top of `scratch` image:
 
 ```bash
-ROOT_IMAGE=alpine:3.11 make package-victoria-metrics
+ROOT_IMAGE=scratch make package-victoria-metrics
 ```
 
 ### Start with docker-compose
@@ -763,7 +774,13 @@ The required resources for query path:
 ### 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:
+2) Pass addresses of these instances to [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) via `-remoteWrite.url` command-line flag:
+
+```bash
+/path/to/vmagent -remoteWrite.url=http://<victoriametrics-addr-1>:8428/api/v1/write -remoteWrite.url=http://<victoriametrics-addr-2>:8428/api/v1/write
+```
+
+Alternatively these addresses may be passed to `remote_write` section in Prometheus config:
 
 ```yml
 remote_write:
@@ -782,6 +799,8 @@ remote_write:
 kill -HUP `pidof prometheus`
 ```
 
+It is recommended to use [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) instead of Prometheus for highly loaded setups.
+
 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.
@@ -792,6 +811,7 @@ to write data to `victoriametrics-addr-1`, while each `r2` should write data to
 Another option is to write data simultaneously from Prometheus HA pair to a pair of VictoriaMetrics instances
 with the enabled de-duplication. See [this section](#deduplication) for details.
 
+
 ### Deduplication
 
 VictoriaMetrics de-duplicates data points if `-dedup.minScrapeInterval` command-line flag
@@ -809,6 +829,8 @@ Data is split in per-month subdirectories inside `<-storageDataPath>/data/small`
 Directories for months outside the configured retention are deleted on the first day of new month.
 In order to keep data according to `-retentionPeriod` max disk space usage is going to be `-retentionPeriod` + 1 month.
 For example if `-retentionPeriod` is set to 1, data for January is deleted on March 1st.
+It is safe to extend `-retentionPeriod` on existing data. If `-retentionPeriod` is set to lower
+value than before then data outside the configured period will be eventually deleted.
 
 ### Multiple retentions
 
@@ -874,6 +896,10 @@ Consider setting the following command-line flags:
 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`.
 
+Prefer authorizing all the incoming requests from untrusted networks with [vmauth](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmauth/README.md)
+or similar auth proxy.
+
+
 ### Tuning
 
 * There is no need for VictoriaMetrics tuning since it uses reasonable defaults for command-line flags,
@@ -910,6 +936,12 @@ The most interesting metrics are:
 * `sum(rate(vm_rows_inserted_total[5m]))` - ingestion rate, i.e. how many samples are inserted int the database per second.
 * `vm_free_disk_space_bytes` - free space left at `-storageDataPath`.
 * `sum(vm_data_size_bytes)` - the total size of data on disk.
+* `increase(vm_slow_row_inserts_total[5m])` - the number of slow inserts during the last 5 minutes.
+  If this number remains high during extended periods of time, then it is likely more RAM is needed for optimal handling
+  of the current number of active time series.
+* `increase(vm_slow_metric_name_loads_total[5m])` - the number of slow loads of metric names during the last 5 minutes.
+  If this number remains high during extended periods of time, then it is likely more RAM is needed for optimal handling
+  of the current number of active time series.
 
 
 ### Troubleshooting
@@ -922,8 +954,9 @@ The most interesting metrics are:
 
 * 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.
+  VictoriaMetrics [exposes](#monitoring) `vm_slow_*` metrics, which could be used as an indicator of low amounts of RAM.
   It is recommended increasing the amount of RAM on the node with VictoriaMetrics in order to improve
-  ingestion performance.
+  ingestion and query performance in this case.
   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.
 
@@ -970,6 +1003,24 @@ the query cache, which could contain incomplete data cached during the backfilli
 Yet another solution is to increase `-search.cacheTimestampOffset` flag value in order to disable caching
 for data with timestamps close to the current time.
 
+
+### Replication
+
+Single-node VictoriaMetrics doesn't support application-level replication. Use cluster version instead.
+See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#replication-and-data-safety) for details.
+
+Storage-level replication may be offloaded to durable persistent storage such as [Google Cloud disks](https://cloud.google.com/compute/docs/disks#pdspecs).
+
+See also [high availability docs](#high-availability) and [backup docs](#backups).
+
+
+### Backups
+
+VictoriaMetrics supports backups via [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md)
+and [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md) tools.
+We also provide provide `vmbackuper` tool for paid enterprise subscribers - see [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/466) for details.
+
+
 ### Profiling
 
 VictoriaMetrics provides handlers for collecting the following [Go profiles](https://blog.golang.org/profiling-go-programs):

app/victoria-metrics/main.go

@@ -3,6 +3,7 @@ package main
 import (
 	"flag"
 	"net/http"
+	"os"
 	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vminsert"
@@ -25,6 +26,8 @@ var (
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
 	envflag.Parse()
 	buildinfo.Init()
 	logger.Init()

app/vmagent/Makefile

@@ -52,8 +52,9 @@ publish-vmagent:
 	APP_NAME=vmagent $(MAKE) publish-via-docker
 
 run-vmagent:
-	mkdir -p vmagent-data
-	DOCKER_OPTS='-v $(shell pwd)/vmagent-data:/vmagent-data' \
+	mkdir -p vmagent-remotewrite-data
+	DOCKER_OPTS='-v $(shell pwd)/vmagent-remotewrite-data:/vmagent-remotewrite-data' \
+	ARGS='-remoteWrite.url=http://localhost:8428/api/v1/write' \
 	APP_NAME=vmagent \
 	$(MAKE) run-via-docker
 

app/vmagent/README.md

@@ -1,8 +1,8 @@
 ## vmagent
 
-`vmagent` is a tiny but brave agent, which helps you collecting metrics from various sources
-and storing them to [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics)
-or any other Prometheus-compatible storage system that supports `remote_write` protocol.
+`vmagent` is a tiny but brave agent, which helps you collect metrics from various sources
+and stores them in [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics)
+or any other Prometheus-compatible storage system that supports the `remote_write` protocol.
 
 <img alt="vmagent" src="vmagent.png">
 
@@ -11,7 +11,7 @@ or any other Prometheus-compatible storage system that supports `remote_write` p
 
 While VictoriaMetrics provides an efficient solution to store and observe metrics, our users needed something fast
 and RAM friendly to scrape metrics from Prometheus-compatible exporters to VictoriaMetrics.
-Also, we found that users’ infrastructure is like snowflakes - never alike, and we decided to add more flexibility
+Also, we found that users’ infrastructure are snowflakes - no two are alike, and we decided to add more flexibility
 to `vmagent` (like the ability to push metrics instead of pulling them). We did our best and plan to do even more.
 
 
@@ -31,7 +31,7 @@ to `vmagent` (like the ability to push metrics instead of pulling them). We did
 * Works in environments with unstable connections to remote storage. If the remote storage is unavailable, the collected metrics
   are buffered at `-remoteWrite.tmpDataPath`. The buffered metrics are sent to remote storage as soon as connection
   to remote storage is recovered. The maximum disk usage for the buffer can be limited with `-remoteWrite.maxDiskUsagePerURL`.
-* Uses lower amounts of RAM, CPU, disk IO and network bandwidth comparing to Prometheus.
+* Uses lower amounts of RAM, CPU, disk IO and network bandwidth compared to Prometheus.
 
 
 ### Quick Start
@@ -40,8 +40,7 @@ Just download `vmutils-*` archive from [releases page](https://github.com/Victor
 and pass the following flags to `vmagent` binary in order to start scraping Prometheus targets:
 
 * `-promscrape.config` with the path to Prometheus config file (it is usually located at `/etc/prometheus/prometheus.yml`)
-* `-remoteWrite.url` with the remote storage endpoint such as VictoriaMetrics. Multiple `-remoteWrite.url` args can be set in parallel
-  in order to replicate data concurrently to multiple remote storage systems.
+* `-remoteWrite.url` with the remote storage endpoint such as VictoriaMetrics. The `-remoteWrite.url` argument can be specified multiple times in order to replicate data concurrently to an arbitrary amount of remote storage systems.
 
 Example command line:
 
@@ -49,7 +48,7 @@ Example command line:
 /path/to/vmagent -promscrape.config=/path/to/prometheus.yml -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write
 ```
 
-If you need collecting only Influx data, then the following command line would be enough:
+If you only need to collect Influx data, then the following is sufficient:
 
 ```
 /path/to/vmagent -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write
@@ -79,14 +78,14 @@ See [the corresponding Makefile rules](https://github.com/VictoriaMetrics/Victor
 #### Drop-in replacement for Prometheus
 
 If you use Prometheus only for scraping metrics from various targets and forwarding these metrics to remote storage,
-then `vmagent` can replace such Prometheus setup. Usually `vmagent` requires lower amounts of RAM, CPU and network bandwidth comparing to Prometheus for such setup.
+then `vmagent` can replace such Prometheus setup. Usually `vmagent` requires lower amounts of RAM, CPU and network bandwidth comparing to Prometheus for such a setup.
 See [these docs](#how-to-collect-metrics-in-prometheus-format) for details.
 
 
 #### Replication and high availability
 
 `vmagent` replicates the collected metrics among multiple remote storage instances configured via `-remoteWrite.url` args.
-If a single remote storage instance temporarily goes out of service, then the collected data remains available in another remote storage instances.
+If a single remote storage instance temporarily is out of service, then the collected data remains available in another remote storage instances.
 `vmagent` buffers the collected data in files at `-remoteWrite.tmpDataPath` until the remote storage becomes available again.
 Then it sends the buffered data to the remote storage in order to prevent data gaps in the remote storage.
 
@@ -94,13 +93,13 @@ Then it sends the buffered data to the remote storage in order to prevent data g
 #### Relabeling and filtering
 
 `vmagent` can add, remove or update labels on the collected data before sending it to remote storage. Additionally,
-it can remove unneeded samples via Prometheus-like relabeling before sending the collected data to remote storage.
+it can remove unwanted samples via Prometheus-like relabeling before sending the collected data to remote storage.
 See [these docs](#relabeling) for details.
 
 
 #### Splitting data streams among multiple systems
 
-`vmagent` supports splitting of the collected data among muliple destinations with the help of `-remoteWrite.urlRelabelConfig`,
+`vmagent` supports splitting the collected data between muliple destinations with the help of `-remoteWrite.urlRelabelConfig`,
 which is applied independently for each configured `-remoteWrite.url` destination. For instance, it is possible to replicate or split
 data among long-term remote storage, short-term remote storage and real-time analytical system [built on top of Kafka](https://github.com/Telefonica/prometheus-kafka-adapter).
 Note that each destination can receive its own subset of the collected data thanks to per-destination relabeling via `-remoteWrite.urlRelabelConfig`.
@@ -148,6 +147,10 @@ The following scrape types in [scrape_config](https://prometheus.io/docs/prometh
 * `dns_sd_configs` - for scraping targets discovered from DNS records (SRV, A and AAAA).
   See [dns_sd_config](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#dns_sd_config) for details.
 
+Note that `vmagent` doesn't support `refresh_interval` option these scrape configs. Use the corresponding `-promscrape.*CheckInterval`
+command-line flag instead. For example, `-promscrape.consulSDCheckInterval=60s` sets `refresh_interval` for all the `consul_sd_configs`
+entries to 60s. Run `vmagent -help` in order to see default values for `-promscrape.*CheckInterval` flags.
+
 
 File feature requests at [our issue tracker](https://github.com/VictoriaMetrics/VictoriaMetrics/issues) if you need other service discovery mechanisms to be supported by `vmagent`.
 
@@ -200,12 +203,12 @@ either via `vmagent` itself or via Prometheus, so the exported metrics could be
 * When `vmagent` scrapes many unreliable targets, it can flood error log with scrape errors. These errors can be suppressed
   by passing `-promscrape.suppressScrapeErrors` command-line flag to `vmagent`. The most recent scrape error per each target can be observed at `http://vmagent-host:8429/targets`.
 
-* It is recommended increasing `-remoteWrite.queues` if `vmagent` collects more than 100K samples per second
+* It is recommended to increase `-remoteWrite.queues` if `vmagent` collects more than 100K samples per second
   and `vmagent_remotewrite_pending_data_bytes` metric exported at `http://vmagent-host:8429/metrics` page constantly grows.
 
 * `vmagent` buffers scraped data at `-remoteWrite.tmpDataPath` directory until it is sent to `-remoteWrite.url`.
-  The directory can grow big when remote storage is unavailable during extended periods of time and if `-remoteWrite.maxDiskUsagePerURL` isn't set.
-  If you don't want sending all the data from the directory to remote storage, just stop `vmagent` and delete the directory.
+  The directory can grow large when remote storage is unavailable for extended periods of time and if `-remoteWrite.maxDiskUsagePerURL` isn't set.
+  If you don't want to send all the data from the directory to remote storage, simply stop `vmagent` and delete the directory.
 
 
 ### How to build from sources

app/vmagent/main.go

@@ -4,6 +4,7 @@ import (
 	"flag"
 	"fmt"
 	"net/http"
+	"os"
 	"strings"
 	"time"
 
@@ -39,6 +40,8 @@ var (
 		"Telnet put messages and HTTP /api/put messages are simultaneously served on TCP port. "+
 		"Usually :4242 must be set. Doesn't work if empty")
 	opentsdbHTTPListenAddr = flag.String("opentsdbHTTPListenAddr", "", "TCP address to listen for OpentTSDB HTTP put requests. Usually :4242 must be set. Doesn't work if empty")
+	dryRun                 = flag.Bool("dryRun", false, "Whether to check only config files without running vmagent. The following files are checked: "+
+		"-promscrape.config, -remoteWrite.relabelConfig, -remoteWrite.urlRelabelConfig . See also -promscrape.config.dryRun")
 )
 
 var (
@@ -49,9 +52,27 @@ var (
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
+	flag.Usage = usage
 	envflag.Parse()
 	buildinfo.Init()
 	logger.Init()
+
+	if *dryRun {
+		if err := flag.Set("promscrape.config.strictParse", "true"); err != nil {
+			logger.Panicf("BUG: cannot set promscrape.config.strictParse=true: %s", err)
+		}
+		if err := remotewrite.CheckRelabelConfigs(); err != nil {
+			logger.Fatalf("error when checking relabel configs: %s", err)
+		}
+		if err := promscrape.CheckConfig(); err != nil {
+			logger.Fatalf("error when checking Prometheus config: %s", err)
+		}
+		logger.Infof("all the configs are ok; exitting with 0 status code")
+		return
+	}
+
 	logger.Infof("starting vmagent at %q...", *httpListenAddr)
 	startTime := time.Now()
 	remotewrite.Init()
@@ -185,3 +206,15 @@ var (
 
 	promscrapeConfigReloadRequests = metrics.NewCounter(`vmagent_http_requests_total{path="/-/reload"}`)
 )
+
+func usage() {
+	const s = `
+vmagent collects metrics data via popular data ingestion protocols and routes it to VictoriaMetrics.
+
+See the docs at https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md .
+`
+
+	f := flag.CommandLine.Output()
+	fmt.Fprintf(f, "%s\n", s)
+	flag.PrintDefaults()
+}

app/vmagent/remotewrite/client.go

@@ -69,7 +69,7 @@ func newClient(argIdx int, remoteWriteURL, urlLabelValue string, fq *persistentq
 	token := bearerToken.GetOptionalArg(argIdx)
 	if len(token) > 0 {
 		if authHeader != "" {
-			logger.Panicf("FATAL: `-remoteWrite.bearerToken`=%q cannot be set when `-remoteWrite.basicAuth.*` flags are set", token)
+			logger.Fatalf("`-remoteWrite.bearerToken`=%q cannot be set when `-remoteWrite.basicAuth.*` flags are set", token)
 		}
 		authHeader = "Bearer " + token
 	}
@@ -85,11 +85,11 @@ func newClient(argIdx int, remoteWriteURL, urlLabelValue string, fq *persistentq
 	switch scheme {
 	case "http", "https":
 	default:
-		logger.Panicf("FATAL: unsupported scheme in -remoteWrite.url=%q: %q. It must be http or https", remoteWriteURL, scheme)
+		logger.Fatalf("unsupported scheme in -remoteWrite.url=%q: %q. It must be http or https", remoteWriteURL, scheme)
 	}
 	host := string(u.Host())
 	if len(host) == 0 {
-		logger.Panicf("FATAL: invalid -remoteWrite.url=%q: host cannot be empty. Make sure the url looks like `http://host:port/path`", remoteWriteURL)
+		logger.Fatalf("invalid -remoteWrite.url=%q: host cannot be empty. Make sure the url looks like `http://host:port/path`", remoteWriteURL)
 	}
 	requestURI := string(u.RequestURI())
 	isTLS := scheme == "https"

app/vmagent/remotewrite/pendingseries.go

@@ -6,7 +6,7 @@ import (
 	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/persistentqueue"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
 	"github.com/VictoriaMetrics/metrics"
@@ -15,7 +15,8 @@ import (
 
 var (
 	flushInterval = flag.Duration("remoteWrite.flushInterval", time.Second, "Interval for flushing the data to remote storage. "+
-		"Higher value reduces network bandwidth usage at the cost of delayed push of scraped data to remote storage")
+		"Higher value reduces network bandwidth usage at the cost of delayed push of scraped data to remote storage. "+
+		"Minimum supported interval is 1 second")
 	maxUnpackedBlockSize = flag.Int("remoteWrite.maxBlockSize", 32*1024*1024, "The maximum size in bytes of unpacked request to send to remote storage. "+
 		"It shouldn't exceed -maxInsertRequestSize from VictoriaMetrics")
 )
@@ -55,6 +56,10 @@ func (ps *pendingSeries) Push(tss []prompbmarshal.TimeSeries) {
 }
 
 func (ps *pendingSeries) periodicFlusher() {
+	flushSeconds := int64(flushInterval.Seconds())
+	if flushSeconds <= 0 {
+		flushSeconds = 1
+	}
 	ticker := time.NewTicker(*flushInterval)
 	defer ticker.Stop()
 	mustStop := false
@@ -63,7 +68,7 @@ func (ps *pendingSeries) periodicFlusher() {
 		case <-ps.stopCh:
 			mustStop = true
 		case <-ticker.C:
-			if time.Since(ps.wr.lastFlushTime) < *flushInterval/2 {
+			if fasttime.UnixTimestamp()-ps.wr.lastFlushTime < uint64(flushSeconds) {
 				continue
 			}
 		}
@@ -76,7 +81,7 @@ func (ps *pendingSeries) periodicFlusher() {
 type writeRequest struct {
 	wr            prompbmarshal.WriteRequest
 	pushBlock     func(block []byte)
-	lastFlushTime time.Time
+	lastFlushTime uint64
 
 	tss []prompbmarshal.TimeSeries
 
@@ -108,7 +113,7 @@ func (wr *writeRequest) reset() {
 
 func (wr *writeRequest) flush() {
 	wr.wr.Timeseries = wr.tss
-	wr.lastFlushTime = time.Now()
+	wr.lastFlushTime = fasttime.UnixTimestamp()
 	pushWriteRequest(&wr.wr, wr.pushBlock)
 	wr.reset()
 }
@@ -144,13 +149,8 @@ func (wr *writeRequest) copyTimeSeries(dst, src *prompbmarshal.TimeSeries) {
 	}
 	dst.Labels = labelsDst[labelsLen:]
 
-	samplesDst = append(samplesDst, prompbmarshal.Sample{})
-	dstSample := &samplesDst[len(samplesDst)-1]
-	if len(src.Samples) != 1 {
-		logger.Panicf("BUG: unexpected number of samples in time series; got %d; want 1", len(src.Samples))
-	}
-	*dstSample = src.Samples[0]
-	dst.Samples = samplesDst[len(samplesDst)-1:]
+	samplesDst = append(samplesDst, src.Samples...)
+	dst.Samples = samplesDst[len(samplesDst)-len(src.Samples):]
 
 	wr.samples = samplesDst
 	wr.labels = labelsDst

app/vmagent/remotewrite/relabel.go

@@ -2,6 +2,7 @@ package remotewrite
 
 import (
 	"flag"
+	"fmt"
 	"strings"
 	"sync"
 
@@ -16,35 +17,60 @@ var (
 		"Pass multiple -remoteWrite.label flags in order to add multiple flags to metrics before sending them to remote storage")
 	relabelConfigPathGlobal = flag.String("remoteWrite.relabelConfig", "", "Optional path to file with relabel_config entries. These entries are applied to all the metrics "+
 		"before sending them to -remoteWrite.url. See https://prometheus.io/docs/prometheus/latest/configuration/configuration/#relabel_config for details")
+	relabelConfigPaths = flagutil.NewArray("remoteWrite.urlRelabelConfig", "Optional path to relabel config for the corresponding -remoteWrite.url")
 )
 
 var labelsGlobal []prompbmarshal.Label
-var prcsGlobal []promrelabel.ParsedRelabelConfig
 
-// initRelabelGlobal must be called after parsing command-line flags.
-func initRelabelGlobal() {
+// CheckRelabelConfigs checks -remoteWrite.relabelConfig and -remoteWrite.urlRelabelConfig.
+func CheckRelabelConfigs() error {
+	_, err := loadRelabelConfigs()
+	return err
+}
+
+func loadRelabelConfigs() (*relabelConfigs, error) {
+	var rcs relabelConfigs
+	if *relabelConfigPathGlobal != "" {
+		global, err := promrelabel.LoadRelabelConfigs(*relabelConfigPathGlobal)
+		if err != nil {
+			return nil, fmt.Errorf("cannot load -remoteWrite.relabelConfig=%q: %s", *relabelConfigPathGlobal, err)
+		}
+		rcs.global = global
+	}
+	if len(*relabelConfigPaths) > len(*remoteWriteURLs) {
+		return nil, fmt.Errorf("too many -remoteWrite.urlRelabelConfig args: %d; it mustn't exceed the number of -remoteWrite.url args: %d",
+			len(*relabelConfigPaths), len(*remoteWriteURLs))
+	}
+	rcs.perURL = make([][]promrelabel.ParsedRelabelConfig, len(*remoteWriteURLs))
+	for i, path := range *relabelConfigPaths {
+		prc, err := promrelabel.LoadRelabelConfigs(path)
+		if err != nil {
+			return nil, fmt.Errorf("cannot load relabel configs from -remoteWrite.urlRelabelConfig=%q: %s", path, err)
+		}
+		rcs.perURL[i] = prc
+	}
+	return &rcs, nil
+}
+
+type relabelConfigs struct {
+	global []promrelabel.ParsedRelabelConfig
+	perURL [][]promrelabel.ParsedRelabelConfig
+}
+
+// initLabelsGlobal must be called after parsing command-line flags.
+func initLabelsGlobal() {
 	// Init labelsGlobal
 	labelsGlobal = nil
 	for _, s := range *unparsedLabelsGlobal {
 		n := strings.IndexByte(s, '=')
 		if n < 0 {
-			logger.Panicf("FATAL: missing '=' in `-remoteWrite.label`. It must contain label in the form `name=value`; got %q", s)
+			logger.Fatalf("missing '=' in `-remoteWrite.label`. It must contain label in the form `name=value`; got %q", s)
 		}
 		labelsGlobal = append(labelsGlobal, prompbmarshal.Label{
 			Name:  s[:n],
 			Value: s[n+1:],
 		})
 	}
-
-	// Init prcsGlobal
-	prcsGlobal = nil
-	if len(*relabelConfigPathGlobal) > 0 {
-		var err error
-		prcsGlobal, err = promrelabel.LoadRelabelConfigs(*relabelConfigPathGlobal)
-		if err != nil {
-			logger.Panicf("FATAL: cannot load relabel configs from -remoteWrite.relabelConfig=%q: %s", *relabelConfigPathGlobal, err)
-		}
-	}
 }
 
 func (rctx *relabelCtx) applyRelabeling(tss []prompbmarshal.TimeSeries, extraLabels []prompbmarshal.Label, prcs []promrelabel.ParsedRelabelConfig) []prompbmarshal.TimeSeries {

app/vmagent/remotewrite/remotewrite.go

@@ -3,6 +3,7 @@ package remotewrite
 import (
 	"flag"
 	"fmt"
+	"sync"
 	"sync/atomic"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
@@ -10,8 +11,8 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/persistentqueue"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/procutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
 	"github.com/VictoriaMetrics/metrics"
 	xxhash "github.com/cespare/xxhash/v2"
 )
@@ -20,9 +21,8 @@ var (
 	remoteWriteURLs = flagutil.NewArray("remoteWrite.url", "Remote storage URL to write data to. It must support Prometheus remote_write API. "+
 		"It is recommended using VictoriaMetrics as remote storage. Example url: http://<victoriametrics-host>:8428/api/v1/write . "+
 		"Pass multiple -remoteWrite.url flags in order to write data concurrently to multiple remote storage systems")
-	relabelConfigPaths = flagutil.NewArray("remoteWrite.urlRelabelConfig", "Optional path to relabel config for the corresponding -remoteWrite.url")
-	tmpDataPath        = flag.String("remoteWrite.tmpDataPath", "vmagent-remotewrite-data", "Path to directory where temporary data for remote write component is stored")
-	queues             = flag.Int("remoteWrite.queues", 1, "The number of concurrent queues to each -remoteWrite.url. Set more queues if a single queue "+
+	tmpDataPath = flag.String("remoteWrite.tmpDataPath", "vmagent-remotewrite-data", "Path to directory where temporary data for remote write component is stored")
+	queues      = flag.Int("remoteWrite.queues", 1, "The number of concurrent queues to each -remoteWrite.url. Set more queues if a single queue "+
 		"isn't enough for sending high volume of collected data to remote storage")
 	showRemoteWriteURL = flag.Bool("remoteWrite.showURL", false, "Whether to show -remoteWrite.url in the exported metrics. "+
 		"It is hidden by default, since it can contain sensistive auth info")
@@ -34,6 +34,9 @@ var (
 
 var rwctxs []*remoteWriteCtx
 
+// Contains the current relabelConfigs.
+var allRelabelConfigs atomic.Value
+
 // Init initializes remotewrite.
 //
 // It must be called after flag.Parse().
@@ -41,14 +44,19 @@ var rwctxs []*remoteWriteCtx
 // Stop must be called for graceful shutdown.
 func Init() {
 	if len(*remoteWriteURLs) == 0 {
-		logger.Panicf("FATAL: at least one `-remoteWrite.url` must be set")
+		logger.Fatalf("at least one `-remoteWrite.url` must be set")
 	}
 
 	if !*showRemoteWriteURL {
 		// remoteWrite.url can contain authentication codes, so hide it at `/metrics` output.
 		httpserver.RegisterSecretFlag("remoteWrite.url")
 	}
-	initRelabelGlobal()
+	initLabelsGlobal()
+	rcs, err := loadRelabelConfigs()
+	if err != nil {
+		logger.Fatalf("cannot load relabel configs: %s", err)
+	}
+	allRelabelConfigs.Store(rcs)
 
 	maxInmemoryBlocks := memory.Allowed() / len(*remoteWriteURLs) / maxRowsPerBlock / 100
 	if maxInmemoryBlocks > 200 {
@@ -61,23 +69,47 @@ func Init() {
 		maxInmemoryBlocks = 2
 	}
 	for i, remoteWriteURL := range *remoteWriteURLs {
-		relabelConfigPath := ""
-		if i < len(*relabelConfigPaths) {
-			relabelConfigPath = (*relabelConfigPaths)[i]
-		}
 		urlLabelValue := fmt.Sprintf("secret-url-%d", i+1)
 		if *showRemoteWriteURL {
 			urlLabelValue = remoteWriteURL
 		}
-		rwctx := newRemoteWriteCtx(i, remoteWriteURL, relabelConfigPath, maxInmemoryBlocks, urlLabelValue)
+		rwctx := newRemoteWriteCtx(i, remoteWriteURL, maxInmemoryBlocks, urlLabelValue)
 		rwctxs = append(rwctxs, rwctx)
 	}
+
+	// Start config reloader.
+	sighupCh := procutil.NewSighupChan()
+	configReloaderWG.Add(1)
+	go func() {
+		defer configReloaderWG.Done()
+		for {
+			select {
+			case <-sighupCh:
+			case <-stopCh:
+				return
+			}
+			logger.Infof("SIGHUP received; reloading relabel configs pointed by -remoteWrite.relabelConfig and -remoteWrite.urlRelabelConfig")
+			rcs, err := loadRelabelConfigs()
+			if err != nil {
+				logger.Errorf("cannot reload relabel configs; preserving the previous configs; error: %s", err)
+				continue
+			}
+			allRelabelConfigs.Store(rcs)
+			logger.Infof("Successfully reloaded relabel configs")
+		}
+	}()
 }
 
+var stopCh = make(chan struct{})
+var configReloaderWG sync.WaitGroup
+
 // Stop stops remotewrite.
 //
 // It is expected that nobody calls Push during and after the call to this func.
 func Stop() {
+	close(stopCh)
+	configReloaderWG.Wait()
+
 	for _, rwctx := range rwctxs {
 		rwctx.MustStop()
 	}
@@ -86,11 +118,11 @@ func Stop() {
 
 // Push sends wr to remote storage systems set via `-remoteWrite.url`.
 //
-// Each timeseries in wr.Timeseries must contain one sample.
-//
 // Note that wr may be modified by Push due to relabeling.
 func Push(wr *prompbmarshal.WriteRequest) {
 	var rctx *relabelCtx
+	rcs := allRelabelConfigs.Load().(*relabelConfigs)
+	prcsGlobal := rcs.global
 	if len(prcsGlobal) > 0 || len(labelsGlobal) > 0 {
 		rctx = getRelabelCtx()
 	}
@@ -124,9 +156,9 @@ func Push(wr *prompbmarshal.WriteRequest) {
 var globalRelabelMetricsDropped = metrics.NewCounter("vmagent_remotewrite_global_relabel_metrics_dropped_total")
 
 type remoteWriteCtx struct {
+	idx        int
 	fq         *persistentqueue.FastQueue
 	c          *client
-	prcs       []promrelabel.ParsedRelabelConfig
 	pss        []*pendingSeries
 	pssNextIdx uint64
 
@@ -135,7 +167,7 @@ type remoteWriteCtx struct {
 	relabelMetricsDropped *metrics.Counter
 }
 
-func newRemoteWriteCtx(argIdx int, remoteWriteURL, relabelConfigPath string, maxInmemoryBlocks int, urlLabelValue string) *remoteWriteCtx {
+func newRemoteWriteCtx(argIdx int, remoteWriteURL string, maxInmemoryBlocks int, urlLabelValue string) *remoteWriteCtx {
 	h := xxhash.Sum64([]byte(remoteWriteURL))
 	path := fmt.Sprintf("%s/persistent-queue/%016X", *tmpDataPath, h)
 	fq := persistentqueue.MustOpenFastQueue(path, remoteWriteURL, maxInmemoryBlocks, *maxPendingBytesPerURL)
@@ -146,23 +178,15 @@ func newRemoteWriteCtx(argIdx int, remoteWriteURL, relabelConfigPath string, max
 		return float64(fq.GetInmemoryQueueLen())
 	})
 	c := newClient(argIdx, remoteWriteURL, urlLabelValue, fq, *queues)
-	var prcs []promrelabel.ParsedRelabelConfig
-	if len(relabelConfigPath) > 0 {
-		var err error
-		prcs, err = promrelabel.LoadRelabelConfigs(relabelConfigPath)
-		if err != nil {
-			logger.Panicf("FATAL: cannot load relabel configs from -remoteWrite.urlRelabelConfig=%q: %s", relabelConfigPath, err)
-		}
-	}
 	pss := make([]*pendingSeries, *queues)
 	for i := range pss {
 		pss[i] = newPendingSeries(fq.MustWriteBlock)
 	}
 	return &remoteWriteCtx{
-		fq:   fq,
-		c:    c,
-		prcs: prcs,
-		pss:  pss,
+		idx: argIdx,
+		fq:  fq,
+		c:   c,
+		pss: pss,
 
 		relabelMetricsDropped: metrics.GetOrCreateCounter(fmt.Sprintf(`vmagent_remotewrite_relabel_metrics_dropped_total{path=%q, url=%q}`, path, urlLabelValue)),
 	}
@@ -172,10 +196,10 @@ func (rwctx *remoteWriteCtx) MustStop() {
 	for _, ps := range rwctx.pss {
 		ps.MustStop()
 	}
+	rwctx.idx = 0
 	rwctx.pss = nil
 	rwctx.fq.MustClose()
 	rwctx.fq = nil
-	rwctx.prcs = nil
 	rwctx.c.MustStop()
 	rwctx.c = nil
 
@@ -184,7 +208,9 @@ func (rwctx *remoteWriteCtx) MustStop() {
 
 func (rwctx *remoteWriteCtx) Push(tss []prompbmarshal.TimeSeries) {
 	var rctx *relabelCtx
-	if len(rwctx.prcs) > 0 {
+	rcs := allRelabelConfigs.Load().(*relabelConfigs)
+	prcs := rcs.perURL[rwctx.idx]
+	if len(prcs) > 0 {
 		// Make a copy of tss before applying relabeling in order to prevent
 		// from affecting time series for other remoteWrite.url configs.
 		// See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/467 for details.
@@ -192,7 +218,7 @@ func (rwctx *remoteWriteCtx) Push(tss []prompbmarshal.TimeSeries) {
 		tss = rwctx.tss
 		rctx = getRelabelCtx()
 		tssLen := len(tss)
-		tss = rctx.applyRelabeling(tss, nil, rwctx.prcs)
+		tss = rctx.applyRelabeling(tss, nil, prcs)
 		rwctx.relabelMetricsDropped.Add(tssLen - len(tss))
 	}
 	pss := rwctx.pss

app/vmalert/Makefile

@@ -52,14 +52,17 @@ publish-vmalert:
 	APP_NAME=vmalert $(MAKE) publish-via-docker
 
 test-vmalert:
-	go test -race -cover ./app/vmalert
+	go test -v -race -cover ./app/vmalert -loggerLevel=ERROR
+	go test -v -race -cover ./app/vmalert/datasource
+	go test -v -race -cover ./app/vmalert/notifier
+	go test -v -race -cover ./app/vmalert/config
 
 run-vmalert: vmalert
-	./bin/vmalert -rule=app/vmalert/testdata/rules0-good.rules \
+	./bin/vmalert -rule=app/vmalert/config/testdata/rules2-good.rules \
 		-datasource.url=http://localhost:8428 \
 		-notifier.url=http://localhost:9093 \
-		-remotewrite.url=http://localhost:8428 \
-		-remoteread.url=http://localhost:8428 \
+		-remoteWrite.url=http://localhost:8428 \
+		-remoteRead.url=http://localhost:8428 \
 		-evaluationInterval=3s
 
 vmalert-amd64:

app/vmalert/README.md

@@ -1,20 +1,18 @@
-## VM Alert
+## vmalert
 
-`vmalert` executes a list of given MetricsQL expressions (rules) and
-sends alerts to [Alert Manager](https://github.com/prometheus/alertmanager).   
+`vmalert` executes a list of given [alerting](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/)
+or [recording](https://prometheus.io/docs/prometheus/latest/configuration/recording_rules/)
+rules against configured address.
 
 ### Features:
 * Integration with [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics) TSDB;
 * VictoriaMetrics [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/MetricsQL)
- expressions validation;
+ support and expressions validation;
 * Prometheus [alerting rules definition format](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/#defining-alerting-rules)
  support;
 * Integration with [Alertmanager](https://github.com/prometheus/alertmanager);
 * Lightweight without extra dependencies.
 
-### TODO:
-* Configuration hot reload.
-
 ### QuickStart
 
 To build `vmalert` from sources:
@@ -26,9 +24,9 @@ make vmalert
 The build binary will be placed to `VictoriaMetrics/bin` folder.
 
 To start using `vmalert` you will need the following things:
-* list of alert rules - PromQL/MetricsQL expressions to execute;
+* list of rules - PromQL/MetricsQL expressions to execute;
 * datasource address - reachable VictoriaMetrics instance for rules execution;
-* notifier address - reachable Alertmanager instance for processing, 
+* notifier address - reachable [Alert Manager](https://github.com/prometheus/alertmanager) instance for processing, 
 aggregating alerts and sending notifications.
 
 Then configure `vmalert` accordingly:
@@ -38,23 +36,28 @@ Then configure `vmalert` accordingly:
         -notifier.url=http://localhost:9093
 ```
 
-Example for `.rules` file may be found [here](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/testdata/rules0-good.rules)
+Example for `.rules` file may be found [here](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/testdata).
+
+`vmalert` may be configured with `-remoteWrite` flag to write recording rules and 
+alerts state in form of timeseries via remote write protocol. Alerts state will be written 
+as `ALERTS` timeseries. These timeseries may be used to recover alerts state on `vmalert` 
+restarts if `-remoteRead` is configured.
 
 `vmalert` runs evaluation for every group in a separate goroutine.
 Rules in group evaluated one-by-one sequentially. 
 
+**Important:** while recording rules execution is sequential, writing of timeseries results to remote
+storage is asynchronous. Hence, user shouldn't rely on recording rules chaining when result of previous
+recording rule is reused in next one.
+
 `vmalert` also runs a web-server (`-httpListenAddr`) for serving metrics and alerts endpoints:
+* `http://<vmalert-addr>/api/v1/groups` - list of all loaded groups and rules;
 * `http://<vmalert-addr>/api/v1/alerts` - list of all active alerts;
 * `http://<vmalert-addr>/api/v1/<groupName>/<alertID>/status" ` - get alert status by ID.
 Used as alert source in AlertManager.
 * `http://<vmalert-addr>/metrics` - application metrics.
 * `http://<vmalert-addr>/-/reload` - hot configuration reload.
 
-`vmalert` may be configured with `-remotewrite` flag to write alerts state in form of timeseries
-via remote write protocol. Alerts state will be written as `ALERTS` timeseries. These timeseries
-may be used to recover alerts state on `vmalert` restarts if `-remoteread` is configured.
-
-
 ### Configuration
 
 The shortlist of configuration flags is the following:
@@ -66,35 +69,35 @@ Usage of vmalert:
         Optional basic auth username for -datasource.url
   -datasource.url string
         Victoria Metrics or VMSelect url. Required parameter. E.g. http://127.0.0.1:8428
-  -enableTCP6
-        Whether to enable IPv6 for listening and dialing. By default only IPv4 TCP is used
   -evaluationInterval duration
-        How often to evaluate the rules. Default 1m (default 1m0s)
+        How often to evaluate the rules (default 1m0s)
   -external.url string
         External URL is used as alert's source for sent alerts to the notifier
-  -http.maxGracefulShutdownDuration duration
-        The maximum duration for graceful shutdown of HTTP server. Highly loaded server may require increased value for graceful shutdown (default 7s)
-  -httpAuth.password string
-        Password for HTTP Basic Auth. The authentication is disabled if -httpAuth.username is empty
-  -httpAuth.username string
-        Username for HTTP Basic Auth. The authentication is disabled if empty. See also -httpAuth.password
   -httpListenAddr string
         Address to listen for http connections (default ":8880")
+  -metricsAuthKey string
+        Auth key for /metrics. It overrides httpAuth settings
   -notifier.url string
         Prometheus alertmanager URL. Required parameter. e.g. http://127.0.0.1:9093
-  -remoteread.basicAuth.password string
-        Optional basic auth password for -remoteread.url
-  -remoteread.basicAuth.username string
-        Optional basic auth username for -remoteread.url
-  -remoteread.lookback duration
+  -remoteRead.basicAuth.password string
+        Optional basic auth password for -remoteRead.url
+  -remoteRead.basicAuth.username string
+        Optional basic auth username for -remoteRead.url
+  -remoteRead.lookback duration
         Lookback defines how far to look into past for alerts timeseries. For example, if lookback=1h then range from now() to now()-1h will be scanned. (default 1h0m0s)
-  -remoteread.url vmalert
-        Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts state. This configuration makes sense only if vmalert was configured with `remotewrite.url` before and has been successfully persisted its state. E.g. http://127.0.0.1:8428
-  -remotewrite.basicAuth.password string
-        Optional basic auth password for -remotewrite.url
-  -remotewrite.basicAuth.username string
-        Optional basic auth username for -remotewrite.url
-  -remotewrite.url string
+  -remoteRead.url vmalert
+        Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts state. This configuration makes sense only if vmalert was configured with `remoteWrite.url` before and has been successfully persisted its state. E.g. http://127.0.0.1:8428
+  -remoteWrite.basicAuth.password string
+        Optional basic auth password for -remoteWrite.url
+  -remoteWrite.basicAuth.username string
+        Optional basic auth username for -remoteWrite.url
+  -remoteWrite.concurrency int
+        Defines number of readers that concurrently write into remote storage (default 1)
+  -remoteWrite.maxBatchSize int
+        Defines defines max number of timeseries to be flushed at once (default 1000)
+  -remoteWrite.maxQueueSize int
+        Defines the max number of pending datapoints to remote write endpoint (default 100000)
+  -remoteWrite.url string
         Optional URL to Victoria Metrics or VMInsert where to persist alerts state in form of timeseries. E.g. http://127.0.0.1:8428
   -rule value
         Path to the file with alert rules. 
@@ -118,3 +121,22 @@ or send GET request to `/-/reload` endpoint.
 `vmalert` is mostly designed and built by VictoriaMetrics community.
 Feel free to share your experience and ideas for improving this 
 software. Please keep simplicity as the main priority.
+
+### How to build from sources
+
+It is recommended using 
+[binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) 
+- `vmalert` is located in `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.13.
+2. Run `make vmalert` from the root folder of the repository.
+   It builds `vmalert` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmalert-prod` from the root folder of the repository.
+   It builds `vmalert-prod` binary and puts it into the `bin` folder.

app/vmalert/alerting.go

@@ -0,0 +1,376 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"hash/fnv"
+	"sort"
+	"strconv"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+// AlertingRule is basic alert entity
+type AlertingRule struct {
+	Name        string
+	Expr        string
+	For         time.Duration
+	Labels      map[string]string
+	Annotations map[string]string
+	GroupID     uint64
+
+	// guard status fields
+	mu sync.RWMutex
+	// stores list of active alerts
+	alerts map[uint64]*notifier.Alert
+	// stores last moment of time Exec was called
+	lastExecTime time.Time
+	// stores last error that happened in Exec func
+	// resets on every successful Exec
+	// may be used as Health state
+	lastExecError error
+}
+
+func newAlertingRule(gID uint64, cfg config.Rule) *AlertingRule {
+	return &AlertingRule{
+		Name:        cfg.Alert,
+		Expr:        cfg.Expr,
+		For:         cfg.For,
+		Labels:      cfg.Labels,
+		Annotations: cfg.Annotations,
+		GroupID:     gID,
+		alerts:      make(map[uint64]*notifier.Alert),
+	}
+}
+
+// String implements Stringer interface
+func (ar *AlertingRule) String() string {
+	return ar.Name
+}
+
+// ID returns unique Rule ID
+// within the parent Group.
+func (ar *AlertingRule) ID() uint64 {
+	hash := fnv.New64a()
+	hash.Write([]byte("alerting"))
+	hash.Write([]byte("\xff"))
+	hash.Write([]byte(ar.Name))
+	return hash.Sum64()
+}
+
+// Exec executes AlertingRule expression via the given Querier.
+// Based on the Querier results AlertingRule maintains notifier.Alerts
+func (ar *AlertingRule) Exec(ctx context.Context, q datasource.Querier, series bool) ([]prompbmarshal.TimeSeries, error) {
+	qMetrics, err := q.Query(ctx, ar.Expr)
+	ar.mu.Lock()
+	defer ar.mu.Unlock()
+
+	ar.lastExecError = err
+	ar.lastExecTime = time.Now()
+	if err != nil {
+		return nil, fmt.Errorf("failed to execute query %q: %s", ar.Expr, err)
+	}
+
+	for h, a := range ar.alerts {
+		// cleanup inactive alerts from previous Exec
+		if a.State == notifier.StateInactive {
+			delete(ar.alerts, h)
+		}
+	}
+
+	updated := make(map[uint64]struct{})
+	// update list of active alerts
+	for _, m := range qMetrics {
+		h := hash(m)
+		updated[h] = struct{}{}
+		if a, ok := ar.alerts[h]; ok {
+			if a.Value != m.Value {
+				// update Value field with latest value
+				a.Value = m.Value
+				// and re-exec template since Value can be used
+				// in templates
+				err = ar.template(a)
+				if err != nil {
+					return nil, err
+				}
+			}
+			continue
+		}
+		a, err := ar.newAlert(m, ar.lastExecTime)
+		if err != nil {
+			ar.lastExecError = err
+			return nil, fmt.Errorf("failed to create alert: %s", err)
+		}
+		a.ID = h
+		a.State = notifier.StatePending
+		ar.alerts[h] = a
+	}
+
+	for h, a := range ar.alerts {
+		// if alert wasn't updated in this iteration
+		// means it is resolved already
+		if _, ok := updated[h]; !ok {
+			if a.State == notifier.StatePending {
+				// alert was in Pending state - it is not
+				// active anymore
+				delete(ar.alerts, h)
+				continue
+			}
+			a.State = notifier.StateInactive
+			continue
+		}
+		if a.State == notifier.StatePending && time.Since(a.Start) >= ar.For {
+			a.State = notifier.StateFiring
+			alertsFired.Inc()
+		}
+	}
+	if series {
+		return ar.toTimeSeries(ar.lastExecTime), nil
+	}
+	return nil, nil
+}
+
+func (ar *AlertingRule) toTimeSeries(timestamp time.Time) []prompbmarshal.TimeSeries {
+	var tss []prompbmarshal.TimeSeries
+	for _, a := range ar.alerts {
+		if a.State == notifier.StateInactive {
+			continue
+		}
+		ts := ar.alertToTimeSeries(a, timestamp)
+		tss = append(tss, ts...)
+	}
+	return tss
+}
+
+// UpdateWith copies all significant fields.
+// alerts state isn't copied since
+// it should be updated in next 2 Execs
+func (ar *AlertingRule) UpdateWith(r Rule) error {
+	nr, ok := r.(*AlertingRule)
+	if !ok {
+		return fmt.Errorf("BUG: attempt to update alerting rule with wrong type %#v", r)
+	}
+	ar.Expr = nr.Expr
+	ar.For = nr.For
+	ar.Labels = nr.Labels
+	ar.Annotations = nr.Annotations
+	return nil
+}
+
+// TODO: consider hashing algorithm in VM
+func hash(m datasource.Metric) uint64 {
+	hash := fnv.New64a()
+	labels := m.Labels
+	sort.Slice(labels, func(i, j int) bool {
+		return labels[i].Name < labels[j].Name
+	})
+	for _, l := range labels {
+		// drop __name__ to be consistent with Prometheus alerting
+		if l.Name == "__name__" {
+			continue
+		}
+		hash.Write([]byte(l.Name))
+		hash.Write([]byte(l.Value))
+		hash.Write([]byte("\xff"))
+	}
+	return hash.Sum64()
+}
+
+func (ar *AlertingRule) newAlert(m datasource.Metric, start time.Time) (*notifier.Alert, error) {
+	a := &notifier.Alert{
+		GroupID: ar.GroupID,
+		Name:    ar.Name,
+		Labels:  map[string]string{},
+		Value:   m.Value,
+		Start:   start,
+		Expr:    ar.Expr,
+	}
+	for _, l := range m.Labels {
+		// drop __name__ to be consistent with Prometheus alerting
+		if l.Name == "__name__" {
+			continue
+		}
+		a.Labels[l.Name] = l.Value
+	}
+	return a, ar.template(a)
+}
+
+func (ar *AlertingRule) template(a *notifier.Alert) error {
+	// 1. template rule labels with data labels
+	rLabels, err := a.ExecTemplate(ar.Labels)
+	if err != nil {
+		return err
+	}
+
+	// 2. merge data labels and rule labels
+	// metric labels may be overridden by
+	// rule labels
+	for k, v := range rLabels {
+		a.Labels[k] = v
+	}
+
+	// 3. template merged labels
+	a.Labels, err = a.ExecTemplate(a.Labels)
+	if err != nil {
+		return err
+	}
+
+	a.Annotations, err = a.ExecTemplate(ar.Annotations)
+	return err
+}
+
+// AlertAPI generates APIAlert object from alert by its id(hash)
+func (ar *AlertingRule) AlertAPI(id uint64) *APIAlert {
+	ar.mu.RLock()
+	defer ar.mu.RUnlock()
+	a, ok := ar.alerts[id]
+	if !ok {
+		return nil
+	}
+	return ar.newAlertAPI(*a)
+}
+
+// RuleAPI returns Rule representation in form
+// of APIAlertingRule
+func (ar *AlertingRule) RuleAPI() APIAlertingRule {
+	var lastErr string
+	if ar.lastExecError != nil {
+		lastErr = ar.lastExecError.Error()
+	}
+	return APIAlertingRule{
+		// encode as strings to avoid rounding
+		ID:          fmt.Sprintf("%d", ar.ID()),
+		GroupID:     fmt.Sprintf("%d", ar.GroupID),
+		Name:        ar.Name,
+		Expression:  ar.Expr,
+		For:         ar.For.String(),
+		LastError:   lastErr,
+		LastExec:    ar.lastExecTime,
+		Labels:      ar.Labels,
+		Annotations: ar.Annotations,
+	}
+}
+
+// AlertsAPI generates list of APIAlert objects from existing alerts
+func (ar *AlertingRule) AlertsAPI() []*APIAlert {
+	var alerts []*APIAlert
+	ar.mu.RLock()
+	for _, a := range ar.alerts {
+		alerts = append(alerts, ar.newAlertAPI(*a))
+	}
+	ar.mu.RUnlock()
+	return alerts
+}
+
+func (ar *AlertingRule) newAlertAPI(a notifier.Alert) *APIAlert {
+	return &APIAlert{
+		// encode as strings to avoid rounding
+		ID:      fmt.Sprintf("%d", a.ID),
+		GroupID: fmt.Sprintf("%d", a.GroupID),
+
+		Name:        a.Name,
+		Expression:  ar.Expr,
+		Labels:      a.Labels,
+		Annotations: a.Annotations,
+		State:       a.State.String(),
+		ActiveAt:    a.Start,
+		Value:       strconv.FormatFloat(a.Value, 'e', -1, 64),
+	}
+}
+
+const (
+	// AlertMetricName is the metric name for synthetic alert timeseries.
+	alertMetricName = "ALERTS"
+	// AlertForStateMetricName is the metric name for 'for' state of alert.
+	alertForStateMetricName = "ALERTS_FOR_STATE"
+
+	// AlertNameLabel is the label name indicating the name of an alert.
+	alertNameLabel = "alertname"
+	// AlertStateLabel is the label name indicating the state of an alert.
+	alertStateLabel = "alertstate"
+)
+
+// alertToTimeSeries converts the given alert with the given timestamp to timeseries
+func (ar *AlertingRule) alertToTimeSeries(a *notifier.Alert, timestamp time.Time) []prompbmarshal.TimeSeries {
+	var tss []prompbmarshal.TimeSeries
+	tss = append(tss, alertToTimeSeries(ar.Name, a, timestamp))
+	if ar.For > 0 {
+		tss = append(tss, alertForToTimeSeries(ar.Name, a, timestamp))
+	}
+	return tss
+}
+
+func alertToTimeSeries(name string, a *notifier.Alert, timestamp time.Time) prompbmarshal.TimeSeries {
+	labels := make(map[string]string)
+	for k, v := range a.Labels {
+		labels[k] = v
+	}
+	labels["__name__"] = alertMetricName
+	labels[alertNameLabel] = name
+	labels[alertStateLabel] = a.State.String()
+	return newTimeSeries(1, labels, timestamp)
+}
+
+// alertForToTimeSeries returns a timeseries that represents
+// state of active alerts, where value is time when alert become active
+func alertForToTimeSeries(name string, a *notifier.Alert, timestamp time.Time) prompbmarshal.TimeSeries {
+	labels := make(map[string]string)
+	for k, v := range a.Labels {
+		labels[k] = v
+	}
+	labels["__name__"] = alertForStateMetricName
+	labels[alertNameLabel] = name
+	return newTimeSeries(float64(a.Start.Unix()), labels, timestamp)
+}
+
+// Restore restores the state of active alerts basing on previously written timeseries.
+// Restore restores only Start field. Field State will be always Pending and supposed
+// to be updated on next Exec, as well as Value field.
+// Only rules with For > 0 will be restored.
+func (ar *AlertingRule) Restore(ctx context.Context, q datasource.Querier, lookback time.Duration) error {
+	if q == nil {
+		return fmt.Errorf("querier is nil")
+	}
+	// Get the last datapoint in range via MetricsQL `last_over_time`.
+	// We don't use plain PromQL since Prometheus doesn't support
+	// remote write protocol which is used for state persistence in vmalert.
+	expr := fmt.Sprintf("last_over_time(%s{alertname=%q}[%ds])",
+		alertForStateMetricName, ar.Name, int(lookback.Seconds()))
+	qMetrics, err := q.Query(ctx, expr)
+	if err != nil {
+		return err
+	}
+
+	for _, m := range qMetrics {
+		labels := m.Labels
+		m.Labels = make([]datasource.Label, 0)
+		// drop all extra labels, so hash key will
+		// be identical to timeseries received in Exec
+		for _, l := range labels {
+			if l.Name == alertNameLabel {
+				continue
+			}
+			// drop all overridden labels
+			if _, ok := ar.Labels[l.Name]; ok {
+				continue
+			}
+			m.Labels = append(m.Labels, l)
+		}
+
+		a, err := ar.newAlert(m, time.Unix(int64(m.Value), 0))
+		if err != nil {
+			return fmt.Errorf("failed to create alert: %s", err)
+		}
+		a.ID = hash(m)
+		a.State = notifier.StatePending
+		ar.alerts[a.ID] = a
+		logger.Infof("alert %q(%d) restored to state at %v", a.Name, a.ID, a.Start)
+	}
+	return nil
+}

app/vmalert/alerting_test.go

@@ -2,7 +2,6 @@ package main
 
 import (
 	"context"
-	"sync"
 	"testing"
 	"time"
 
@@ -11,30 +10,15 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
 )
 
-func TestRule_Validate(t *testing.T) {
-	if err := (&Rule{}).Validate(); err == nil {
-		t.Errorf("exptected empty name error")
-	}
-	if err := (&Rule{Name: "alert"}).Validate(); err == nil {
-		t.Errorf("exptected empty expr error")
-	}
-	if err := (&Rule{Name: "alert", Expr: "test{"}).Validate(); err == nil {
-		t.Errorf("exptected invalid expr error")
-	}
-	if err := (&Rule{Name: "alert", Expr: "test>0"}).Validate(); err != nil {
-		t.Errorf("exptected valid rule got %s", err)
-	}
-}
-
-func TestRule_AlertToTimeSeries(t *testing.T) {
+func TestAlertingRule_ToTimeSeries(t *testing.T) {
 	timestamp := time.Now()
 	testCases := []struct {
-		rule  *Rule
+		rule  *AlertingRule
 		alert *notifier.Alert
 		expTS []prompbmarshal.TimeSeries
 	}{
 		{
-			newTestRule("instant", 0),
+			newTestAlertingRule("instant", 0),
 			&notifier.Alert{State: notifier.StateFiring},
 			[]prompbmarshal.TimeSeries{
 				newTimeSeries(1, map[string]string{
@@ -45,7 +29,7 @@ func TestRule_AlertToTimeSeries(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("instant extra labels", 0),
+			newTestAlertingRule("instant extra labels", 0),
 			&notifier.Alert{State: notifier.StateFiring, Labels: map[string]string{
 				"job":      "foo",
 				"instance": "bar",
@@ -61,7 +45,7 @@ func TestRule_AlertToTimeSeries(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("instant labels override", 0),
+			newTestAlertingRule("instant labels override", 0),
 			&notifier.Alert{State: notifier.StateFiring, Labels: map[string]string{
 				alertStateLabel: "foo",
 				"__name__":      "bar",
@@ -75,7 +59,7 @@ func TestRule_AlertToTimeSeries(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for", time.Second),
+			newTestAlertingRule("for", time.Second),
 			&notifier.Alert{State: notifier.StateFiring, Start: timestamp.Add(time.Second)},
 			[]prompbmarshal.TimeSeries{
 				newTimeSeries(1, map[string]string{
@@ -90,7 +74,7 @@ func TestRule_AlertToTimeSeries(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for pending", 10*time.Second),
+			newTestAlertingRule("for pending", 10*time.Second),
 			&notifier.Alert{State: notifier.StatePending, Start: timestamp.Add(time.Second)},
 			[]prompbmarshal.TimeSeries{
 				newTimeSeries(1, map[string]string{
@@ -107,58 +91,28 @@ func TestRule_AlertToTimeSeries(t *testing.T) {
 	}
 	for _, tc := range testCases {
 		t.Run(tc.rule.Name, func(t *testing.T) {
-			tss := tc.rule.AlertToTimeSeries(tc.alert, timestamp)
-			if len(tc.expTS) != len(tss) {
-				t.Fatalf("expected number of timeseries %d; got %d", len(tc.expTS), len(tss))
-			}
-			for i := range tc.expTS {
-				expTS, gotTS := tc.expTS[i], tss[i]
-				if len(expTS.Samples) != len(gotTS.Samples) {
-					t.Fatalf("expected number of samples %d; got %d", len(expTS.Samples), len(gotTS.Samples))
-				}
-				for i, exp := range expTS.Samples {
-					got := gotTS.Samples[i]
-					if got.Value != exp.Value {
-						t.Errorf("expected value %.2f; got %.2f", exp.Value, got.Value)
-					}
-					if got.Timestamp != exp.Timestamp {
-						t.Errorf("expected timestamp %d; got %d", exp.Timestamp, got.Timestamp)
-					}
-				}
-				if len(expTS.Labels) != len(gotTS.Labels) {
-					t.Fatalf("expected number of labels %d; got %d", len(expTS.Labels), len(gotTS.Labels))
-				}
-				for i, exp := range expTS.Labels {
-					got := gotTS.Labels[i]
-					if got.Name != exp.Name {
-						t.Errorf("expected label name %q; got %q", exp.Name, got.Name)
-					}
-					if got.Value != exp.Value {
-						t.Errorf("expected label value %q; got %q", exp.Value, got.Value)
-					}
-				}
+			tc.rule.alerts[tc.alert.ID] = tc.alert
+			tss := tc.rule.toTimeSeries(timestamp)
+			if err := compareTimeSeries(t, tc.expTS, tss); err != nil {
+				t.Fatalf("timeseries missmatch: %s", err)
 			}
 		})
 	}
 }
 
-func newTestRule(name string, waitFor time.Duration) *Rule {
-	return &Rule{Name: name, alerts: make(map[uint64]*notifier.Alert), For: waitFor}
-}
-
-func TestRule_Exec(t *testing.T) {
+func TestAlertingRule_Exec(t *testing.T) {
 	testCases := []struct {
-		rule      *Rule
+		rule      *AlertingRule
 		steps     [][]datasource.Metric
 		expAlerts map[uint64]*notifier.Alert
 	}{
 		{
-			newTestRule("empty", 0),
+			newTestAlertingRule("empty", 0),
 			[][]datasource.Metric{},
 			map[uint64]*notifier.Alert{},
 		},
 		{
-			newTestRule("empty labels", 0),
+			newTestAlertingRule("empty labels", 0),
 			[][]datasource.Metric{
 				{datasource.Metric{}},
 			},
@@ -167,7 +121,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("single-firing", 0),
+			newTestAlertingRule("single-firing", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 			},
@@ -176,7 +130,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("single-firing=>inactive", 0),
+			newTestAlertingRule("single-firing=>inactive", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{},
@@ -186,7 +140,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("single-firing=>inactive=>firing", 0),
+			newTestAlertingRule("single-firing=>inactive=>firing", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{},
@@ -197,7 +151,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("single-firing=>inactive=>firing=>inactive", 0),
+			newTestAlertingRule("single-firing=>inactive=>firing=>inactive", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{},
@@ -209,7 +163,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("single-firing=>inactive=>firing=>inactive=>empty", 0),
+			newTestAlertingRule("single-firing=>inactive=>firing=>inactive=>empty", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{},
@@ -220,7 +174,7 @@ func TestRule_Exec(t *testing.T) {
 			map[uint64]*notifier.Alert{},
 		},
 		{
-			newTestRule("single-firing=>inactive=>firing=>inactive=>empty=>firing", 0),
+			newTestAlertingRule("single-firing=>inactive=>firing=>inactive=>empty=>firing", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{},
@@ -234,7 +188,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("multiple-firing", 0),
+			newTestAlertingRule("multiple-firing", 0),
 			[][]datasource.Metric{
 				{
 					metricWithLabels(t, "name", "foo"),
@@ -249,7 +203,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("multiple-steps-firing", 0),
+			newTestAlertingRule("multiple-steps-firing", 0),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{metricWithLabels(t, "name", "foo1")},
@@ -264,7 +218,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("duplicate", 0),
+			newTestAlertingRule("duplicate", 0),
 			[][]datasource.Metric{
 				{
 					// metrics with the same labelset should result in one alert
@@ -277,7 +231,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for-pending", time.Minute),
+			newTestAlertingRule("for-pending", time.Minute),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 			},
@@ -286,7 +240,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for-fired", time.Millisecond),
+			newTestAlertingRule("for-fired", time.Millisecond),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{metricWithLabels(t, "name", "foo")},
@@ -296,7 +250,17 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for-pending=>inactive", time.Millisecond),
+			newTestAlertingRule("for-pending=>empty", time.Second),
+			[][]datasource.Metric{
+				{metricWithLabels(t, "name", "foo")},
+				{metricWithLabels(t, "name", "foo")},
+				// empty step to reset and delete pending alerts
+				{},
+			},
+			map[uint64]*notifier.Alert{},
+		},
+		{
+			newTestAlertingRule("for-pending=>firing=>inactive", time.Millisecond),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{metricWithLabels(t, "name", "foo")},
@@ -308,22 +272,10 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for-pending=>firing=>inactive", time.Millisecond),
+			newTestAlertingRule("for-pending=>firing=>inactive=>pending", time.Millisecond),
 			[][]datasource.Metric{
-				{metricWithLabels(t, "name", "foo")},
-				{metricWithLabels(t, "name", "foo")},
-				// empty step to reset pending alerts
-				{},
-			},
-			map[uint64]*notifier.Alert{
-				hash(metricWithLabels(t, "name", "foo")): {State: notifier.StateInactive},
-			},
-		},
-		{
-			newTestRule("for-pending=>firing=>inactive=>pending", time.Millisecond),
-			[][]datasource.Metric{
-				{metricWithLabels(t, "name", "foo")},
-				{metricWithLabels(t, "name", "foo")},
+				//{metricWithLabels(t, "name", "foo")},
+				//{metricWithLabels(t, "name", "foo")},
 				// empty step to reset pending alerts
 				{},
 				{metricWithLabels(t, "name", "foo")},
@@ -333,7 +285,7 @@ func TestRule_Exec(t *testing.T) {
 			},
 		},
 		{
-			newTestRule("for-pending=>firing=>inactive=>pending=>firing", time.Millisecond),
+			newTestAlertingRule("for-pending=>firing=>inactive=>pending=>firing", time.Millisecond),
 			[][]datasource.Metric{
 				{metricWithLabels(t, "name", "foo")},
 				{metricWithLabels(t, "name", "foo")},
@@ -351,11 +303,11 @@ func TestRule_Exec(t *testing.T) {
 	for _, tc := range testCases {
 		t.Run(tc.rule.Name, func(t *testing.T) {
 			fq := &fakeQuerier{}
-			tc.rule.group = fakeGroup
+			tc.rule.GroupID = fakeGroup.ID()
 			for _, step := range tc.steps {
 				fq.reset()
 				fq.add(step...)
-				if err := tc.rule.Exec(context.TODO(), fq); err != nil {
+				if _, err := tc.rule.Exec(context.TODO(), fq, false); err != nil {
 					t.Fatalf("unexpected err: %s", err)
 				}
 				// artificial delay between applying steps
@@ -377,49 +329,9 @@ func TestRule_Exec(t *testing.T) {
 	}
 }
 
-func metricWithLabels(t *testing.T, labels ...string) datasource.Metric {
-	t.Helper()
-	if len(labels) == 0 || len(labels)%2 != 0 {
-		t.Fatalf("expected to get even number of labels")
-	}
-	m := datasource.Metric{}
-	for i := 0; i < len(labels); i += 2 {
-		m.Labels = append(m.Labels, datasource.Label{
-			Name:  labels[i],
-			Value: labels[i+1],
-		})
-	}
-	return m
-}
-
-type fakeQuerier struct {
-	sync.Mutex
-	metrics []datasource.Metric
-}
-
-func (fq *fakeQuerier) reset() {
-	fq.Lock()
-	fq.metrics = fq.metrics[:0]
-	fq.Unlock()
-}
-
-func (fq *fakeQuerier) add(metrics ...datasource.Metric) {
-	fq.Lock()
-	fq.metrics = append(fq.metrics, metrics...)
-	fq.Unlock()
-}
-
-func (fq *fakeQuerier) Query(_ context.Context, _ string) ([]datasource.Metric, error) {
-	fq.Lock()
-	cpy := make([]datasource.Metric, len(fq.metrics))
-	copy(cpy, fq.metrics)
-	fq.Unlock()
-	return cpy, nil
-}
-
-func TestRule_Restore(t *testing.T) {
+func TestAlertingRule_Restore(t *testing.T) {
 	testCases := []struct {
-		rule      *Rule
+		rule      *AlertingRule
 		metrics   []datasource.Metric
 		expAlerts map[uint64]*notifier.Alert
 	}{
@@ -504,7 +416,7 @@ func TestRule_Restore(t *testing.T) {
 	for _, tc := range testCases {
 		t.Run(tc.rule.Name, func(t *testing.T) {
 			fq := &fakeQuerier{}
-			tc.rule.group = fakeGroup
+			tc.rule.GroupID = fakeGroup.ID()
 			fq.add(tc.metrics...)
 			if err := tc.rule.Restore(context.TODO(), fq, time.Hour); err != nil {
 				t.Fatalf("unexpected err: %s", err)
@@ -528,8 +440,8 @@ func TestRule_Restore(t *testing.T) {
 	}
 }
 
-func newTestRuleWithLabels(name string, labels ...string) *Rule {
-	r := newTestRule(name, 0)
+func newTestRuleWithLabels(name string, labels ...string) *AlertingRule {
+	r := newTestAlertingRule(name, 0)
 	r.Labels = make(map[string]string)
 	for i := 0; i < len(labels); i += 2 {
 		r.Labels[labels[i]] = labels[i+1]
@@ -537,9 +449,6 @@ func newTestRuleWithLabels(name string, labels ...string) *Rule {
 	return r
 }
 
-func metricWithValueAndLabels(t *testing.T, value float64, labels ...string) datasource.Metric {
-	t.Helper()
-	m := metricWithLabels(t, labels...)
-	m.Value = value
-	return m
+func newTestAlertingRule(name string, waitFor time.Duration) *AlertingRule {
+	return &AlertingRule{Name: name, alerts: make(map[uint64]*notifier.Alert), For: waitFor}
 }

app/vmalert/config.go

@@ -1,73 +0,0 @@
-package main
-
-import (
-	"fmt"
-	"gopkg.in/yaml.v2"
-	"io/ioutil"
-	"path/filepath"
-	"strings"
-
-	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
-)
-
-// Parse parses rule configs from given file patterns
-func Parse(pathPatterns []string, validateAnnotations bool) ([]Group, error) {
-	var fp []string
-	for _, pattern := range pathPatterns {
-		matches, err := filepath.Glob(pattern)
-		if err != nil {
-			return nil, fmt.Errorf("error reading file patther %s:%v", pattern, err)
-		}
-		fp = append(fp, matches...)
-	}
-	var groups []Group
-	for _, file := range fp {
-		groupsNames := map[string]struct{}{}
-		gr, err := parseFile(file)
-		if err != nil {
-			return nil, fmt.Errorf("file %s: %w", file, err)
-		}
-		for _, g := range gr {
-			if _, ok := groupsNames[g.Name]; ok {
-				return nil, fmt.Errorf("one file can not contain groups with the same name %s, filepath:%s", g.Name, file)
-			}
-			g.File = file
-			g.done = make(chan struct{})
-			g.finished = make(chan struct{})
-
-			groupsNames[g.Name] = struct{}{}
-			for _, rule := range g.Rules {
-				if err = rule.Validate(); err != nil {
-					return nil, fmt.Errorf("invalid rule filepath: %s, group %s: %w", file, g.Name, err)
-				}
-				if validateAnnotations {
-					if err = notifier.ValidateTemplates(rule.Annotations); err != nil {
-						return nil, fmt.Errorf("invalid annotations filepath: %s, group %s: %w", file, g.Name, err)
-					}
-					if err = notifier.ValidateTemplates(rule.Labels); err != nil {
-						return nil, fmt.Errorf("invalid labels filepath: %s, group %s: %w", file, g.Name, err)
-					}
-				}
-				rule.group = g
-				rule.alerts = make(map[uint64]*notifier.Alert)
-			}
-			groups = append(groups, g)
-		}
-	}
-	if len(groups) < 1 {
-		return nil, fmt.Errorf("no groups found in %s", strings.Join(pathPatterns, ";"))
-	}
-	return groups, nil
-}
-
-func parseFile(path string) ([]Group, error) {
-	data, err := ioutil.ReadFile(path)
-	if err != nil {
-		return nil, fmt.Errorf("error reading alert rule file: %w", err)
-	}
-	g := struct {
-		Groups []Group `yaml:"groups"`
-	}{}
-	err = yaml.Unmarshal(data, &g)
-	return g.Groups, err
-}

app/vmalert/config/config.go

@@ -0,0 +1,147 @@
+package config
+
+import (
+	"fmt"
+	"gopkg.in/yaml.v2"
+	"io/ioutil"
+	"path/filepath"
+	"strings"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
+	"github.com/VictoriaMetrics/metricsql"
+)
+
+// Group contains list of Rules grouped into
+// entity with one name and evaluation interval
+type Group struct {
+	File     string
+	Name     string        `yaml:"name"`
+	Interval time.Duration `yaml:"interval,omitempty"`
+	Rules    []Rule        `yaml:"rules"`
+
+	// Catches all undefined fields and must be empty after parsing.
+	XXX map[string]interface{} `yaml:",inline"`
+}
+
+// Validate check for internal Group or Rule configuration errors
+func (g *Group) Validate(validateAnnotations bool) error {
+	if g.Name == "" {
+		return fmt.Errorf("group name must be set")
+	}
+	if len(g.Rules) == 0 {
+		return fmt.Errorf("group %q can't contain no rules", g.Name)
+	}
+	uniqueRules := map[string]struct{}{}
+	for _, r := range g.Rules {
+		ruleName := r.Record
+		if r.Alert != "" {
+			ruleName = r.Alert
+		}
+		if _, ok := uniqueRules[ruleName]; ok {
+			return fmt.Errorf("rule name %q duplicate", ruleName)
+		}
+		uniqueRules[ruleName] = struct{}{}
+		if err := r.Validate(); err != nil {
+			return fmt.Errorf("invalid rule %q.%q: %s", g.Name, ruleName, err)
+		}
+		if !validateAnnotations {
+			continue
+		}
+		if err := notifier.ValidateTemplates(r.Annotations); err != nil {
+			return fmt.Errorf("invalid annotations for rule %q.%q: %s", g.Name, ruleName, err)
+		}
+		if err := notifier.ValidateTemplates(r.Labels); err != nil {
+			return fmt.Errorf("invalid labels for rule %q.%q: %s", g.Name, ruleName, err)
+		}
+	}
+	return checkOverflow(g.XXX, fmt.Sprintf("group %q", g.Name))
+}
+
+// Rule describes entity that represent either
+// recording rule or alerting rule.
+type Rule struct {
+	Record      string            `yaml:"record,omitempty"`
+	Alert       string            `yaml:"alert,omitempty"`
+	Expr        string            `yaml:"expr"`
+	For         time.Duration     `yaml:"for,omitempty"`
+	Labels      map[string]string `yaml:"labels,omitempty"`
+	Annotations map[string]string `yaml:"annotations,omitempty"`
+}
+
+// Validate check for Rule configuration errors
+func (r *Rule) Validate() error {
+	if (r.Record == "" && r.Alert == "") || (r.Record != "" && r.Alert != "") {
+		return fmt.Errorf("either `record` or `alert` must be set")
+	}
+	if r.Expr == "" {
+		return fmt.Errorf("expression can't be empty")
+	}
+	if _, err := metricsql.Parse(r.Expr); err != nil {
+		return fmt.Errorf("invalid expression: %w", err)
+	}
+	return nil
+}
+
+// Parse parses rule configs from given file patterns
+func Parse(pathPatterns []string, validateAnnotations bool) ([]Group, error) {
+	var fp []string
+	for _, pattern := range pathPatterns {
+		matches, err := filepath.Glob(pattern)
+		if err != nil {
+			return nil, fmt.Errorf("error reading file pattern %s: %v", pattern, err)
+		}
+		fp = append(fp, matches...)
+	}
+	var groups []Group
+	for _, file := range fp {
+		uniqueGroups := map[string]struct{}{}
+		gr, err := parseFile(file)
+		if err != nil {
+			return nil, fmt.Errorf("failed to parse file %q: %w", file, err)
+		}
+		for _, g := range gr {
+			if err := g.Validate(validateAnnotations); err != nil {
+				return nil, fmt.Errorf("invalid group %q in file %q: %s", g.Name, file, err)
+			}
+			if _, ok := uniqueGroups[g.Name]; ok {
+				return nil, fmt.Errorf("group name %q duplicate in file %q", g.Name, file)
+			}
+			uniqueGroups[g.Name] = struct{}{}
+			g.File = file
+			groups = append(groups, g)
+		}
+	}
+	if len(groups) < 1 {
+		return nil, fmt.Errorf("no groups found in %s", strings.Join(pathPatterns, ";"))
+	}
+	return groups, nil
+}
+
+func parseFile(path string) ([]Group, error) {
+	data, err := ioutil.ReadFile(path)
+	if err != nil {
+		return nil, fmt.Errorf("error reading alert rule file: %w", err)
+	}
+	g := struct {
+		Groups []Group `yaml:"groups"`
+		// Catches all undefined fields and must be empty after parsing.
+		XXX map[string]interface{} `yaml:",inline"`
+	}{}
+	err = yaml.Unmarshal(data, &g)
+	if err != nil {
+		return nil, err
+	}
+	return g.Groups, checkOverflow(g.XXX, "config")
+}
+
+func checkOverflow(m map[string]interface{}, ctx string) error {
+	if len(m) > 0 {
+		var keys []string
+		for k := range m {
+			keys = append(keys, k)
+		}
+		return fmt.Errorf("unknown fields in %s: %s", ctx, strings.Join(keys, ", "))
+	}
+	return nil
+}

app/vmalert/config/config_test.go

@@ -0,0 +1,83 @@
+package config
+
+import (
+	"net/url"
+	"os"
+	"strings"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
+)
+
+func TestMain(m *testing.M) {
+	u, _ := url.Parse("https://victoriametrics.com/path")
+	notifier.InitTemplateFunc(u)
+	os.Exit(m.Run())
+}
+
+func TestParseGood(t *testing.T) {
+	if _, err := Parse([]string{"testdata/*good.rules", "testdata/dir/*good.*"}, true); err != nil {
+		t.Errorf("error parsing files %s", err)
+	}
+}
+
+func TestParseBad(t *testing.T) {
+	testCases := []struct {
+		path   []string
+		expErr string
+	}{
+		{
+			[]string{"testdata/rules0-bad.rules"},
+			"unexpected token",
+		},
+		{
+			[]string{"testdata/dir/rules0-bad.rules"},
+			"error parsing annotation",
+		},
+		{
+			[]string{"testdata/dir/rules1-bad.rules"},
+			"duplicate in file",
+		},
+		{
+			[]string{"testdata/dir/rules2-bad.rules"},
+			"function \"value\" not defined",
+		},
+		{
+			[]string{"testdata/dir/rules3-bad.rules"},
+			"either `record` or `alert` must be set",
+		},
+		{
+			[]string{"testdata/dir/rules4-bad.rules"},
+			"either `record` or `alert` must be set",
+		},
+		{
+			[]string{"testdata/*.yaml"},
+			"no groups found",
+		},
+	}
+	for _, tc := range testCases {
+		_, err := Parse(tc.path, true)
+		if err == nil {
+			t.Errorf("expected to get error")
+			return
+		}
+		if !strings.Contains(err.Error(), tc.expErr) {
+			t.Errorf("expected err to contain %q; got %q instead", tc.expErr, err)
+		}
+	}
+}
+
+func TestRule_Validate(t *testing.T) {
+	if err := (&Rule{}).Validate(); err == nil {
+		t.Errorf("exptected empty name error")
+	}
+	if err := (&Rule{Alert: "alert"}).Validate(); err == nil {
+		t.Errorf("exptected empty expr error")
+	}
+	if err := (&Rule{Alert: "alert", Expr: "test{"}).Validate(); err == nil {
+		t.Errorf("exptected invalid expr error")
+	}
+	if err := (&Rule{Alert: "alert", Expr: "test>0"}).Validate(); err != nil {
+		t.Errorf("exptected valid rule got %s", err)
+	}
+}

app/vmalert/config/testdata/dir/rules0-good.rules

@@ -6,6 +6,7 @@ groups:
         expr: vm_rows > 0
         labels:
           label: bar
+          expr: "{{ $expr|queryEscape }}"
         annotations:
           summary: "{{ $value|humanize }}"
           description: "{{$labels}}"

app/vmalert/config/testdata/dir/rules1-good.rules

@@ -9,5 +9,3 @@ groups:
         annotations:
           summary: "{{ $value }}"
           description: "{{$labels}}"
-
-

app/vmalert/config/testdata/dir/rules3-bad.rules

@@ -0,0 +1,5 @@
+groups:
+  - name: group
+    rules:
+      - for: 5m
+        expr: vm_rows > 0

app/vmalert/config/testdata/dir/rules4-bad.rules

@@ -0,0 +1,7 @@
+groups:
+  - name: group
+    rules:
+      - alert: rows
+        record: record
+        for: 5m
+        expr: vm_rows > 0

app/vmalert/config/testdata/dir/rules5-bad.rules

@@ -0,0 +1,7 @@
+groups:
+  - name: group
+    rules:
+      - alert: rows
+        expr: vm_rows > 0
+      - record: rows
+        expr: sum(vm_rows)

app/vmalert/config/testdata/rules2-good.rules

@@ -0,0 +1,28 @@
+groups:
+  - name: TestGroup
+    interval: 2s
+    rules:
+      - alert: Conns
+        expr: sum(vm_tcplistener_conns) by(instance) > 1
+        for: 3m
+        annotations:
+          summary: "Too high connection number for {{$labels.instance}}"
+          description: "It is {{ $value }} connections for {{$labels.instance}}"
+      - alert: ExampleAlertAlwaysFiring
+        expr: sum by(job)
+          (up == 1)
+      - record: handler:requests:rate5m
+        expr: sum(rate(prometheus_http_requests_total[5m])) by (handler)
+        labels:
+          recording: true
+      - record: code:requests:rate5m
+        expr: sum(rate(promhttp_metric_handler_requests_total[5m])) by (code)
+        labels:
+          recording: true
+      - record: successful_requests:ratio_rate5m
+        labels:
+          recording: true
+        expr: |2
+            sum(code:requests:rate5m{code="200"})
+          /
+            sum(code:requests:rate5m)

app/vmalert/config_test.go

@@ -1,39 +0,0 @@
-package main
-
-import (
-	"net/url"
-	"os"
-	"testing"
-
-	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
-)
-
-func TestMain(m *testing.M) {
-	u, _ := url.Parse("https://victoriametrics.com/path")
-	notifier.InitTemplateFunc(u)
-	os.Exit(m.Run())
-}
-
-func TestParseGood(t *testing.T) {
-	if _, err := Parse([]string{"testdata/*good.rules", "testdata/dir/*good.*"}, true); err != nil {
-		t.Errorf("error parsing files %s", err)
-	}
-}
-
-func TestParseBad(t *testing.T) {
-	if _, err := Parse([]string{"testdata/rules0-bad.rules"}, true); err == nil {
-		t.Errorf("expected syntaxt error")
-	}
-	if _, err := Parse([]string{"testdata/dir/rules0-bad.rules"}, true); err == nil {
-		t.Errorf("expected template annotation error")
-	}
-	if _, err := Parse([]string{"testdata/dir/rules1-bad.rules"}, true); err == nil {
-		t.Errorf("expected same group error")
-	}
-	if _, err := Parse([]string{"testdata/dir/rules2-bad.rules"}, true); err == nil {
-		t.Errorf("expected template label error")
-	}
-	if _, err := Parse([]string{"testdata/*.yaml"}, true); err == nil {
-		t.Errorf("expected empty group")
-	}
-}

app/vmalert/group.go

@@ -4,8 +4,10 @@ import (
 	"context"
 	"fmt"
 	"hash/fnv"
+	"sync"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/remotewrite"
@@ -15,17 +17,49 @@ import (
 
 // Group is an entity for grouping rules
 type Group struct {
-	Name  string
-	File  string
-	Rules []*Rule
+	Name     string
+	File     string
+	Rules    []Rule
+	Interval time.Duration
 
-	done     chan struct{}
-	finished chan struct{}
+	doneCh     chan struct{}
+	finishedCh chan struct{}
+	// channel accepts new Group obj
+	// which supposed to update current group
+	updateCh chan *Group
+	mu       sync.RWMutex
+}
+
+func newGroup(cfg config.Group, defaultInterval time.Duration) *Group {
+	g := &Group{
+		Name:       cfg.Name,
+		File:       cfg.File,
+		Interval:   cfg.Interval,
+		doneCh:     make(chan struct{}),
+		finishedCh: make(chan struct{}),
+		updateCh:   make(chan *Group),
+	}
+	if g.Interval == 0 {
+		g.Interval = defaultInterval
+	}
+	rules := make([]Rule, len(cfg.Rules))
+	for i, r := range cfg.Rules {
+		rules[i] = g.newRule(r)
+	}
+	g.Rules = rules
+	return g
+}
+
+func (g *Group) newRule(rule config.Rule) Rule {
+	if rule.Alert != "" {
+		return newAlertingRule(g.ID(), rule)
+	}
+	return newRecordingRule(g.ID(), rule)
 }
 
 // ID return unique group ID that consists of
 // rules file and group name
-func (g Group) ID() uint64 {
+func (g *Group) ID() uint64 {
 	hash := fnv.New64a()
 	hash.Write([]byte(g.File))
 	hash.Write([]byte("\xff"))
@@ -33,50 +67,62 @@ func (g Group) ID() uint64 {
 	return hash.Sum64()
 }
 
-// Restore restores alerts state for all group rules with For > 0
+// Restore restores alerts state for group rules
 func (g *Group) Restore(ctx context.Context, q datasource.Querier, lookback time.Duration) error {
 	for _, rule := range g.Rules {
-		if rule.For == 0 {
-			return nil
+		rr, ok := rule.(*AlertingRule)
+		if !ok {
+			continue
 		}
-		if err := rule.Restore(ctx, q, lookback); err != nil {
-			return fmt.Errorf("error while restoring rule %q: %s", rule.Name, err)
+		if rr.For < 1 {
+			continue
+		}
+		if err := rr.Restore(ctx, q, lookback); err != nil {
+			return fmt.Errorf("error while restoring rule %q: %s", rule, err)
 		}
 	}
 	return nil
 }
 
 // updateWith updates existing group with
-// passed group object.
-func (g *Group) updateWith(newGroup Group) {
-	rulesRegistry := make(map[string]*Rule)
+// passed group object. This function ignores group
+// evaluation interval change. It supposed to be updated
+// in group.start function.
+// Not thread-safe.
+func (g *Group) updateWith(newGroup *Group) error {
+	rulesRegistry := make(map[uint64]Rule)
 	for _, nr := range newGroup.Rules {
-		rulesRegistry[nr.id()] = nr
+		rulesRegistry[nr.ID()] = nr
 	}
 
 	for i, or := range g.Rules {
-		nr, ok := rulesRegistry[or.id()]
+		nr, ok := rulesRegistry[or.ID()]
 		if !ok {
 			// old rule is not present in the new list
-			// and must be removed
-			or = nil
-			g.Rules = append(g.Rules[:i], g.Rules[i+1:]...)
+			// so we mark it for removing
+			g.Rules[i] = nil
 			continue
 		}
-
-		// copy all significant fields.
-		// alerts state isn't copied since
-		// it should be updated in next 2 Evals
-		or.For = nr.For
-		or.Expr = nr.Expr
-		or.Labels = nr.Labels
-		or.Annotations = nr.Annotations
-		delete(rulesRegistry, nr.id())
+		if err := or.UpdateWith(nr); err != nil {
+			return err
+		}
+		delete(rulesRegistry, nr.ID())
 	}
 
+	var newRules []Rule
+	for _, r := range g.Rules {
+		if r == nil {
+			// skip nil rules
+			continue
+		}
+		newRules = append(newRules, r)
+	}
+	// add the rest of rules from registry
 	for _, nr := range rulesRegistry {
-		g.Rules = append(g.Rules, nr)
+		newRules = append(newRules, nr)
 	}
+	g.Rules = newRules
+	return nil
 }
 
 var (
@@ -96,28 +142,48 @@ var (
 )
 
 func (g *Group) close() {
-	if g.done == nil {
+	if g.doneCh == nil {
 		return
 	}
-	close(g.done)
-	<-g.finished
+	close(g.doneCh)
+	<-g.finishedCh
 }
 
-func (g *Group) start(ctx context.Context, interval time.Duration,
-	querier datasource.Querier, nr notifier.Notifier, rw *remotewrite.Client) {
-	logger.Infof("group %q started", g.Name)
-	t := time.NewTicker(interval)
+func (g *Group) start(ctx context.Context, querier datasource.Querier, nr notifier.Notifier, rw *remotewrite.Client) {
+	logger.Infof("group %q started with interval %v", g.Name, g.Interval)
+
+	var returnSeries bool
+	if rw != nil {
+		returnSeries = true
+	}
+
+	t := time.NewTicker(g.Interval)
 	defer t.Stop()
 	for {
 		select {
 		case <-ctx.Done():
 			logger.Infof("group %q: context cancelled", g.Name)
-			close(g.finished)
+			close(g.finishedCh)
 			return
-		case <-g.done:
+		case <-g.doneCh:
 			logger.Infof("group %q: received stop signal", g.Name)
-			close(g.finished)
+			close(g.finishedCh)
 			return
+		case ng := <-g.updateCh:
+			g.mu.Lock()
+			err := g.updateWith(ng)
+			if err != nil {
+				logger.Errorf("group %q: failed to update: %s", g.Name, err)
+				g.mu.Unlock()
+				continue
+			}
+			if g.Interval != ng.Interval {
+				g.Interval = ng.Interval
+				t.Stop()
+				t = time.NewTicker(g.Interval)
+				logger.Infof("group %q: changed evaluation interval to %v", g.Name, g.Interval)
+			}
+			g.mu.Unlock()
 		case <-t.C:
 			iterationTotal.Inc()
 			iterationStart := time.Now()
@@ -125,39 +191,53 @@ func (g *Group) start(ctx context.Context, interval time.Duration,
 				execTotal.Inc()
 
 				execStart := time.Now()
-				err := rule.Exec(ctx, querier)
+				tss, err := rule.Exec(ctx, querier, returnSeries)
 				execDuration.UpdateDuration(execStart)
 
 				if err != nil {
 					execErrors.Inc()
-					logger.Errorf("failed to execute rule %q.%q: %s", g.Name, rule.Name, err)
+					logger.Errorf("failed to execute rule %q.%q: %s", g.Name, rule, err)
 					continue
 				}
 
-				var alertsToSend []notifier.Alert
-				for _, a := range rule.alerts {
-					if a.State != notifier.StatePending {
-						alertsToSend = append(alertsToSend, *a)
-					}
-					if a.State == notifier.StateInactive || rw == nil {
-						continue
-					}
-					tss := rule.AlertToTimeSeries(a, execStart)
+				if len(tss) > 0 {
+					remoteWriteSent.Add(len(tss))
 					for _, ts := range tss {
-						remoteWriteSent.Inc()
 						if err := rw.Push(ts); err != nil {
 							remoteWriteErrors.Inc()
-							logger.Errorf("failed to push timeseries to remotewrite: %s", err)
+							logger.Errorf("failed to remote write for rule %q.%q: %s", g.Name, rule, err)
 						}
 					}
 				}
-				if len(alertsToSend) > 0 {
-					alertsSent.Add(len(alertsToSend))
-					if err := nr.Send(ctx, alertsToSend); err != nil {
-						alertsSendErrors.Inc()
-						logger.Errorf("failed to send alert for rule %q.%q: %s", g.Name, rule.Name, err)
+
+				ar, ok := rule.(*AlertingRule)
+				if !ok {
+					continue
+				}
+				var alerts []notifier.Alert
+				for _, a := range ar.alerts {
+					switch a.State {
+					case notifier.StateFiring:
+						// set End to execStart + 3 intervals
+						// so notifier can resolve it automatically if `vmalert`
+						// won't be able to send resolve for some reason
+						a.End = execStart.Add(3 * g.Interval)
+						alerts = append(alerts, *a)
+					case notifier.StateInactive:
+						// set End to execStart to notify
+						// that it was just resolved
+						a.End = execStart
+						alerts = append(alerts, *a)
 					}
 				}
+				if len(alerts) < 1 {
+					continue
+				}
+				alertsSent.Add(len(alerts))
+				if err := nr.Send(ctx, alerts); err != nil {
+					alertsSendErrors.Inc()
+					logger.Errorf("failed to send alert for rule %q.%q: %s", g.Name, rule, err)
+				}
 			}
 			iterationDuration.UpdateDuration(iterationStart)
 		}

app/vmalert/group_test.go

@@ -4,27 +4,28 @@ import (
 	"context"
 	"reflect"
 	"sort"
-	"sync"
 	"testing"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
 )
 
 func TestUpdateWith(t *testing.T) {
 	testCases := []struct {
 		name         string
-		currentRules []*Rule
-		newRules     []*Rule
+		currentRules []Rule
+		// rules must be sorted by ID
+		newRules []Rule
 	}{
 		{
 			"new rule",
-			[]*Rule{},
-			[]*Rule{{Name: "bar"}},
+			[]Rule{},
+			[]Rule{&AlertingRule{Name: "bar"}},
 		},
 		{
-			"update rule",
-			[]*Rule{{
+			"update alerting rule",
+			[]Rule{&AlertingRule{
 				Name: "foo",
 				Expr: "up > 0",
 				For:  time.Second,
@@ -36,8 +37,8 @@ func TestUpdateWith(t *testing.T) {
 					"description": "{{$labels}}",
 				},
 			}},
-			[]*Rule{{
-				Name: "bar",
+			[]Rule{&AlertingRule{
+				Name: "foo",
 				Expr: "up > 10",
 				For:  time.Second,
 				Labels: map[string]string{
@@ -48,43 +49,82 @@ func TestUpdateWith(t *testing.T) {
 				},
 			}},
 		},
+		{
+			"update recording rule",
+			[]Rule{&RecordingRule{
+				Name: "foo",
+				Expr: "max(up)",
+				Labels: map[string]string{
+					"bar": "baz",
+				},
+			}},
+			[]Rule{&RecordingRule{
+				Name: "foo",
+				Expr: "min(up)",
+				Labels: map[string]string{
+					"baz": "bar",
+				},
+			}},
+		},
 		{
 			"empty rule",
-			[]*Rule{{Name: "foo"}},
-			[]*Rule{},
+			[]Rule{&AlertingRule{Name: "foo"}, &RecordingRule{Name: "bar"}},
+			[]Rule{},
 		},
 		{
 			"multiple rules",
-			[]*Rule{{Name: "foo"}, {Name: "bar"}, {Name: "baz"}},
-			[]*Rule{{Name: "foo"}, {Name: "baz"}},
+			[]Rule{
+				&AlertingRule{Name: "bar"},
+				&AlertingRule{Name: "baz"},
+				&RecordingRule{Name: "foo"},
+			},
+			[]Rule{
+				&AlertingRule{Name: "baz"},
+				&RecordingRule{Name: "foo"},
+			},
+		},
+		{
+			"replace rule",
+			[]Rule{&AlertingRule{Name: "foo1"}},
+			[]Rule{&AlertingRule{Name: "foo2"}},
+		},
+		{
+			"replace multiple rules",
+			[]Rule{
+				&AlertingRule{Name: "foo1"},
+				&RecordingRule{Name: "foo2"},
+				&AlertingRule{Name: "foo3"},
+			},
+			[]Rule{
+				&AlertingRule{Name: "foo3"},
+				&AlertingRule{Name: "foo4"},
+				&RecordingRule{Name: "foo5"},
+			},
 		},
 	}
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
 			g := &Group{Rules: tc.currentRules}
-			g.updateWith(Group{Rules: tc.newRules})
+			err := g.updateWith(&Group{Rules: tc.newRules})
+			if err != nil {
+				t.Fatal(err)
+			}
 
 			if len(g.Rules) != len(tc.newRules) {
 				t.Fatalf("expected to have %d rules; got: %d",
 					len(g.Rules), len(tc.newRules))
 			}
+			sort.Slice(g.Rules, func(i, j int) bool {
+				return g.Rules[i].ID() < g.Rules[j].ID()
+			})
 			for i, r := range g.Rules {
 				got, want := r, tc.newRules[i]
-				if got.Name != want.Name {
-					t.Fatalf("expected to have rule %q; got %q", want.Name, got.Name)
+				if got.ID() != want.ID() {
+					t.Fatalf("expected to have rule %q; got %q", want, got)
 				}
-				if got.Expr != want.Expr {
-					t.Fatalf("expected to have expression %q; got %q", want.Expr, got.Expr)
-				}
-				if got.For != want.For {
-					t.Fatalf("expected to have for %q; got %q", want.For, got.For)
-				}
-				if !reflect.DeepEqual(got.Annotations, want.Annotations) {
-					t.Fatalf("expected to have annotations %#v; got %#v", want.Annotations, got.Annotations)
-				}
-				if !reflect.DeepEqual(got.Labels, want.Labels) {
-					t.Fatalf("expected to have labels %#v; got %#v", want.Labels, got.Labels)
+				if err := compareRules(t, got, want); err != nil {
+					t.Fatalf("comparsion error: %s", err)
 				}
 			}
 		})
@@ -93,11 +133,12 @@ func TestUpdateWith(t *testing.T) {
 
 func TestGroupStart(t *testing.T) {
 	// TODO: make parsing from string instead of file
-	groups, err := Parse([]string{"testdata/rules1-good.rules"}, true)
+	groups, err := config.Parse([]string{"config/testdata/rules1-good.rules"}, true)
 	if err != nil {
 		t.Fatalf("failed to parse rules: %s", err)
 	}
-	g := groups[0]
+	const evalInterval = time.Millisecond
+	g := newGroup(groups[0], evalInterval)
 
 	fn := &fakeNotifier{}
 	fs := &fakeQuerier{}
@@ -106,27 +147,26 @@ func TestGroupStart(t *testing.T) {
 	m1 := metricWithLabels(t, "instance", inst1, "job", job)
 	m2 := metricWithLabels(t, "instance", inst2, "job", job)
 
-	r := g.Rules[0]
-	alert1, err := r.newAlert(m1)
+	r := g.Rules[0].(*AlertingRule)
+	alert1, err := r.newAlert(m1, time.Now())
 	if err != nil {
 		t.Fatalf("faield to create alert: %s", err)
 	}
 	alert1.State = notifier.StateFiring
 	alert1.ID = hash(m1)
 
-	alert2, err := r.newAlert(m2)
+	alert2, err := r.newAlert(m2, time.Now())
 	if err != nil {
 		t.Fatalf("faield to create alert: %s", err)
 	}
 	alert2.State = notifier.StateFiring
 	alert2.ID = hash(m2)
 
-	const evalInterval = time.Millisecond
 	finished := make(chan struct{})
 	fs.add(m1)
 	fs.add(m2)
 	go func() {
-		g.start(context.Background(), evalInterval, fs, fn, nil)
+		g.start(context.Background(), fs, fn, nil)
 		close(finished)
 	}()
 
@@ -183,21 +223,3 @@ func compareAlerts(t *testing.T, as, bs []notifier.Alert) {
 		}
 	}
 }
-
-type fakeNotifier struct {
-	sync.Mutex
-	alerts []notifier.Alert
-}
-
-func (fn *fakeNotifier) Send(_ context.Context, alerts []notifier.Alert) error {
-	fn.Lock()
-	defer fn.Unlock()
-	fn.alerts = alerts
-	return nil
-}
-
-func (fn *fakeNotifier) getAlerts() []notifier.Alert {
-	fn.Lock()
-	defer fn.Unlock()
-	return fn.alerts
-}

app/vmalert/helpers_test.go

@@ -0,0 +1,200 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"reflect"
+	"sync"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+type fakeQuerier struct {
+	sync.Mutex
+	metrics []datasource.Metric
+	err     error
+}
+
+func (fq *fakeQuerier) setErr(err error) {
+	fq.Lock()
+	fq.err = err
+	fq.Unlock()
+}
+
+func (fq *fakeQuerier) reset() {
+	fq.Lock()
+	fq.err = nil
+	fq.metrics = fq.metrics[:0]
+	fq.Unlock()
+}
+
+func (fq *fakeQuerier) add(metrics ...datasource.Metric) {
+	fq.Lock()
+	fq.metrics = append(fq.metrics, metrics...)
+	fq.Unlock()
+}
+
+func (fq *fakeQuerier) Query(_ context.Context, _ string) ([]datasource.Metric, error) {
+	fq.Lock()
+	defer fq.Unlock()
+	if fq.err != nil {
+		return nil, fq.err
+	}
+	cp := make([]datasource.Metric, len(fq.metrics))
+	copy(cp, fq.metrics)
+	return cp, nil
+}
+
+type fakeNotifier struct {
+	sync.Mutex
+	alerts []notifier.Alert
+}
+
+func (fn *fakeNotifier) Send(_ context.Context, alerts []notifier.Alert) error {
+	fn.Lock()
+	defer fn.Unlock()
+	fn.alerts = alerts
+	return nil
+}
+
+func (fn *fakeNotifier) getAlerts() []notifier.Alert {
+	fn.Lock()
+	defer fn.Unlock()
+	return fn.alerts
+}
+
+func metricWithValueAndLabels(t *testing.T, value float64, labels ...string) datasource.Metric {
+	t.Helper()
+	m := metricWithLabels(t, labels...)
+	m.Value = value
+	return m
+}
+
+func metricWithLabels(t *testing.T, labels ...string) datasource.Metric {
+	t.Helper()
+	if len(labels) == 0 || len(labels)%2 != 0 {
+		t.Fatalf("expected to get even number of labels")
+	}
+	m := datasource.Metric{}
+	for i := 0; i < len(labels); i += 2 {
+		m.Labels = append(m.Labels, datasource.Label{
+			Name:  labels[i],
+			Value: labels[i+1],
+		})
+	}
+	return m
+}
+
+func compareGroups(t *testing.T, a, b *Group) {
+	t.Helper()
+	if a.Name != b.Name {
+		t.Fatalf("expected group name %q; got %q", a.Name, b.Name)
+	}
+	if a.File != b.File {
+		t.Fatalf("expected group %q file name %q; got %q", a.Name, a.File, b.File)
+	}
+	if a.Interval != b.Interval {
+		t.Fatalf("expected group %q interval %v; got %v", a.Name, a.Interval, b.Interval)
+	}
+	if len(a.Rules) != len(b.Rules) {
+		t.Fatalf("expected group %s to have %d rules; got: %d",
+			a.Name, len(a.Rules), len(b.Rules))
+	}
+	for i, r := range a.Rules {
+		got, want := r, b.Rules[i]
+		if a.ID() != b.ID() {
+			t.Fatalf("expected to have rule %q; got %q", want.ID(), got.ID())
+		}
+		if err := compareRules(t, want, got); err != nil {
+			t.Fatalf("comparsion error: %s", err)
+		}
+	}
+}
+
+func compareRules(t *testing.T, a, b Rule) error {
+	t.Helper()
+	switch v := a.(type) {
+	case *AlertingRule:
+		br, ok := b.(*AlertingRule)
+		if !ok {
+			return fmt.Errorf("rule %q supposed to be of type AlertingRule", b.ID())
+		}
+		return compareAlertingRules(t, v, br)
+	case *RecordingRule:
+		br, ok := b.(*RecordingRule)
+		if !ok {
+			return fmt.Errorf("rule %q supposed to be of type RecordingRule", b.ID())
+		}
+		return compareRecordingRules(t, v, br)
+	default:
+		return fmt.Errorf("unexpected rule type received %T", a)
+	}
+}
+
+func compareRecordingRules(t *testing.T, a, b *RecordingRule) error {
+	t.Helper()
+	if a.Expr != b.Expr {
+		return fmt.Errorf("expected to have expression %q; got %q", a.Expr, b.Expr)
+	}
+	if !reflect.DeepEqual(a.Labels, b.Labels) {
+		return fmt.Errorf("expected to have labels %#v; got %#v", a.Labels, b.Labels)
+	}
+	return nil
+}
+
+func compareAlertingRules(t *testing.T, a, b *AlertingRule) error {
+	t.Helper()
+	if a.Expr != b.Expr {
+		return fmt.Errorf("expected to have expression %q; got %q", a.Expr, b.Expr)
+	}
+	if a.For != b.For {
+		return fmt.Errorf("expected to have for %q; got %q", a.For, b.For)
+	}
+	if !reflect.DeepEqual(a.Annotations, b.Annotations) {
+		return fmt.Errorf("expected to have annotations %#v; got %#v", a.Annotations, b.Annotations)
+	}
+	if !reflect.DeepEqual(a.Labels, b.Labels) {
+		return fmt.Errorf("expected to have labels %#v; got %#v", a.Labels, b.Labels)
+	}
+	return nil
+}
+
+func compareTimeSeries(t *testing.T, a, b []prompbmarshal.TimeSeries) error {
+	t.Helper()
+	if len(a) != len(b) {
+		return fmt.Errorf("expected number of timeseries %d; got %d", len(a), len(b))
+	}
+	for i := range a {
+		expTS, gotTS := a[i], b[i]
+		if len(expTS.Samples) != len(gotTS.Samples) {
+			return fmt.Errorf("expected number of samples %d; got %d", len(expTS.Samples), len(gotTS.Samples))
+		}
+		for i, exp := range expTS.Samples {
+			got := gotTS.Samples[i]
+			if got.Value != exp.Value {
+				return fmt.Errorf("expected value %.2f; got %.2f", exp.Value, got.Value)
+			}
+			// timestamp validation isn't always correct for now.
+			// this must be improved with time mock.
+			/*if got.Timestamp != exp.Timestamp {
+				return fmt.Errorf("expected timestamp %d; got %d", exp.Timestamp, got.Timestamp)
+			}*/
+		}
+		if len(expTS.Labels) != len(gotTS.Labels) {
+			return fmt.Errorf("expected number of labels %d; got %d", len(expTS.Labels), len(gotTS.Labels))
+		}
+		for i, exp := range expTS.Labels {
+			got := gotTS.Labels[i]
+			if got.Name != exp.Name {
+				return fmt.Errorf("expected label name %q; got %q", exp.Name, got.Name)
+			}
+			if got.Value != exp.Value {
+				return fmt.Errorf("expected label value %q; got %q", exp.Value, got.Value)
+			}
+		}
+	}
+	return nil
+}

app/vmalert/main.go

@@ -15,6 +15,7 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/remotewrite"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/buildinfo"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/envflag"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/flagutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httpserver"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
@@ -37,25 +38,31 @@ absolute path to all .yaml files in root.`)
 	basicAuthUsername = flag.String("datasource.basicAuth.username", "", "Optional basic auth username for -datasource.url")
 	basicAuthPassword = flag.String("datasource.basicAuth.password", "", "Optional basic auth password for -datasource.url")
 
-	remoteWriteURL = flag.String("remotewrite.url", "", "Optional URL to Victoria Metrics or VMInsert where to persist alerts state"+
+	remoteWriteURL = flag.String("remoteWrite.url", "", "Optional URL to Victoria Metrics or VMInsert where to persist alerts state"+
 		" in form of timeseries. E.g. http://127.0.0.1:8428")
-	remoteWriteUsername = flag.String("remotewrite.basicAuth.username", "", "Optional basic auth username for -remotewrite.url")
-	remoteWritePassword = flag.String("remotewrite.basicAuth.password", "", "Optional basic auth password for -remotewrite.url")
+	remoteWriteUsername     = flag.String("remoteWrite.basicAuth.username", "", "Optional basic auth username for -remoteWrite.url")
+	remoteWritePassword     = flag.String("remoteWrite.basicAuth.password", "", "Optional basic auth password for -remoteWrite.url")
+	remoteWriteMaxQueueSize = flag.Int("remoteWrite.maxQueueSize", 1e5, "Defines the max number of pending datapoints to remote write endpoint")
+	remoteWriteMaxBatchSize = flag.Int("remoteWrite.maxBatchSize", 1e3, "Defines defines max number of timeseries to be flushed at once")
+	remoteWriteConcurrency  = flag.Int("remoteWrite.concurrency", 1, "Defines number of readers that concurrently write into remote storage")
 
-	remoteReadURL = flag.String("remoteread.url", "", "Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts"+
-		" state. This configuration makes sense only if `vmalert` was configured with `remotewrite.url` before and has been successfully persisted its state."+
+	remoteReadURL = flag.String("remoteRead.url", "", "Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts"+
+		" state. This configuration makes sense only if `vmalert` was configured with `remoteWrite.url` before and has been successfully persisted its state."+
 		" E.g. http://127.0.0.1:8428")
-	remoteReadUsername = flag.String("remoteread.basicAuth.username", "", "Optional basic auth username for -remoteread.url")
-	remoteReadPassword = flag.String("remoteread.basicAuth.password", "", "Optional basic auth password for -remoteread.url")
-	remoteReadLookBack = flag.Duration("remoteread.lookback", time.Hour, "Lookback defines how far to look into past for alerts timeseries."+
+	remoteReadUsername = flag.String("remoteRead.basicAuth.username", "", "Optional basic auth username for -remoteRead.url")
+	remoteReadPassword = flag.String("remoteRead.basicAuth.password", "", "Optional basic auth password for -remoteRead.url")
+	remoteReadLookBack = flag.Duration("remoteRead.lookback", time.Hour, "Lookback defines how far to look into past for alerts timeseries."+
 		" For example, if lookback=1h then range from now() to now()-1h will be scanned.")
 
-	evaluationInterval = flag.Duration("evaluationInterval", time.Minute, "How often to evaluate the rules. Default 1m")
+	evaluationInterval = flag.Duration("evaluationInterval", time.Minute, "How often to evaluate the rules")
 	notifierURL        = flag.String("notifier.url", "", "Prometheus alertmanager URL. Required parameter. e.g. http://127.0.0.1:9093")
 	externalURL        = flag.String("external.url", "", "External URL is used as alert's source for sent alerts to the notifier")
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
+	flag.Usage = usage
 	envflag.Parse()
 	buildinfo.Init()
 	logger.Init()
@@ -77,6 +84,9 @@ func main() {
 	if *remoteWriteURL != "" {
 		c, err := remotewrite.NewClient(ctx, remotewrite.Config{
 			Addr:          *remoteWriteURL,
+			Concurrency:   *remoteWriteConcurrency,
+			MaxQueueSize:  *remoteWriteMaxQueueSize,
+			MaxBatchSize:  *remoteWriteMaxBatchSize,
 			FlushInterval: *evaluationInterval,
 			BasicAuthUser: *remoteWriteUsername,
 			BasicAuthPass: *remoteWritePassword,
@@ -97,7 +107,7 @@ func main() {
 	go func() {
 		// init reload metrics with positive values to improve alerting conditions
 		configSuccess.Set(1)
-		configTimestamp.Set(uint64(time.Now().UnixNano()) / 1e9)
+		configTimestamp.Set(fasttime.UnixTimestamp())
 		sigHup := procutil.NewSighupChan()
 		for {
 			<-sigHup
@@ -110,13 +120,13 @@ func main() {
 				continue
 			}
 			configSuccess.Set(1)
-			configTimestamp.Set(uint64(time.Now().UnixNano()) / 1e9)
+			configTimestamp.Set(fasttime.UnixTimestamp())
 			logger.Infof("Rules reloaded successfully from %q", *rulePath)
 		}
 	}()
 
 	rh := &requestHandler{m: manager}
-	go httpserver.Serve(*httpListenAddr, (rh).handler)
+	go httpserver.Serve(*httpListenAddr, rh.handler)
 
 	sig := procutil.WaitForSigterm()
 	logger.Infof("service received signal %s", sig)
@@ -163,3 +173,15 @@ func checkFlags() {
 		logger.Fatalf("datasource.url is empty")
 	}
 }
+
+func usage() {
+	const s = `
+vmalert processes alerts and recording rules.
+
+See the docs at https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/README.md .
+`
+
+	f := flag.CommandLine.Output()
+	fmt.Fprintf(f, "%s\n", s)
+	flag.PrintDefaults()
+}

app/vmalert/manager.go

@@ -6,12 +6,14 @@ import (
 	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/remotewrite"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 )
 
+// manager controls group states
 type manager struct {
 	storage  datasource.Querier
 	notifier notifier.Notifier
@@ -25,7 +27,7 @@ type manager struct {
 	groups   map[uint64]*Group
 }
 
-// AlertAPI generates APIAlert object from alert by its id(hash)
+// AlertAPI generates APIAlert object from alert by its ID(hash)
 func (m *manager) AlertAPI(gID, aID uint64) (*APIAlert, error) {
 	m.groupsMu.RLock()
 	defer m.groupsMu.RUnlock()
@@ -35,11 +37,15 @@ func (m *manager) AlertAPI(gID, aID uint64) (*APIAlert, error) {
 		return nil, fmt.Errorf("can't find group with id %q", gID)
 	}
 	for _, rule := range g.Rules {
-		if apiAlert := rule.AlertAPI(aID); apiAlert != nil {
+		ar, ok := rule.(*AlertingRule)
+		if !ok {
+			continue
+		}
+		if apiAlert := ar.AlertAPI(aID); apiAlert != nil {
 			return apiAlert, nil
 		}
 	}
-	return nil, fmt.Errorf("can't func alert with id %q in group %q", aID, g.Name)
+	return nil, fmt.Errorf("can't find alert with id %q in group %q", aID, g.Name)
 }
 
 func (m *manager) start(ctx context.Context, path []string, validate bool) error {
@@ -56,8 +62,8 @@ func (m *manager) close() {
 	m.wg.Wait()
 }
 
-func (m *manager) startGroup(ctx context.Context, group Group, restore bool) {
-	if restore {
+func (m *manager) startGroup(ctx context.Context, group *Group, restore bool) {
+	if restore && m.rr != nil {
 		err := group.Restore(ctx, m.rr, *remoteReadLookBack)
 		if err != nil {
 			logger.Errorf("error while restoring state for group %q: %s", group.Name, err)
@@ -67,28 +73,28 @@ func (m *manager) startGroup(ctx context.Context, group Group, restore bool) {
 	m.wg.Add(1)
 	id := group.ID()
 	go func() {
-		group.start(ctx, *evaluationInterval, m.storage, m.notifier, m.rw)
+		group.start(ctx, m.storage, m.notifier, m.rw)
 		m.wg.Done()
 	}()
-	m.groups[id] = &group
+	m.groups[id] = group
 }
 
 func (m *manager) update(ctx context.Context, path []string, validate, restore bool) error {
-	logger.Infof("reading alert rules configuration file from %q", strings.Join(path, ";"))
-	newGroups, err := Parse(path, validate)
+	logger.Infof("reading rules configuration file from %q", strings.Join(path, ";"))
+	groupsCfg, err := config.Parse(path, validate)
 	if err != nil {
 		return fmt.Errorf("cannot parse configuration file: %s", err)
 	}
 
-	groupsRegistry := make(map[uint64]Group)
-	for _, ng := range newGroups {
+	groupsRegistry := make(map[uint64]*Group)
+	for _, cfg := range groupsCfg {
+		ng := newGroup(cfg, *evaluationInterval)
 		groupsRegistry[ng.ID()] = ng
 	}
 
 	m.groupsMu.Lock()
 	for _, og := range m.groups {
-		id := og.ID()
-		ng, ok := groupsRegistry[id]
+		ng, ok := groupsRegistry[og.ID()]
 		if !ok {
 			// old group is not present in new list
 			// and must be stopped and deleted
@@ -97,7 +103,7 @@ func (m *manager) update(ctx context.Context, path []string, validate, restore b
 			og = nil
 			continue
 		}
-		og.updateWith(ng)
+		og.updateCh <- ng
 		delete(groupsRegistry, ng.ID())
 	}
 
@@ -107,3 +113,22 @@ func (m *manager) update(ctx context.Context, path []string, validate, restore b
 	m.groupsMu.Unlock()
 	return nil
 }
+
+func (g *Group) toAPI() APIGroup {
+	ag := APIGroup{
+		// encode as strings to avoid rounding
+		ID:       fmt.Sprintf("%d", g.ID()),
+		Name:     g.Name,
+		File:     g.File,
+		Interval: g.Interval.String(),
+	}
+	for _, r := range g.Rules {
+		switch v := r.(type) {
+		case *AlertingRule:
+			ag.AlertingRules = append(ag.AlertingRules, v.RuleAPI())
+		case *RecordingRule:
+			ag.RecordingRules = append(ag.RecordingRules, v.RuleAPI())
+		}
+	}
+	return ag
+}

app/vmalert/manager_test.go

@@ -3,12 +3,22 @@ package main
 import (
 	"context"
 	"math/rand"
+	"net/url"
+	"os"
 	"strings"
 	"sync"
 	"testing"
 	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
 )
 
+func TestMain(m *testing.M) {
+	u, _ := url.Parse("https://victoriametrics.com/path")
+	notifier.InitTemplateFunc(u)
+	os.Exit(m.Run())
+}
+
 func TestManagerUpdateError(t *testing.T) {
 	m := &manager{groups: make(map[uint64]*Group)}
 	path := []string{"foo/bar"}
@@ -26,24 +36,36 @@ func TestManagerUpdateError(t *testing.T) {
 // execution of configuration update.
 // Should be executed with -race flag
 func TestManagerUpdateConcurrent(t *testing.T) {
-	m := &manager{groups: make(map[uint64]*Group)}
+	m := &manager{
+		groups:   make(map[uint64]*Group),
+		storage:  &fakeQuerier{},
+		notifier: &fakeNotifier{},
+	}
 	paths := []string{
-		"testdata/dir/rules0-good.rules",
-		"testdata/dir/rules1-good.rules",
-		"testdata/rules0-good.rules",
+		"config/testdata/dir/rules0-good.rules",
+		"config/testdata/dir/rules0-bad.rules",
+		"config/testdata/dir/rules1-good.rules",
+		"config/testdata/dir/rules1-bad.rules",
+		"config/testdata/rules0-good.rules",
+		"config/testdata/rules1-good.rules",
+		"config/testdata/rules2-good.rules",
+	}
+	*evaluationInterval = time.Millisecond
+	if err := m.start(context.Background(), []string{paths[0]}, true); err != nil {
+		t.Fatalf("failed to start: %s", err)
 	}
 
-	const n = 500
+	const workers = 500
+	const iterations = 10
 	wg := sync.WaitGroup{}
-	wg.Add(n)
-	for i := 0; i < n; i++ {
+	wg.Add(workers)
+	for i := 0; i < workers; i++ {
 		go func() {
 			defer wg.Done()
-			rnd := rand.Intn(len(paths))
-			path := []string{paths[rnd]}
-			err := m.update(context.Background(), path, true, false)
-			if err != nil {
-				t.Errorf("update error: %s", err)
+			for i := 0; i < iterations; i++ {
+				rnd := rand.Intn(len(paths))
+				path := []string{paths[rnd]}
+				_ = m.update(context.Background(), path, true, false)
 			}
 		}()
 	}
@@ -53,6 +75,41 @@ func TestManagerUpdateConcurrent(t *testing.T) {
 // TestManagerUpdate tests sequential configuration
 // updates.
 func TestManagerUpdate(t *testing.T) {
+	const defaultEvalInterval = time.Second * 30
+	currentEvalInterval := *evaluationInterval
+	*evaluationInterval = defaultEvalInterval
+	defer func() {
+		*evaluationInterval = currentEvalInterval
+	}()
+
+	var (
+		VMRows = &AlertingRule{
+			Name: "VMRows",
+			Expr: "vm_rows > 0",
+			For:  10 * time.Second,
+			Labels: map[string]string{
+				"label": "bar",
+				"host":  "{{ $labels.instance }}",
+			},
+			Annotations: map[string]string{
+				"summary":     "{{ $value|humanize }}",
+				"description": "{{$labels}}",
+			},
+		}
+		Conns = &AlertingRule{
+			Name: "Conns",
+			Expr: "sum(vm_tcplistener_conns) by(instance) > 1",
+			Annotations: map[string]string{
+				"summary":     "Too high connection number for {{$labels.instance}}",
+				"description": "It is {{ $value }} connections for {{$labels.instance}}",
+			},
+		}
+		ExampleAlertAlwaysFiring = &AlertingRule{
+			Name: "ExampleAlertAlwaysFiring",
+			Expr: "sum by(job) (up == 1)",
+		}
+	)
+
 	testCases := []struct {
 		name       string
 		initPath   string
@@ -61,49 +118,65 @@ func TestManagerUpdate(t *testing.T) {
 	}{
 		{
 			name:       "update good rules",
-			initPath:   "testdata/rules0-good.rules",
-			updatePath: "testdata/dir/rules1-good.rules",
+			initPath:   "config/testdata/rules0-good.rules",
+			updatePath: "config/testdata/dir/rules1-good.rules",
 			want: []*Group{
 				{
-					File:  "testdata/dir/rules1-good.rules",
-					Name:  "duplicatedGroupDiffFiles",
-					Rules: []*Rule{newTestRule("VMRows", time.Second*10)},
+					File:     "config/testdata/dir/rules1-good.rules",
+					Name:     "duplicatedGroupDiffFiles",
+					Interval: defaultEvalInterval,
+					Rules: []Rule{
+						&AlertingRule{
+							Name:   "VMRows",
+							Expr:   "vm_rows > 0",
+							For:    5 * time.Minute,
+							Labels: map[string]string{"label": "bar"},
+							Annotations: map[string]string{
+								"summary":     "{{ $value }}",
+								"description": "{{$labels}}",
+							},
+						},
+					},
 				},
 			},
 		},
 		{
 			name:       "update good rules from 1 to 2 groups",
-			initPath:   "testdata/dir/rules1-good.rules",
-			updatePath: "testdata/rules0-good.rules",
+			initPath:   "config/testdata/dir/rules1-good.rules",
+			updatePath: "config/testdata/rules0-good.rules",
 			want: []*Group{
 				{
-					File: "testdata/rules0-good.rules",
-					Name: "groupGorSingleAlert", Rules: []*Rule{
-						newTestRule("VMRows", time.Second*10),
-					}},
+					File:     "config/testdata/rules0-good.rules",
+					Name:     "groupGorSingleAlert",
+					Rules:    []Rule{VMRows},
+					Interval: defaultEvalInterval,
+				},
 				{
-					File: "testdata/rules0-good.rules",
-					Name: "TestGroup", Rules: []*Rule{
-						newTestRule("Conns", time.Duration(0)),
-						newTestRule("ExampleAlertAlwaysFiring", time.Duration(0)),
+					File:     "config/testdata/rules0-good.rules",
+					Interval: defaultEvalInterval,
+					Name:     "TestGroup", Rules: []Rule{
+						Conns,
+						ExampleAlertAlwaysFiring,
 					}},
 			},
 		},
 		{
 			name:       "update with one bad rule file",
-			initPath:   "testdata/rules0-good.rules",
-			updatePath: "testdata/dir/rules2-bad.rules",
+			initPath:   "config/testdata/rules0-good.rules",
+			updatePath: "config/testdata/dir/rules2-bad.rules",
 			want: []*Group{
 				{
-					File: "testdata/rules0-good.rules",
-					Name: "groupGorSingleAlert", Rules: []*Rule{
-						newTestRule("VMRows", time.Second*10),
-					}},
+					File:     "config/testdata/rules0-good.rules",
+					Name:     "groupGorSingleAlert",
+					Interval: defaultEvalInterval,
+					Rules:    []Rule{VMRows},
+				},
 				{
-					File: "testdata/rules0-good.rules",
-					Name: "TestGroup", Rules: []*Rule{
-						newTestRule("Conns", time.Duration(0)),
-						newTestRule("ExampleAlertAlwaysFiring", time.Duration(0)),
+					File:     "config/testdata/rules0-good.rules",
+					Interval: defaultEvalInterval,
+					Name:     "TestGroup", Rules: []Rule{
+						Conns,
+						ExampleAlertAlwaysFiring,
 					}},
 			},
 		},
@@ -111,7 +184,7 @@ func TestManagerUpdate(t *testing.T) {
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.TODO())
-			m := &manager{groups: make(map[uint64]*Group)}
+			m := &manager{groups: make(map[uint64]*Group), storage: &fakeQuerier{}}
 			path := []string{tc.initPath}
 			if err := m.update(ctx, path, true, false); err != nil {
 				t.Fatalf("failed to complete initial rules update: %s", err)
@@ -128,7 +201,7 @@ func TestManagerUpdate(t *testing.T) {
 				if !ok {
 					t.Fatalf("expected to have group %q", wantG.Name)
 				}
-				compareGroups(t, gotG, wantG)
+				compareGroups(t, wantG, gotG)
 			}
 
 			cancel()
@@ -136,17 +209,3 @@ func TestManagerUpdate(t *testing.T) {
 		})
 	}
 }
-
-func compareGroups(t *testing.T, a, b *Group) {
-	t.Helper()
-	if len(a.Rules) != len(b.Rules) {
-		t.Fatalf("expected group %s to have %d rules; got: %d",
-			a.Name, len(a.Rules), len(b.Rules))
-	}
-	for i, r := range a.Rules {
-		got, want := r, b.Rules[i]
-		if got.Name != want.Name {
-			t.Fatalf("expected to have rule %q; got %q", want.Name, got.Name)
-		}
-	}
-}

app/vmalert/notifier/alert.go

@@ -18,6 +18,7 @@ type Alert struct {
 	Annotations map[string]string
 	State       AlertState
 
+	Expr  string
 	Start time.Time
 	End   time.Time
 	Value float64
@@ -52,14 +53,15 @@ func (as AlertState) String() string {
 type alertTplData struct {
 	Labels map[string]string
 	Value  float64
+	Expr   string
 }
 
-const tplHeader = `{{ $value := .Value }}{{ $labels := .Labels }}`
+const tplHeader = `{{ $value := .Value }}{{ $labels := .Labels }}{{ $expr := .Expr }}`
 
 // ExecTemplate executes the Alert template for give
 // map of annotations.
 func (a *Alert) ExecTemplate(annotations map[string]string) (map[string]string, error) {
-	tplData := alertTplData{Value: a.Value, Labels: a.Labels}
+	tplData := alertTplData{Value: a.Value, Labels: a.Labels, Expr: a.Expr}
 	return templateAnnotations(annotations, tplHeader, tplData)
 }
 
@@ -85,7 +87,7 @@ func templateAnnotations(annotations map[string]string, header string, data aler
 		builder.WriteString(header)
 		builder.WriteString(text)
 		if err := templateAnnotation(&buf, builder.String(), data); err != nil {
-			eg.errs = append(eg.errs, fmt.Sprintf("key %s, template %s:%s", key, text, err))
+			eg.errs = append(eg.errs, fmt.Sprintf("key %q, template %q: %s", key, text, err))
 			continue
 		}
 		r[key] = buf.String()
@@ -96,10 +98,10 @@ func templateAnnotations(annotations map[string]string, header string, data aler
 func templateAnnotation(dst io.Writer, text string, data alertTplData) error {
 	tpl, err := template.New("").Funcs(tmplFunc).Option("missingkey=zero").Parse(text)
 	if err != nil {
-		return fmt.Errorf("error parsing annotation:%w", err)
+		return fmt.Errorf("error parsing annotation: %w", err)
 	}
 	if err = tpl.Execute(dst, data); err != nil {
-		return fmt.Errorf("error evaluating annotation template:%w", err)
+		return fmt.Errorf("error evaluating annotation template: %w", err)
 	}
 	return nil
 }

app/vmalert/notifier/alert_test.go

@@ -1,22 +1,27 @@
 package notifier
 
 import (
-	"fmt"
+	"net/url"
 	"testing"
 )
 
 func TestAlert_ExecTemplate(t *testing.T) {
+	u, _ := url.Parse("https://victoriametrics.com/path")
+	InitTemplateFunc(u)
 	testCases := []struct {
+		name        string
 		alert       *Alert
 		annotations map[string]string
 		expTpl      map[string]string
 	}{
 		{
+			name:        "empty-alert",
 			alert:       &Alert{},
 			annotations: map[string]string{},
 			expTpl:      map[string]string{},
 		},
 		{
+			name: "no-template",
 			alert: &Alert{
 				Value: 1e4,
 				Labels: map[string]string{
@@ -27,6 +32,7 @@ func TestAlert_ExecTemplate(t *testing.T) {
 			expTpl:      map[string]string{},
 		},
 		{
+			name: "label-template",
 			alert: &Alert{
 				Value: 1e4,
 				Labels: map[string]string{
@@ -43,10 +49,24 @@ func TestAlert_ExecTemplate(t *testing.T) {
 				"description": "It is 10000 connections for localhost",
 			},
 		},
+		{
+			name: "expression-template",
+			alert: &Alert{
+				Expr: `vm_rows{"label"="bar"}>0`,
+			},
+			annotations: map[string]string{
+				"exprEscapedQuery": "{{ $expr|quotesEscape|queryEscape }}",
+				"exprEscapedPath":  "{{ $expr|quotesEscape|pathEscape }}",
+			},
+			expTpl: map[string]string{
+				"exprEscapedQuery": "vm_rows%7B%5C%22label%5C%22%3D%5C%22bar%5C%22%7D%3E0",
+				"exprEscapedPath":  "vm_rows%7B%5C%22label%5C%22=%5C%22bar%5C%22%7D%3E0",
+			},
+		},
 	}
 
-	for i, tc := range testCases {
-		t.Run(fmt.Sprintf("%d", i), func(t *testing.T) {
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
 			tpl, err := tc.alert.ExecTemplate(tc.annotations)
 			if err != nil {
 				t.Fatal(err)

app/vmalert/notifier/template_func.go

@@ -142,6 +142,15 @@ func InitTemplateFunc(externalURL *url.URL) {
 		"externalURL": func() string {
 			return externalURL.String()
 		},
+		"pathEscape": func(u string) string {
+			return url.PathEscape(u)
+		},
+		"queryEscape": func(q string) string {
+			return url.QueryEscape(q)
+		},
+		"quotesEscape": func(q string) string {
+			return strings.Replace(q, `"`, `\"`, -1)
+		},
 	}
 }
 

app/vmalert/notifier/utils.go

@@ -17,5 +17,5 @@ func (eg *errGroup) err() error {
 }
 
 func (eg *errGroup) Error() string {
-	return fmt.Sprintf("errors:%s", strings.Join(eg.errs, "\n"))
+	return fmt.Sprintf("errors: %s", strings.Join(eg.errs, "\n"))
 }

app/vmalert/recording.go

@@ -0,0 +1,151 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"hash/fnv"
+	"sort"
+	"sync"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/config"
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+// RecordingRule is a Rule that supposed
+// to evaluate configured Expression and
+// return TimeSeries as result.
+type RecordingRule struct {
+	Name    string
+	Expr    string
+	Labels  map[string]string
+	GroupID uint64
+
+	// guard status fields
+	mu sync.RWMutex
+	// stores last moment of time Exec was called
+	lastExecTime time.Time
+	// stores last error that happened in Exec func
+	// resets on every successful Exec
+	// may be used as Health state
+	lastExecError error
+}
+
+// String implements Stringer interface
+func (rr *RecordingRule) String() string {
+	return rr.Name
+}
+
+// ID returns unique Rule ID
+// within the parent Group.
+func (rr *RecordingRule) ID() uint64 {
+	hash := fnv.New64a()
+	hash.Write([]byte("alerting"))
+	hash.Write([]byte("\xff"))
+	hash.Write([]byte(rr.Name))
+	return hash.Sum64()
+}
+
+func newRecordingRule(gID uint64, cfg config.Rule) *RecordingRule {
+	return &RecordingRule{
+		Name:    cfg.Record,
+		Expr:    cfg.Expr,
+		Labels:  cfg.Labels,
+		GroupID: gID,
+	}
+}
+
+var errDuplicate = errors.New("result contains metrics with the same labelset after applying rule labels")
+
+// Exec executes RecordingRule expression via the given Querier.
+func (rr *RecordingRule) Exec(ctx context.Context, q datasource.Querier, series bool) ([]prompbmarshal.TimeSeries, error) {
+	if !series {
+		return nil, nil
+	}
+
+	qMetrics, err := q.Query(ctx, rr.Expr)
+
+	rr.mu.Lock()
+	defer rr.mu.Unlock()
+
+	rr.lastExecTime = time.Now()
+	rr.lastExecError = err
+	if err != nil {
+		return nil, fmt.Errorf("failed to execute query %q: %s", rr.Expr, err)
+	}
+
+	duplicates := make(map[uint64]prompbmarshal.TimeSeries, len(qMetrics))
+	var tss []prompbmarshal.TimeSeries
+	for _, r := range qMetrics {
+		ts := rr.toTimeSeries(r, rr.lastExecTime)
+		h := hashTimeSeries(ts)
+		if _, ok := duplicates[h]; ok {
+			rr.lastExecError = errDuplicate
+			return nil, errDuplicate
+		}
+		duplicates[h] = ts
+		tss = append(tss, ts)
+	}
+	return tss, nil
+}
+
+func hashTimeSeries(ts prompbmarshal.TimeSeries) uint64 {
+	hash := fnv.New64a()
+	labels := ts.Labels
+	sort.Slice(labels, func(i, j int) bool {
+		return labels[i].Name < labels[j].Name
+	})
+	for _, l := range labels {
+		hash.Write([]byte(l.Name))
+		hash.Write([]byte(l.Value))
+		hash.Write([]byte("\xff"))
+	}
+	return hash.Sum64()
+}
+
+func (rr *RecordingRule) toTimeSeries(m datasource.Metric, timestamp time.Time) prompbmarshal.TimeSeries {
+	labels := make(map[string]string)
+	for _, l := range m.Labels {
+		labels[l.Name] = l.Value
+	}
+	labels["__name__"] = rr.Name
+	// override existing labels with configured ones
+	for k, v := range rr.Labels {
+		labels[k] = v
+	}
+	return newTimeSeries(m.Value, labels, timestamp)
+}
+
+// UpdateWith copies all significant fields.
+// alerts state isn't copied since
+// it should be updated in next 2 Execs
+func (rr *RecordingRule) UpdateWith(r Rule) error {
+	nr, ok := r.(*RecordingRule)
+	if !ok {
+		return fmt.Errorf("BUG: attempt to update recroding rule with wrong type %#v", r)
+	}
+	rr.Expr = nr.Expr
+	rr.Labels = nr.Labels
+	return nil
+}
+
+// RuleAPI returns Rule representation in form
+// of APIRecordingRule
+func (rr *RecordingRule) RuleAPI() APIRecordingRule {
+	var lastErr string
+	if rr.lastExecError != nil {
+		lastErr = rr.lastExecError.Error()
+	}
+	return APIRecordingRule{
+		// encode as strings to avoid rounding
+		ID:         fmt.Sprintf("%d", rr.ID()),
+		GroupID:    fmt.Sprintf("%d", rr.GroupID),
+		Name:       rr.Name,
+		Expression: rr.Expr,
+		LastError:  lastErr,
+		LastExec:   rr.lastExecTime,
+		Labels:     rr.Labels,
+	}
+}

app/vmalert/recording_test.go

@@ -0,0 +1,121 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+func TestRecoridngRule_ToTimeSeries(t *testing.T) {
+	timestamp := time.Now()
+	testCases := []struct {
+		rule    *RecordingRule
+		metrics []datasource.Metric
+		expTS   []prompbmarshal.TimeSeries
+	}{
+		{
+			&RecordingRule{Name: "foo"},
+			[]datasource.Metric{metricWithValueAndLabels(t, 10,
+				"__name__", "bar",
+			)},
+			[]prompbmarshal.TimeSeries{
+				newTimeSeries(10, map[string]string{
+					"__name__": "foo",
+				}, timestamp),
+			},
+		},
+		{
+			&RecordingRule{Name: "foobarbaz"},
+			[]datasource.Metric{
+				metricWithValueAndLabels(t, 1, "__name__", "foo", "job", "foo"),
+				metricWithValueAndLabels(t, 2, "__name__", "bar", "job", "bar"),
+				metricWithValueAndLabels(t, 3, "__name__", "baz", "job", "baz"),
+			},
+			[]prompbmarshal.TimeSeries{
+				newTimeSeries(1, map[string]string{
+					"__name__": "foobarbaz",
+					"job":      "foo",
+				}, timestamp),
+				newTimeSeries(2, map[string]string{
+					"__name__": "foobarbaz",
+					"job":      "bar",
+				}, timestamp),
+				newTimeSeries(3, map[string]string{
+					"__name__": "foobarbaz",
+					"job":      "baz",
+				}, timestamp),
+			},
+		},
+		{
+			&RecordingRule{Name: "job:foo", Labels: map[string]string{
+				"source": "test",
+			}},
+			[]datasource.Metric{
+				metricWithValueAndLabels(t, 2, "__name__", "foo", "job", "foo"),
+				metricWithValueAndLabels(t, 1, "__name__", "bar", "job", "bar")},
+			[]prompbmarshal.TimeSeries{
+				newTimeSeries(2, map[string]string{
+					"__name__": "job:foo",
+					"job":      "foo",
+					"source":   "test",
+				}, timestamp),
+				newTimeSeries(1, map[string]string{
+					"__name__": "job:foo",
+					"job":      "bar",
+					"source":   "test",
+				}, timestamp),
+			},
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.rule.Name, func(t *testing.T) {
+			fq := &fakeQuerier{}
+			fq.add(tc.metrics...)
+			tss, err := tc.rule.Exec(context.TODO(), fq, true)
+			if err != nil {
+				t.Fatalf("unexpected Exec err: %s", err)
+			}
+			if err := compareTimeSeries(t, tc.expTS, tss); err != nil {
+				t.Fatalf("timeseries missmatch: %s", err)
+			}
+		})
+	}
+}
+
+func TestRecoridngRule_ToTimeSeriesNegative(t *testing.T) {
+	rr := &RecordingRule{Name: "job:foo", Labels: map[string]string{
+		"job": "test",
+	}}
+
+	fq := &fakeQuerier{}
+	expErr := "connection reset by peer"
+	fq.setErr(errors.New(expErr))
+
+	_, err := rr.Exec(context.TODO(), fq, true)
+	if err == nil {
+		t.Fatalf("expected to get err; got nil")
+	}
+	if !strings.Contains(err.Error(), expErr) {
+		t.Fatalf("expected to get err %q; got %q insterad", expErr, err)
+	}
+
+	fq.reset()
+
+	// add metrics which differs only by `job` label
+	// which will be overridden by rule
+	fq.add(metricWithValueAndLabels(t, 1, "__name__", "foo", "job", "foo"))
+	fq.add(metricWithValueAndLabels(t, 2, "__name__", "foo", "job", "bar"))
+
+	_, err = rr.Exec(context.TODO(), fq, true)
+	if err == nil {
+		t.Fatalf("expected to get err; got nil")
+	}
+	if !strings.Contains(err.Error(), errDuplicate.Error()) {
+		t.Fatalf("expected to get err %q; got %q insterad", errDuplicate, err)
+	}
+}

app/vmalert/remotewrite/remotewrite.go

@@ -38,11 +38,15 @@ type Config struct {
 	BasicAuthUser string
 	BasicAuthPass string
 
+	// Concurrency defines number of readers that
+	// concurrently read from the queue and flush data
+	Concurrency int
 	// MaxBatchSize defines max number of timeseries
 	// to be flushed at once
 	MaxBatchSize int
 	// MaxQueueSize defines max length of input queue
-	// populated by Push method
+	// populated by Push method.
+	// Push will be rejected once queue is full.
 	MaxQueueSize int
 	// FlushInterval defines time interval for flushing batches
 	FlushInterval time.Duration
@@ -52,9 +56,10 @@ type Config struct {
 }
 
 const (
+	defaultConcurrency   = 4
 	defaultMaxBatchSize  = 1e3
-	defaultMaxQueueSize  = 100
-	defaultFlushInterval = 5 * time.Second
+	defaultMaxQueueSize  = 1e5
+	defaultFlushInterval = time.Second
 	defaultWriteTimeout  = 30 * time.Second
 )
 
@@ -90,7 +95,13 @@ func NewClient(ctx context.Context, cfg Config) (*Client, error) {
 		doneCh:        make(chan struct{}),
 		input:         make(chan prompbmarshal.TimeSeries, cfg.MaxQueueSize),
 	}
-	c.run(ctx)
+	cc := defaultConcurrency
+	if cfg.Concurrency > 0 {
+		cc = cfg.Concurrency
+	}
+	for i := 0; i < cc; i++ {
+		c.run(ctx)
+	}
 	return c, nil
 }
 
@@ -103,7 +114,8 @@ func (c *Client) Push(s prompbmarshal.TimeSeries) error {
 	case c.input <- s:
 		return nil
 	default:
-		return fmt.Errorf("failed to push timeseries - queue is full (%d entries)",
+		return fmt.Errorf("failed to push timeseries - queue is full (%d entries). "+
+			"Queue size is controlled by -remoteWrite.maxQueueSize flag",
 			c.maxQueueSize)
 	}
 }
@@ -127,7 +139,10 @@ func (c *Client) run(ctx context.Context) {
 		for ts := range c.input {
 			wr.Timeseries = append(wr.Timeseries, ts)
 		}
-		lastCtx, cancel := context.WithTimeout(context.Background(), time.Second*10)
+		if len(wr.Timeseries) < 1 {
+			return
+		}
+		lastCtx, cancel := context.WithTimeout(context.Background(), defaultWriteTimeout)
 		c.flush(lastCtx, wr)
 		cancel()
 	}

app/vmalert/rule.go

@@ -2,334 +2,23 @@ package main
 
 import (
 	"context"
-	"errors"
-	"fmt"
-	"hash/fnv"
-	"sort"
-	"strconv"
-	"sync"
-	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/datasource"
-	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmalert/notifier"
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
-	"github.com/VictoriaMetrics/metricsql"
 )
 
-// Rule is basic alert entity
-type Rule struct {
-	Name        string            `yaml:"alert"`
-	Expr        string            `yaml:"expr"`
-	For         time.Duration     `yaml:"for"`
-	Labels      map[string]string `yaml:"labels"`
-	Annotations map[string]string `yaml:"annotations"`
-
-	group Group
-
-	// guard status fields
-	mu sync.RWMutex
-	// stores list of active alerts
-	alerts map[uint64]*notifier.Alert
-	// stores last moment of time Exec was called
-	lastExecTime time.Time
-	// stores last error that happened in Exec func
-	// resets on every successful Exec
-	// may be used as Health state
-	lastExecError error
-}
-
-func (r *Rule) id() string {
-	return r.Name
-}
-
-// Validate validates rule
-func (r *Rule) Validate() error {
-	if r.Name == "" {
-		return errors.New("rule name can not be empty")
-	}
-	if r.Expr == "" {
-		return fmt.Errorf("expression for rule %q can't be empty", r.Name)
-	}
-	if _, err := metricsql.Parse(r.Expr); err != nil {
-		return fmt.Errorf("invalid expression for rule %q: %w", r.Name, err)
-	}
-	return nil
-}
-
-// Exec executes Rule expression via the given Querier.
-// Based on the Querier results Rule maintains notifier.Alerts
-func (r *Rule) Exec(ctx context.Context, q datasource.Querier) error {
-	qMetrics, err := q.Query(ctx, r.Expr)
-	r.mu.Lock()
-	defer r.mu.Unlock()
-
-	r.lastExecError = err
-	r.lastExecTime = time.Now()
-	if err != nil {
-		return fmt.Errorf("failed to execute query %q: %s", r.Expr, err)
-	}
-
-	for h, a := range r.alerts {
-		// cleanup inactive alerts from previous Eval
-		if a.State == notifier.StateInactive {
-			delete(r.alerts, h)
-		}
-	}
-
-	updated := make(map[uint64]struct{})
-	// update list of active alerts
-	for _, m := range qMetrics {
-		h := hash(m)
-		updated[h] = struct{}{}
-		if a, ok := r.alerts[h]; ok {
-			if a.Value != m.Value {
-				// update Value field with latest value
-				a.Value = m.Value
-				// and re-exec template since Value can be used
-				// in templates
-				err = r.template(a)
-				if err != nil {
-					return err
-				}
-			}
-			continue
-		}
-		a, err := r.newAlert(m)
-		if err != nil {
-			r.lastExecError = err
-			return fmt.Errorf("failed to create alert: %s", err)
-		}
-		a.ID = h
-		a.State = notifier.StatePending
-		r.alerts[h] = a
-	}
-
-	for h, a := range r.alerts {
-		// if alert wasn't updated in this iteration
-		// means it is resolved already
-		if _, ok := updated[h]; !ok {
-			a.State = notifier.StateInactive
-			// set endTime to last execution time
-			// so it can be sent by notifier on next step
-			a.End = r.lastExecTime
-			continue
-		}
-		if a.State == notifier.StatePending && time.Since(a.Start) >= r.For {
-			a.State = notifier.StateFiring
-			alertsFired.Inc()
-		}
-		if a.State == notifier.StateFiring {
-			a.End = r.lastExecTime.Add(3 * *evaluationInterval)
-		}
-	}
-	return nil
-}
-
-// TODO: consider hashing algorithm in VM
-func hash(m datasource.Metric) uint64 {
-	hash := fnv.New64a()
-	labels := m.Labels
-	sort.Slice(labels, func(i, j int) bool {
-		return labels[i].Name < labels[j].Name
-	})
-	for _, l := range labels {
-		// drop __name__ to be consistent with Prometheus alerting
-		if l.Name == "__name__" {
-			continue
-		}
-		hash.Write([]byte(l.Name))
-		hash.Write([]byte(l.Value))
-		hash.Write([]byte("\xff"))
-	}
-	return hash.Sum64()
-}
-
-func (r *Rule) newAlert(m datasource.Metric) (*notifier.Alert, error) {
-	a := &notifier.Alert{
-		GroupID: r.group.ID(),
-		Name:    r.Name,
-		Labels:  map[string]string{},
-		Value:   m.Value,
-		Start:   time.Now(),
-		// TODO: support End time
-	}
-	for _, l := range m.Labels {
-		// drop __name__ to be consistent with Prometheus alerting
-		if l.Name == "__name__" {
-			continue
-		}
-		a.Labels[l.Name] = l.Value
-	}
-	return a, r.template(a)
-}
-
-func (r *Rule) template(a *notifier.Alert) error {
-	// 1. template rule labels with data labels
-	rLabels, err := a.ExecTemplate(r.Labels)
-	if err != nil {
-		return err
-	}
-
-	// 2. merge data labels and rule labels
-	// metric labels may be overridden by
-	// rule labels
-	for k, v := range rLabels {
-		a.Labels[k] = v
-	}
-
-	// 3. template merged labels
-	a.Labels, err = a.ExecTemplate(a.Labels)
-	if err != nil {
-		return err
-	}
-
-	a.Annotations, err = a.ExecTemplate(r.Annotations)
-	return err
-}
-
-// AlertAPI generates APIAlert object from alert by its id(hash)
-func (r *Rule) AlertAPI(id uint64) *APIAlert {
-	r.mu.RLock()
-	defer r.mu.RUnlock()
-	a, ok := r.alerts[id]
-	if !ok {
-		return nil
-	}
-	return r.newAlertAPI(*a)
-}
-
-// AlertsAPI generates list of APIAlert objects from existing alerts
-func (r *Rule) AlertsAPI() []*APIAlert {
-	var alerts []*APIAlert
-	r.mu.RLock()
-	for _, a := range r.alerts {
-		alerts = append(alerts, r.newAlertAPI(*a))
-	}
-	r.mu.RUnlock()
-	return alerts
-}
-
-func (r *Rule) newAlertAPI(a notifier.Alert) *APIAlert {
-	return &APIAlert{
-		// encode as strings to avoid rounding
-		ID:      fmt.Sprintf("%d", a.ID),
-		GroupID: fmt.Sprintf("%d", a.GroupID),
-
-		Name:        a.Name,
-		Expression:  r.Expr,
-		Labels:      a.Labels,
-		Annotations: a.Annotations,
-		State:       a.State.String(),
-		ActiveAt:    a.Start,
-		Value:       strconv.FormatFloat(a.Value, 'e', -1, 64),
-	}
-}
-
-const (
-	// AlertMetricName is the metric name for synthetic alert timeseries.
-	alertMetricName = "ALERTS"
-	// AlertForStateMetricName is the metric name for 'for' state of alert.
-	alertForStateMetricName = "ALERTS_FOR_STATE"
-
-	// AlertNameLabel is the label name indicating the name of an alert.
-	alertNameLabel = "alertname"
-	// AlertStateLabel is the label name indicating the state of an alert.
-	alertStateLabel = "alertstate"
-)
-
-// AlertToTimeSeries converts the given alert with the given timestamp to timeseries
-func (r *Rule) AlertToTimeSeries(a *notifier.Alert, timestamp time.Time) []prompbmarshal.TimeSeries {
-	var tss []prompbmarshal.TimeSeries
-	tss = append(tss, alertToTimeSeries(r.Name, a, timestamp))
-	if r.For > 0 {
-		tss = append(tss, alertForToTimeSeries(r.Name, a, timestamp))
-	}
-	return tss
-}
-
-func alertToTimeSeries(name string, a *notifier.Alert, timestamp time.Time) prompbmarshal.TimeSeries {
-	labels := make(map[string]string)
-	for k, v := range a.Labels {
-		labels[k] = v
-	}
-	labels["__name__"] = alertMetricName
-	labels[alertNameLabel] = name
-	labels[alertStateLabel] = a.State.String()
-	return newTimeSeries(1, labels, timestamp)
-}
-
-// alertForToTimeSeries returns a timeseries that represents
-// state of active alerts, where value is time when alert become active
-func alertForToTimeSeries(name string, a *notifier.Alert, timestamp time.Time) prompbmarshal.TimeSeries {
-	labels := make(map[string]string)
-	for k, v := range a.Labels {
-		labels[k] = v
-	}
-	labels["__name__"] = alertForStateMetricName
-	labels[alertNameLabel] = name
-	return newTimeSeries(float64(a.Start.Unix()), labels, timestamp)
-}
-
-func newTimeSeries(value float64, labels map[string]string, timestamp time.Time) prompbmarshal.TimeSeries {
-	ts := prompbmarshal.TimeSeries{}
-	ts.Samples = append(ts.Samples, prompbmarshal.Sample{
-		Value:     value,
-		Timestamp: timestamp.UnixNano() / 1e6,
-	})
-	keys := make([]string, 0, len(labels))
-	for k := range labels {
-		keys = append(keys, k)
-	}
-	sort.Strings(keys)
-	for _, key := range keys {
-		ts.Labels = append(ts.Labels, prompbmarshal.Label{
-			Name:  key,
-			Value: labels[key],
-		})
-	}
-	return ts
-}
-
-// Restore restores the state of active alerts basing on previously written timeseries.
-// Restore restores only Start field. Field State will be always Pending and supposed
-// to be updated on next Eval, as well as Value field.
-func (r *Rule) Restore(ctx context.Context, q datasource.Querier, lookback time.Duration) error {
-	// Get the last datapoint in range via MetricsQL `last_over_time`.
-	// We don't use plain PromQL since Prometheus doesn't support
-	// remote write protocol which is used for state persistence in vmalert.
-	expr := fmt.Sprintf("last_over_time(%s{alertname=%q}[%ds])",
-		alertForStateMetricName, r.Name, int(lookback.Seconds()))
-	qMetrics, err := q.Query(ctx, expr)
-	if err != nil {
-		return err
-	}
-
-	for _, m := range qMetrics {
-		labels := m.Labels
-		m.Labels = make([]datasource.Label, 0)
-		// drop all extra labels, so hash key will
-		// be identical to timeseries received in Eval
-		for _, l := range labels {
-			if l.Name == alertNameLabel {
-				continue
-			}
-			// drop all overridden labels
-			if _, ok := r.Labels[l.Name]; ok {
-				continue
-			}
-			m.Labels = append(m.Labels, l)
-		}
-
-		a, err := r.newAlert(m)
-		if err != nil {
-			return fmt.Errorf("failed to create alert: %s", err)
-		}
-		a.ID = hash(m)
-		a.State = notifier.StatePending
-		a.Start = time.Unix(int64(m.Value), 0)
-		r.alerts[a.ID] = a
-		logger.Infof("alert %q(%d) restored to state at %v", a.Name, a.ID, a.Start)
-	}
-	return nil
+// Rule represents alerting or recording rule
+// that has unique ID, can be Executed and
+// updated with other Rule.
+type Rule interface {
+	// Returns unique ID that may be used for
+	// identifying this Rule among others.
+	ID() uint64
+	// Exec executes the rule with given context
+	// and Querier. If returnSeries is true, Exec
+	// may return TimeSeries as result of execution
+	Exec(ctx context.Context, q datasource.Querier, returnSeries bool) ([]prompbmarshal.TimeSeries, error)
+	// UpdateWith performs modification of current Rule
+	// with fields of the given Rule.
+	UpdateWith(Rule) error
 }

app/vmalert/utils.go

@@ -0,0 +1,27 @@
+package main
+
+import (
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+	"sort"
+	"time"
+)
+
+func newTimeSeries(value float64, labels map[string]string, timestamp time.Time) prompbmarshal.TimeSeries {
+	ts := prompbmarshal.TimeSeries{}
+	ts.Samples = append(ts.Samples, prompbmarshal.Sample{
+		Value:     value,
+		Timestamp: timestamp.UnixNano() / 1e6,
+	})
+	keys := make([]string, 0, len(labels))
+	for k := range labels {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	for _, key := range keys {
+		ts.Labels = append(ts.Labels, prompbmarshal.Label{
+			Name:  key,
+			Value: labels[key],
+		})
+	}
+	return ts
+}

app/vmalert/web.go

@@ -7,32 +7,18 @@ import (
 	"sort"
 	"strconv"
 	"strings"
-	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/httpserver"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/procutil"
 )
 
-// APIAlert represents an notifier.Alert state
-// for WEB view
-type APIAlert struct {
-	ID          string            `json:"id"`
-	Name        string            `json:"name"`
-	GroupID     string            `json:"group_id"`
-	Expression  string            `json:"expression"`
-	State       string            `json:"state"`
-	Value       string            `json:"value"`
-	Labels      map[string]string `json:"labels"`
-	Annotations map[string]string `json:"annotations"`
-	ActiveAt    time.Time         `json:"activeAt"`
-}
-
 type requestHandler struct {
 	m *manager
 }
 
 var pathList = [][]string{
+	{"/api/v1/groups", "list all loaded groups and rules"},
 	{"/api/v1/alerts", "list all active alerts"},
 	{"/api/v1/groupID/alertID/status", "get alert status by ID"},
 	// /metrics is served by httpserver by default
@@ -49,8 +35,11 @@ func (rh *requestHandler) handler(w http.ResponseWriter, r *http.Request) bool {
 			fmt.Fprintf(w, "<a href='%s'>%q</a> - %s<br/>", p, p, doc)
 		}
 		return true
+	case "/api/v1/groups":
+		resph.handle(rh.listGroups())
+		return true
 	case "/api/v1/alerts":
-		resph.handle(rh.list())
+		resph.handle(rh.listAlerts())
 		return true
 	case "/-/reload":
 		logger.Infof("api config reload was called, sending sighup")
@@ -67,6 +56,37 @@ func (rh *requestHandler) handler(w http.ResponseWriter, r *http.Request) bool {
 	}
 }
 
+type listGroupsResponse struct {
+	Data struct {
+		Groups []APIGroup `json:"groups"`
+	} `json:"data"`
+	Status string `json:"status"`
+}
+
+func (rh *requestHandler) listGroups() ([]byte, error) {
+	rh.m.groupsMu.RLock()
+	defer rh.m.groupsMu.RUnlock()
+
+	lr := listGroupsResponse{Status: "success"}
+	for _, g := range rh.m.groups {
+		lr.Data.Groups = append(lr.Data.Groups, g.toAPI())
+	}
+
+	// sort list of alerts for deterministic output
+	sort.Slice(lr.Data.Groups, func(i, j int) bool {
+		return lr.Data.Groups[i].Name < lr.Data.Groups[j].Name
+	})
+
+	b, err := json.Marshal(lr)
+	if err != nil {
+		return nil, &httpserver.ErrorWithStatusCode{
+			Err:        fmt.Errorf(`error encoding list of active alerts: %s`, err),
+			StatusCode: http.StatusInternalServerError,
+		}
+	}
+	return b, nil
+}
+
 type listAlertsResponse struct {
 	Data struct {
 		Alerts []*APIAlert `json:"alerts"`
@@ -74,13 +94,18 @@ type listAlertsResponse struct {
 	Status string `json:"status"`
 }
 
-func (rh *requestHandler) list() ([]byte, error) {
+func (rh *requestHandler) listAlerts() ([]byte, error) {
 	rh.m.groupsMu.RLock()
 	defer rh.m.groupsMu.RUnlock()
+
 	lr := listAlertsResponse{Status: "success"}
 	for _, g := range rh.m.groups {
 		for _, r := range g.Rules {
-			lr.Data.Alerts = append(lr.Data.Alerts, r.AlertsAPI()...)
+			a, ok := r.(*AlertingRule)
+			if !ok {
+				continue
+			}
+			lr.Data.Alerts = append(lr.Data.Alerts, a.AlertsAPI()...)
 		}
 	}
 

app/vmalert/web_test.go

@@ -11,7 +11,7 @@ import (
 )
 
 func TestHandler(t *testing.T) {
-	rule := &Rule{
+	ar := &AlertingRule{
 		Name: "alert",
 		alerts: map[uint64]*notifier.Alert{
 			0: {},
@@ -19,7 +19,7 @@ func TestHandler(t *testing.T) {
 	}
 	g := &Group{
 		Name:  "group",
-		Rules: []*Rule{rule},
+		Rules: []Rule{ar},
 	}
 	m := &manager{groups: make(map[uint64]*Group)}
 	m.groups[0] = g
@@ -54,10 +54,17 @@ func TestHandler(t *testing.T) {
 			t.Errorf("expected 1 alert got %d", length)
 		}
 	})
+	t.Run("/api/v1/groups", func(t *testing.T) {
+		lr := listGroupsResponse{}
+		getResp(ts.URL+"/api/v1/groups", &lr, 200)
+		if length := len(lr.Data.Groups); length != 1 {
+			t.Errorf("expected 1 group got %d", length)
+		}
+	})
 	t.Run("/api/v1/0/0/status", func(t *testing.T) {
 		alert := &APIAlert{}
 		getResp(ts.URL+"/api/v1/0/0/status", alert, 200)
-		expAlert := rule.newAlertAPI(*rule.alerts[0])
+		expAlert := ar.newAlertAPI(*ar.alerts[0])
 		if !reflect.DeepEqual(alert, expAlert) {
 			t.Errorf("expected %v is equal to %v", alert, expAlert)
 		}

app/vmalert/web_types.go

@@ -0,0 +1,53 @@
+package main
+
+import (
+	"time"
+)
+
+// APIAlert represents an notifier.AlertingRule state
+// for WEB view
+type APIAlert struct {
+	ID          string            `json:"id"`
+	Name        string            `json:"name"`
+	GroupID     string            `json:"group_id"`
+	Expression  string            `json:"expression"`
+	State       string            `json:"state"`
+	Value       string            `json:"value"`
+	Labels      map[string]string `json:"labels"`
+	Annotations map[string]string `json:"annotations"`
+	ActiveAt    time.Time         `json:"activeAt"`
+}
+
+// APIGroup represents Group for WEB view
+type APIGroup struct {
+	Name           string             `json:"name"`
+	ID             string             `json:"id"`
+	File           string             `json:"file"`
+	Interval       string             `json:"interval"`
+	AlertingRules  []APIAlertingRule  `json:"alerting_rules"`
+	RecordingRules []APIRecordingRule `json:"recording_rules"`
+}
+
+// APIAlertingRule represents AlertingRule for WEB view
+type APIAlertingRule struct {
+	ID          string            `json:"id"`
+	Name        string            `json:"name"`
+	GroupID     string            `json:"group_id"`
+	Expression  string            `json:"expression"`
+	For         string            `json:"for"`
+	LastError   string            `json:"last_error"`
+	LastExec    time.Time         `json:"last_exec"`
+	Labels      map[string]string `json:"labels"`
+	Annotations map[string]string `json:"annotations"`
+}
+
+// APIRecordingRule represents RecordingRule for WEB view
+type APIRecordingRule struct {
+	ID         string            `json:"id"`
+	Name       string            `json:"name"`
+	GroupID    string            `json:"group_id"`
+	Expression string            `json:"expression"`
+	LastError  string            `json:"last_error"`
+	LastExec   time.Time         `json:"last_exec"`
+	Labels     map[string]string `json:"labels"`
+}

app/vmauth/Makefile

@@ -53,6 +53,8 @@ publish-vmauth:
 
 run-vmauth:
 	APP_NAME=vmauth \
+	DOCKER_OPTS='-v $(shell pwd)/app/vmauth/:/app/vmauth' \
+	ARGS='-auth.config=app/vmauth/example_config.yml' \
 	$(MAKE) run-via-docker
 
 vmauth-amd64:

app/vmauth/auth_config.go

@@ -36,11 +36,11 @@ type UserInfo struct {
 
 func initAuthConfig() {
 	if len(*authConfigPath) == 0 {
-		logger.Panicf("FATAL: missing required `-auth.config` command-line flag")
+		logger.Fatalf("missing required `-auth.config` command-line flag")
 	}
 	m, err := readAuthConfig(*authConfigPath)
 	if err != nil {
-		logger.Panicf("FATAL: cannot load auth config from `-auth.config=%s`: %s", *authConfigPath, err)
+		logger.Fatalf("cannot load auth config from `-auth.config=%s`: %s", *authConfigPath, err)
 	}
 	authConfig.Store(m)
 	stopCh = make(chan struct{})
@@ -63,12 +63,14 @@ func authConfigReloader() {
 		case <-stopCh:
 			return
 		case <-sighupCh:
+			logger.Infof("SIGHUP received; loading -auth.config=%q", *authConfigPath)
 			m, err := readAuthConfig(*authConfigPath)
 			if err != nil {
-				logger.Errorf("failed to load auth config; using the last successfully loaded config; error: %s", err)
+				logger.Errorf("failed to load -auth.config=%q; using the last successfully loaded config; error: %s", *authConfigPath, err)
 				continue
 			}
 			authConfig.Store(m)
+			logger.Infof("Successfully reloaded -auth.config=%q", *authConfigPath)
 		}
 	}
 }

app/vmauth/example_config.yml

@@ -0,0 +1,31 @@
+# Arbitrary number of usernames may be put here.
+# Usernames must be unique.
+
+users:
+
+  # The user for querying local single-node VictoriaMetrics.
+  # All the requests to http://vmauth:8427 with the given Basic Auth (username:password)
+  # will be routed to http://localhost:8428 .
+  # For example, http://vmauth:8427/api/v1/query is routed to http://localhost:8428/api/v1/query
+- username: "local-single-node"
+  password: "***"
+  url_prefix: "http://localhost:8428"
+
+  # The user for querying account 123 in VictoriaMetrics cluster
+  # See https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#url-format
+  # All the requests to http://vmauth:8427 with the given Basic Auth (username:password)
+  # will be routed to http://vmselect:8481/select/123/prometheus .
+  # For example, http://vmauth:8427/api/v1/query is routed to http://vmselect:8481/select/123/prometheus/api/v1/select
+- username: "cluster-select-account-123"
+  password: "***"
+  url_prefix: "http://vmselect:8481/select/123/prometheus"
+
+  # The user for inserting Prometheus data into VictoriaMetrics cluster under account 42
+  # See https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#url-format
+  # All the reuqests to http://vmauth:8427 with the given Basic Auth (username:password)
+  # will be routed to http://vminsert:8480/insert/42/prometheus .
+  # For example, http://vmauth:8427/api/v1/write is routed to http://vminsert:8480/insert/42/prometheus/api/v1/write
+- username: "cluster-insert-account-42"
+  password: "***"
+  url_prefix: "http://vminsert:8480/insert/42/prometheus"
+

app/vmauth/main.go

@@ -2,9 +2,11 @@ package main
 
 import (
 	"flag"
+	"fmt"
 	"net/http"
 	"net/http/httputil"
 	"net/url"
+	"os"
 	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/buildinfo"
@@ -19,6 +21,9 @@ var (
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
+	flag.Usage = usage
 	envflag.Parse()
 	buildinfo.Init()
 	logger.Init()
@@ -74,6 +79,26 @@ var reverseProxy = &httputil.ReverseProxy{
 		}
 		r.URL = target
 	},
+	Transport: func() *http.Transport {
+		tr := http.DefaultTransport.(*http.Transport).Clone()
+		// Automatic compression must be disabled in order to fix https://github.com/VictoriaMetrics/VictoriaMetrics/issues/535
+		tr.DisableCompression = true
+		// Disable HTTP/2.0, since VictoriaMetrics components don't support HTTP/2.0 (because there is no sense in this).
+		tr.ForceAttemptHTTP2 = false
+		return tr
+	}(),
 	FlushInterval: time.Second,
 	ErrorLog:      logger.StdErrorLogger(),
 }
+
+func usage() {
+	const s = `
+vmauth authenticates and authorizes incoming requests and proxies them to VictoriaMetrics.
+
+See the docs at https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmauth/README.md .
+`
+
+	f := flag.CommandLine.Output()
+	fmt.Fprintf(f, "%s\n", s)
+	flag.PrintDefaults()
+}

app/vmbackup/README.md

@@ -19,6 +19,9 @@ Backed up data can be restored with [vmrestore](https://github.com/VictoriaMetri
 
 See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
 
+See also [vmbackuper](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/466) tool built on top of `vmbackup`. This tool simplifies
+creation of hourly, daily, weekly and monthly backups.
+
 
 ### Use cases
 

app/vmbackup/main.go

@@ -31,6 +31,8 @@ var (
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
 	flag.Usage = usage
 	envflag.Parse()
 	buildinfo.Init()

app/vmrestore/README.md

@@ -21,7 +21,8 @@ vmrestore -src=gcs://<bucket>/<path/to/backup> -storageDataPath=<local/path/to/r
 * `<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.
+The original `-storageDataPath` directory may contain old files. They will be susbstituted by the files from backup,
+i.e. the end result would be similar to [rsync --delete](https://askubuntu.com/questions/476041/how-do-i-make-rsync-delete-files-that-have-been-deleted-from-the-source-folder).
 
 
 ### Troubleshooting
@@ -52,7 +53,7 @@ Run `vmrestore -help` in order to see all the available options:
   -envflag.prefix string
     	Prefix for environment variables if -envflag.enable is set
   -fs.disableMmap
-    	Whether to use pread() instead of mmap() for reading data files
+    	Whether to use pread() instead of mmap() for reading data files. By default mmap() is used for 64-bit arches and pread() is used for 32-bit arches, since they cannot data files bigger than 2^32 bytes in memory
   -loggerFormat string
     	Format for logs. Possible values: default, json (default "default")
   -loggerLevel string
@@ -68,8 +69,8 @@ Run `vmrestore -help` in order to see all the available options:
   -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
+    	Destination path where backup must be restored. VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case the contents of -storageDataPath dir is synchronized with -src contents, i.e. it works like 'rsync --delete' (default "victoria-metrics-data")
+ -version
     	Show VictoriaMetrics version
 ```
 

app/vmrestore/main.go

@@ -3,6 +3,7 @@ package main
 import (
 	"flag"
 	"fmt"
+	"os"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/actions"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/backup/common"
@@ -16,13 +17,16 @@ 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")
+		"VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case the contents of -storageDataPath dir "+
+		"is synchronized with -src contents, i.e. it works like 'rsync --delete'")
 	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")
 	skipBackupCompleteCheck = flag.Bool("skipBackupCompleteCheck", false, "Whether to skip checking for 'backup complete' file in -src. This may be useful for restoring from old backups, which were created without 'backup complete' file")
 )
 
 func main() {
+	// Write flags and help message to stdout, since it is easier to grep or pipe.
+	flag.CommandLine.SetOutput(os.Stdout)
 	flag.Usage = usage
 	envflag.Parse()
 	buildinfo.Init()

app/vmselect/prometheus/prometheus.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/netstorage"
 	"github.com/VictoriaMetrics/VictoriaMetrics/app/vmselect/promql"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
 	"github.com/VictoriaMetrics/metrics"
@@ -395,14 +396,14 @@ func TSDBStatusHandler(startTime time.Time, w http.ResponseWriter, r *http.Reque
 	if err := r.ParseForm(); err != nil {
 		return fmt.Errorf("cannot parse form values: %s", err)
 	}
-	date := time.Now().Unix() / secsPerDay
+	date := fasttime.UnixDate()
 	dateStr := r.FormValue("date")
 	if len(dateStr) > 0 {
 		t, err := time.Parse("2006-01-02", dateStr)
 		if err != nil {
 			return fmt.Errorf("cannot parse `date` arg %q: %s", dateStr, err)
 		}
-		date = t.Unix() / secsPerDay
+		date = uint64(t.Unix()) / secsPerDay
 	}
 	topN := 10
 	topNStr := r.FormValue("topN")
@@ -419,7 +420,7 @@ func TSDBStatusHandler(startTime time.Time, w http.ResponseWriter, r *http.Reque
 		}
 		topN = n
 	}
-	status, err := netstorage.GetTSDBStatusForDate(deadline, uint64(date), topN)
+	status, err := netstorage.GetTSDBStatusForDate(deadline, date, topN)
 	if err != nil {
 		return fmt.Errorf(`cannot obtain tsdb status for date=%d, topN=%d: %s`, date, topN, err)
 	}
@@ -992,7 +993,7 @@ func getBool(r *http.Request, argKey string) bool {
 }
 
 func currentTime() int64 {
-	return int64(time.Now().UTC().Unix()) * 1e3
+	return int64(fasttime.UnixTimestamp() * 1000)
 }
 
 func getTagFilterssFromMatches(matches []string) ([][]storage.TagFilter, error) {

app/vmselect/promql/aggr.go

@@ -44,6 +44,7 @@ var aggrFuncs = map[string]aggrFunc{
 	"bottomk_avg":    newAggrFuncRangeTopK(avgValue, true),
 	"bottomk_median": newAggrFuncRangeTopK(medianValue, true),
 	"any":            newAggrFunc(aggrFuncAny),
+	"outliersk":      aggrFuncOutliersK,
 }
 
 type aggrFunc func(afa *aggrFuncArg) ([]*timeseries, error)
@@ -483,11 +484,6 @@ func newAggrFuncTopK(isReverse bool) aggrFunc {
 	}
 }
 
-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
@@ -499,34 +495,42 @@ func newAggrFuncRangeTopK(f func(values []float64) float64, isReverse bool) aggr
 			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 getRangeTopKTimeseries(tss, ks, f, isReverse)
 		}
 		return aggrFuncExt(afe, args[1], &afa.ae.Modifier, afa.ae.Limit, true)
 	}
 }
 
+func getRangeTopKTimeseries(tss []*timeseries, ks []float64, f func(values []float64) float64, isReverse bool) []*timeseries {
+	type tsWithValue struct {
+		ts    *timeseries
+		value float64
+	}
+	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)
+}
+
 func fillNaNsAtIdx(idx int, k float64, tss []*timeseries) {
 	if math.IsNaN(k) {
 		k = 0
@@ -588,16 +592,54 @@ func avgValue(values []float64) float64 {
 func medianValue(values []float64) float64 {
 	h := histogram.GetFast()
 	for _, v := range values {
-		if math.IsNaN(v) {
-			continue
+		if !math.IsNaN(v) {
+			h.Update(v)
 		}
-		h.Update(v)
 	}
 	value := h.Quantile(0.5)
 	histogram.PutFast(h)
 	return value
 }
 
+func aggrFuncOutliersK(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 {
+		// Calculate medians for each point across tss.
+		medians := make([]float64, len(ks))
+		h := histogram.GetFast()
+		for n := range ks {
+			h.Reset()
+			for j := range tss {
+				v := tss[j].Values[n]
+				if !math.IsNaN(v) {
+					h.Update(v)
+				}
+			}
+			medians[n] = h.Quantile(0.5)
+		}
+		histogram.PutFast(h)
+
+		// Return topK time series with the highest variance from median.
+		f := func(values []float64) float64 {
+			sum2 := float64(0)
+			for n, v := range values {
+				d := v - medians[n]
+				sum2 += d * d
+			}
+			return sum2
+		}
+		return getRangeTopKTimeseries(tss, ks, f, false)
+	}
+	return aggrFuncExt(afe, args[1], &afa.ae.Modifier, afa.ae.Limit, true)
+}
+
 func aggrFuncLimitK(afa *aggrFuncArg) ([]*timeseries, error) {
 	args := afa.args
 	if err := expectTransformArgsNum(args, 2); err != nil {
@@ -658,24 +700,18 @@ func aggrFuncMedian(afa *aggrFuncArg) ([]*timeseries, error) {
 func newAggrQuantileFunc(phis []float64) func(tss []*timeseries) []*timeseries {
 	return func(tss []*timeseries) []*timeseries {
 		dst := tss[0]
+		h := histogram.GetFast()
+		defer histogram.PutFast(h)
 		for n := range dst.Values {
-			sort.Slice(tss, func(i, j int) bool {
-				a := tss[i].Values[n]
-				b := tss[j].Values[n]
-				return lessWithNaNs(a, b)
-			})
+			h.Reset()
+			for j := range tss {
+				v := tss[j].Values[n]
+				if !math.IsNaN(v) {
+					h.Update(v)
+				}
+			}
 			phi := phis[n]
-			if math.IsNaN(phi) {
-				phi = 1
-			}
-			if phi < 0 {
-				phi = 0
-			}
-			if phi > 1 {
-				phi = 1
-			}
-			idx := int(math.Round(float64(len(tss)-1) * phi))
-			dst.Values[n] = tss[idx].Values[n]
+			dst.Values[n] = h.Quantile(phi)
 		}
 		tss[0] = dst
 		return tss[:1]

app/vmselect/promql/aggr_incremental.go

@@ -465,7 +465,6 @@ func finalizeAggrGeomean(iac *incrementalAggrContext) {
 }
 
 func updateAggrAny(iac *incrementalAggrContext, values []float64) {
-	dstValues := iac.ts.Values
 	dstCounts := iac.values
 	if dstCounts[0] > 0 {
 		return
@@ -473,17 +472,16 @@ func updateAggrAny(iac *incrementalAggrContext, values []float64) {
 	for i := range values {
 		dstCounts[i] = 1
 	}
-	dstValues = append(dstValues[:0], values...)
+	iac.ts.Values = append(iac.ts.Values[:0], values...)
 }
 
 func mergeAggrAny(dst, src *incrementalAggrContext) {
 	srcValues := src.ts.Values
-	dstValues := dst.ts.Values
 	srcCounts := src.values
 	dstCounts := dst.values
 	if dstCounts[0] > 0 {
 		return
 	}
 	dstCounts[0] = srcCounts[0]
-	dstValues = append(dstValues[:0], srcValues...)
+	dst.ts.Values = append(dst.ts.Values[:0], srcValues...)
 }

app/vmselect/promql/exec_test.go

@@ -3577,7 +3577,7 @@ func TestExecSuccess(t *testing.T) {
 		q := `sort(histogram_over_time(alias(label_set(rand(0)*1.3+1.1, "foo", "bar"), "xxx")[200s:5s]))`
 		r1 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{14, 15, 12, 13, 15, 11},
+			Values:     []float64{14, 16, 12, 13, 15, 11},
 			Timestamps: timestampsExpected,
 		}
 		r1.MetricName.Tags = []storage.Tag{
@@ -3607,7 +3607,7 @@ func TestExecSuccess(t *testing.T) {
 		}
 		r3 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{13, 11, 16, 19, 13, 16},
+			Values:     []float64{13, 10, 16, 19, 13, 16},
 			Timestamps: timestampsExpected,
 		}
 		r3.MetricName.Tags = []storage.Tag{
@@ -3628,7 +3628,7 @@ func TestExecSuccess(t *testing.T) {
 		q := `sort(sum(histogram_over_time(alias(label_set(rand(0)*1.3+1.1, "foo", "bar"), "xxx")[200s:5s])) by (vmrange))`
 		r1 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{14, 15, 12, 13, 15, 11},
+			Values:     []float64{14, 16, 12, 13, 15, 11},
 			Timestamps: timestampsExpected,
 		}
 		r1.MetricName.Tags = []storage.Tag{
@@ -3650,7 +3650,7 @@ func TestExecSuccess(t *testing.T) {
 		}
 		r3 := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{13, 11, 16, 19, 13, 16},
+			Values:     []float64{13, 10, 16, 19, 13, 16},
 			Timestamps: timestampsExpected,
 		}
 		r3.MetricName.Tags = []storage.Tag{
@@ -3678,7 +3678,7 @@ func TestExecSuccess(t *testing.T) {
 		q := `topk_max(1, histogram_over_time(alias(label_set(rand(0)*1.3+1.1, "foo", "bar"), "xxx")[200s:5s]))`
 		r := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{13, 11, 16, 19, 13, 16},
+			Values:     []float64{13, 10, 16, 19, 13, 16},
 			Timestamps: timestampsExpected,
 		}
 		r.MetricName.Tags = []storage.Tag{
@@ -4196,14 +4196,63 @@ func TestExecSuccess(t *testing.T) {
 	t.Run(`quantile(NaN)`, func(t *testing.T) {
 		t.Parallel()
 		q := `quantile(NaN, label_set(10, "foo", "bar") or label_set(time()/150, "baz", "sss"))`
+		resultExpected := []netstorage.Result{}
+		f(q, resultExpected)
+	})
+	t.Run(`outliersk(0)`, func(t *testing.T) {
+		t.Parallel()
+		q := `outliersk(0, (
+			label_set(1300, "foo", "bar"),
+			label_set(time(), "baz", "sss"),
+		))`
+		resultExpected := []netstorage.Result{}
+		f(q, resultExpected)
+	})
+	t.Run(`outliersk(1)`, func(t *testing.T) {
+		t.Parallel()
+		q := `outliersk(1, (
+			label_set(2000, "foo", "bar"),
+			label_set(time(), "baz", "sss"),
+		))`
 		r := netstorage.Result{
 			MetricName: metricNameExpected,
-			Values:     []float64{10, 10, 10, 10.666666666666666, 12, 13.333333333333334},
+			Values:     []float64{1000, 1200, 1400, 1600, 1800, 2000},
 			Timestamps: timestampsExpected,
 		}
+		r.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
 		resultExpected := []netstorage.Result{r}
 		f(q, resultExpected)
 	})
+	t.Run(`outliersk(3)`, func(t *testing.T) {
+		t.Parallel()
+		q := `sort_desc(outliersk(3, (
+			label_set(1300, "foo", "bar"),
+			label_set(time(), "baz", "sss"),
+		)))`
+		r1 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1000, 1200, 1400, 1600, 1800, 2000},
+			Timestamps: timestampsExpected,
+		}
+		r1.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("baz"),
+			Value: []byte("sss"),
+		}}
+		r2 := netstorage.Result{
+			MetricName: metricNameExpected,
+			Values:     []float64{1300, 1300, 1300, 1300, 1300, 1300},
+			Timestamps: timestampsExpected,
+		}
+		r2.MetricName.Tags = []storage.Tag{{
+			Key:   []byte("foo"),
+			Value: []byte("bar"),
+		}}
+		resultExpected := []netstorage.Result{r1, r2}
+		f(q, resultExpected)
+	})
 	t.Run(`range_quantile(0.5)`, func(t *testing.T) {
 		t.Parallel()
 		q := `range_quantile(0.5, time())`
@@ -5531,6 +5580,8 @@ func TestExecError(t *testing.T) {
 	f(`hoeffding_bound_upper()`)
 	f(`hoeffding_bound_upper(1)`)
 	f(`hoeffding_bound_upper(0.99, foo, 1)`)
+	f(`outliersk()`)
+	f(`outliersk(1)`)
 
 	// Invalid argument type
 	f(`median_over_time({}, 2)`)
@@ -5570,6 +5621,7 @@ func TestExecError(t *testing.T) {
 	f(`alias(1, 2)`)
 	f(`aggr_over_time(1, 2)`)
 	f(`aggr_over_time(("foo", "bar"), 3)`)
+	f(`outliersk((label_set(1, "foo", "bar"), label_set(2, "x", "y")), 123)`)
 
 	// Duplicate timeseries
 	f(`(label_set(1, "foo", "bar") or label_set(2, "foo", "baz"))

app/vmselect/promql/rollup.go

@@ -72,6 +72,8 @@ var rollupFuncs = map[string]newRollupFunc{
 	"aggr_over_time":        newRollupFuncTwoArgs(rollupFake),
 	"hoeffding_bound_upper": newRollupHoeffdingBoundUpper,
 	"hoeffding_bound_lower": newRollupHoeffdingBoundLower,
+	"ascent_over_time":      newRollupFuncOneArg(rollupAscentOverTime),
+	"descent_over_time":     newRollupFuncOneArg(rollupDescentOverTime),
 
 	// `timestamp` function must return timestamp for the last datapoint on the current window
 	// in order to properly handle offset and timestamps unaligned to the current step.
@@ -116,6 +118,9 @@ var rollupAggrFuncs = map[string]rollupFunc{
 	"scrape_interval":     rollupScrapeInterval,
 	"tmin_over_time":      rollupTmin,
 	"tmax_over_time":      rollupTmax,
+	"ascent_over_time":    rollupAscentOverTime,
+	"descent_over_time":   rollupDescentOverTime,
+	"timestamp":           rollupTimestamp,
 }
 
 var rollupFuncsCannotAdjustWindow = map[string]bool{
@@ -138,6 +143,8 @@ var rollupFuncsCannotAdjustWindow = map[string]bool{
 	"increases_over_time": true,
 	"decreases_over_time": true,
 	"integrate":           true,
+	"ascent_over_time":    true,
+	"descent_over_time":   true,
 }
 
 var rollupFuncsRemoveCounterResets = map[string]bool{
@@ -1527,6 +1534,52 @@ func rollupTimestamp(rfa *rollupFuncArg) float64 {
 	return float64(timestamps[len(timestamps)-1]) / 1e3
 }
 
+func rollupAscentOverTime(rfa *rollupFuncArg) float64 {
+	// There is no need in handling NaNs here, since they must be cleaned up
+	// before calling rollup funcs.
+	values := rfa.values
+	prevValue := rfa.prevValue
+	if math.IsNaN(prevValue) {
+		if len(values) == 0 {
+			return nan
+		}
+		prevValue = values[0]
+		values = values[1:]
+	}
+	s := float64(0)
+	for _, v := range values {
+		d := v - prevValue
+		if d > 0 {
+			s += d
+		}
+		prevValue = v
+	}
+	return s
+}
+
+func rollupDescentOverTime(rfa *rollupFuncArg) float64 {
+	// There is no need in handling NaNs here, since they must be cleaned up
+	// before calling rollup funcs.
+	values := rfa.values
+	prevValue := rfa.prevValue
+	if math.IsNaN(prevValue) {
+		if len(values) == 0 {
+			return nan
+		}
+		prevValue = values[0]
+		values = values[1:]
+	}
+	s := float64(0)
+	for _, v := range values {
+		d := prevValue - v
+		if d > 0 {
+			s += d
+		}
+		prevValue = v
+	}
+	return s
+}
+
 func rollupFirst(rfa *rollupFuncArg) float64 {
 	// There is no need in handling NaNs here, since they must be cleaned up
 	// before calling rollup funcs.

app/vmselect/promql/rollup_result_cache.go

@@ -10,6 +10,7 @@ import (
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/workingsetcache"
@@ -64,18 +65,18 @@ func InitRollupResultCache(cachePath string) {
 
 	stats := &fastcache.Stats{}
 	var statsLock sync.Mutex
-	var statsLastUpdate time.Time
+	var statsLastUpdate uint64
 	fcs := func() *fastcache.Stats {
 		statsLock.Lock()
 		defer statsLock.Unlock()
 
-		if time.Since(statsLastUpdate) < time.Second {
+		if fasttime.UnixTimestamp()-statsLastUpdate < 2 {
 			return stats
 		}
 		var fcs fastcache.Stats
 		c.UpdateStats(&fcs)
 		stats = &fcs
-		statsLastUpdate = time.Now()
+		statsLastUpdate = fasttime.UnixTimestamp()
 		return stats
 	}
 	if len(rollupResultCachePath) > 0 {

app/vmselect/promql/rollup_test.go

@@ -389,6 +389,9 @@ func TestRollupNewRollupFuncSuccess(t *testing.T) {
 	f("ideriv", 0)
 	f("decreases_over_time", 5)
 	f("increases_over_time", 5)
+	f("ascent_over_time", 142)
+	f("descent_over_time", 231)
+	f("timestamp", 0.13)
 }
 
 func TestRollupNewRollupFuncError(t *testing.T) {

app/vmstorage/main.go

@@ -409,6 +409,16 @@ func registerStorageMetrics() {
 		return float64(m().AddRowsConcurrencyCurrent)
 	})
 
+	metrics.NewGauge(`vm_slow_row_inserts_total`, func() float64 {
+		return float64(m().SlowRowInserts)
+	})
+	metrics.NewGauge(`vm_slow_per_day_index_inserts_total`, func() float64 {
+		return float64(m().SlowPerDayIndexInserts)
+	})
+	metrics.NewGauge(`vm_slow_metric_name_loads_total`, func() float64 {
+		return float64(m().SlowMetricNameLoads)
+	})
+
 	metrics.NewGauge(`vm_rows{type="storage/big"}`, func() float64 {
 		return float64(tm().BigRowsCount)
 	})

deployment/docker/Makefile

@@ -2,9 +2,9 @@
 
 DOCKER_NAMESPACE := victoriametrics
 
-ROOT_IMAGE ?= scratch
-CERTS_IMAGE := alpine:3.11
-GO_BUILDER_IMAGE := golang:1.14.2
+ROOT_IMAGE ?= alpine:3.12
+CERTS_IMAGE := alpine:3.12
+GO_BUILDER_IMAGE := golang:1.14.4
 BUILDER_IMAGE := local/builder:2.0.0-$(shell echo $(GO_BUILDER_IMAGE) | tr : _)
 BASE_IMAGE := local/base:1.1.1-$(shell echo $(ROOT_IMAGE) | tr : _)-$(shell echo $(CERTS_IMAGE) | tr : _)
 
@@ -33,7 +33,7 @@ app-via-docker: package-base package-builder
 		$(DOCKER_OPTS) \
 		$(BUILDER_IMAGE) \
 		go build $(RACE) -mod=vendor -trimpath \
-			-ldflags "-s -w -extldflags '-static' $(GO_BUILDINFO)" \
+			-ldflags "-extldflags '-static' $(GO_BUILDINFO)" \
 			-tags 'netgo osusergo nethttpomithttp2' \
 			-o bin/$(APP_NAME)$(APP_SUFFIX)-prod $(PKG_PREFIX)/app/$(APP_NAME)
 

deployment/docker/docker-compose.yml

@@ -2,7 +2,7 @@ version: '3.5'
 services:
   prometheus:
     container_name: prometheus
-    image: prom/prometheus:v2.17.2
+    image: prom/prometheus:v2.18.1
     depends_on:
       - "victoriametrics"
     ports:
@@ -35,7 +35,7 @@ services:
     restart: always
   grafana:
     container_name: grafana
-    image: grafana/grafana:6.7.2
+    image: grafana/grafana:7.0.2
     entrypoint: >
       /bin/sh -c "
       cd /var/lib/grafana &&

docs/Articles.md

@@ -26,11 +26,14 @@
 * [Billy: how VictoriaMetrics deals with more than 500 billion rows](https://medium.com/@valyala/billy-how-victoriametrics-deals-with-more-than-500-billion-rows-e82ff8f725da)
 
 
-## Third-party articles
+## Third-party articles and slides
 
 * [Better Prometheus rate() function with VictoriaMetrics](https://www.percona.com/blog/2020/02/28/better-prometheus-rate-function-with-victoriametrics/)
 * [Infrastructure monitoring with Prometheus at Zerodha](https://zerodha.tech/blog/infra-monitoring-at-zerodha/)
+* [Sismology: Iguana Solutions’ Monitoring System](https://medium.com/@IG1.com/sismology-iguana-solutions-monitoring-system-f46e4170447f)
+* [Monitoring K8S with VictoriaMetrics](https://docs.google.com/presentation/d/1g7yUyVEaAp4tPuRy-MZbPXKqJ1z78_5VKuV841aQfsg/edit)
 * [CMS monitoring R&D: Real-time monitoring and alerts](https://indico.cern.ch/event/877333/contributions/3696707/attachments/1972189/3281133/CMS_mon_RD_for_opInt.pdf)
 * [Disk usage: VictoriaMetrics vs Prometheus](https://stas.starikevich.com/posts/disk-usage-for-vm-versus-prometheus/)
 * [Benchmarking time series workloads on Apache Kudu using TSBS](https://blog.cloudera.com/benchmarking-time-series-workloads-on-apache-kudu-using-tsbs/)
 * [What are Open Source Time Series Databases?](https://www.iunera.com/kraken/fabric/time-series-database/)
+* [Evaluating performance and correctness](https://www.robustperception.io/evaluating-performance-and-correctness)

docs/CaseStudies.md

@@ -3,6 +3,8 @@
 Below are approved public case studies and talks from VictoriaMetrics users. Join our [community Slack channel](http://slack.victoriametrics.com/)
 and feel free asking for references, reviews and additional case studies from real VictoriaMetrics users there.
 
+See also [articles about VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Articles).
+
 
 ## Adidas
 
@@ -39,6 +41,28 @@ See [slides](https://speakerdeck.com/inletorder/monitoring-platform-with-victori
 from `Large-scale, super-load system monitoring platform built with VictoriaMetrics` talk at [Prometheus Meetup Tokyo #3](https://prometheus.connpass.com/event/157721/).
 
 
+## Zerodha
+
+[Zerodha](https://zerodha.com/) is India's largest stock broker. Monitoring team at Zerodha faced with the following requirements:
+
+* Multiple K8s clusters to monitor
+* Consistent monitoring infra for each cluster across the fleet
+* Ability to handle billions of timeseries events at any point of time
+* Easier to operate and cost effective
+
+Thanos, Cortex and VictoriaMetrics were evaluated as a long-term storage for Prometheus. VictoriaMetrics has been selected due to the following reasons:
+
+* Blazing fast benchmarks for a single node setup.
+* Single binary mode. Easy to scale vertically, very less operational headache.
+* Considerable [improvements on creating Histograms](https://medium.com/@valyala/improving-histogram-usability-for-prometheus-and-grafana-bc7e5df0e350).
+* [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/MetricsQL) gives us the ability to extend PromQL with more aggregation operators.
+* API is compatible with Prometheus, almost all standard PromQL queries just work out of the box.
+* Handles storage well, with periodic compaction. Makes it easy to take snapshots.
+
+See [Monitoring K8S with VictoriaMetrics](https://docs.google.com/presentation/d/1g7yUyVEaAp4tPuRy-MZbPXKqJ1z78_5VKuV841aQfsg/edit) slides,
+[video](https://youtu.be/ZJQYW-cFOms) and [Infrastructure monitoring with Prometheus at Zerodha](https://zerodha.tech/blog/infra-monitoring-at-zerodha/) blog post for more details.
+
+
 ## Wix.com
 
 [Wix.com](https://en.wikipedia.org/wiki/Wix.com) is the leading web development platform.
@@ -209,7 +233,7 @@ Such a scheme has the following benefits comparing to Prometheus:
 
 Cons are the following:
 
-- VictoriaMetrics doesn't support replication - we run extra instance of VictoriaMetrics and Promxy in front of VictoriaMetrics pair for high availability.
+- VictoriaMetrics didn't support replication (it [supports replication now](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#replication-and-data-safety)) - we run extra instance of VictoriaMetrics and Promxy in front of VictoriaMetrics pair for high availability.
 - VictoriaMetrics stores 1 extra month for defined retention (if retention is set to N months, then VM stores N+1 months of data), but this is still better than other solutions.
 
 Some numbers from our single-node VictoriaMetrics setup:

docs/Cluster-VictoriaMetrics.md

@@ -16,8 +16,9 @@ Join [our Slack](http://slack.victoriametrics.com/) or [contact us](mailto:info@
 ## 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).
+- Performance and capacity scales horizontally. See [these docs for details](#cluster-resizing-and-scalability).
+- Supports multiple independent namespaces for time series data (aka multi-tenancy). See [these docs for details](#multitenancy).
+- Supports replication. See [these docs for details](#replication-and-data-safety).
 
 
 ## Architecture overview
@@ -45,7 +46,8 @@ See [these docs](#url-format) for details. Some facts about tenants in VictoriaM
 * Each `accountID` and `projectID` is identified by an arbitrary 32-bit integer in the range `[0 .. 2^32)`.
 If `projectID` is missing, then it is automatically assigned to `0`. It is expected that other information about tenants
 such as auth tokens, tenant names, limits, accounting, etc. is stored in a separate relational database. This database must be managed
-by a separate service sitting in front of VictoriaMetrics cluster.
+by a separate service sitting in front of VictoriaMetrics cluster such as [vmauth](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmauth/README.md).
+[Contact us](mailto:info@victoriametrics.com) if you need help with creating such a service.
 
 * Tenants are automatically created when the first data point is written into the given tenant.
 
@@ -110,11 +112,11 @@ Run `make package`. It will build the following docker images 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`.
 
-By default images are built on top of `scratch` image. It is possible to build on top of any other base image
-by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds images on top of `alpine:3.11` image:
+By default images are built on top of `alpine` image in order to improve debuggability. It is possible to build an image on top of any other base image
+by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds images on top of `scratch` image:
 
 ```bash
-ROOT_IMAGE=alpine:3.11 make package
+ROOT_IMAGE=scratch make package
 ```
 
 ## Operation
@@ -130,7 +132,7 @@ A minimal cluster must contain the following nodes:
 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:
+An http load balancer such as `nginx` 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.
 
@@ -156,7 +158,8 @@ By default the following TCP ports are used:
 - `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
+It is recommended setting up [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md)
+or 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)
 or [an alternative dashboard for VictoriaMetrics cluster](https://grafana.com/grafana/dashboards/11831).
 
@@ -167,21 +170,25 @@ or [an alternative dashboard for VictoriaMetrics cluster](https://grafana.com/gr
   - `<accountID>` is an arbitrary 32-bit integer identifying namespace for data ingestion (aka tenant). It is possible to set it as `accountID:projectID`,
     where `projectID` is also arbitrary 32-bit integer. If `projectID` isn't set, then it equals to `0`.
   - `<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/)
+     - `prometheus` and `prometheus/api/v1/write` - for inserting data with [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write)
+     - `influx/write` and `influx/api/v2/write` - for inserting data with [Influx line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/).
      - `opentsdb/api/put` - for accepting [OpenTSDB HTTP /api/put requests](http://opentsdb.net/docs/build/html/api_http/put.html).
+       This handler is disabled by default. It is exposed on a distinct TCP address set via `-opentsdbHTTPListenAddr` command-line flag.
+       See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#sending-opentsdb-data-via-http-apiput-requests) for details.
      - `prometheus/api/v1/import` - for importing data obtained via `api/v1/export` on `vmselect` (see below).
+     - `prometheus/api/v1/import/csv` - for importing arbitrary CSV data. See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-import-csv-data) for details.
 
 * 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
+    - `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.
+    - `api/v1/status/tsdb` - for time series stats. See [these docs](https://prometheus.io/docs/prometheus/latest/querying/api/#tsdb-stats) 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
@@ -199,7 +206,7 @@ or [an alternative dashboard for VictoriaMetrics cluster](https://grafana.com/gr
   across `vmstorage` nodes.
 
 
-### Cluster resizing and scalability.
+### Cluster resizing and scalability
 
 Cluster performance and capacity scales with adding new nodes.
 
@@ -246,6 +253,8 @@ Each instance type - `vminsert`, `vmselect` and `vmstorage` - can run on the mos
 * 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.
+  The maximum amount of used RAM per `vminsert` node can be tuned with `-memory.allowedPercent` command-line flag. For instance, `-memory.allowedPercent=20`
+  limits the maximum amount of used RAM to 20% of the available RAM on the host system.
 * 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`.
 
@@ -277,7 +286,18 @@ Upgrade follows `Cluster resizing procedure` under the hood.
 
 ### Replication and data safety
 
-VictoriaMetrics offloads replication to the underlying storage pointed by `-storageDataPath`.
+In order to enable application-level replication, `-replicationFactor=N` command-line flag must be passed to `vminsert`.
+This guarantees that all the data remains available for querying if up to `N-1` `vmstorage` nodes are unavailable.
+For example, when `-replicationFactor=3` is passed to `vminsert`, then it replicates all the ingested data to 3 distinct `vmstorage` nodes.
+
+When the replication is enabled, `-dedup.minScrapeInterval=1ms` command-line flag must be passed to `vmselect`
+in order to de-duplicate replicated data during queries. It is OK if `-dedup.minScrapeInterval` exceeds 1ms
+when [deduplication](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#deduplication) is used additionally to replication.
+
+Note that [replication doesn't save from disaster](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883),
+so it is recommended performing regular backups. See [these docs](#backups) for details.
+
+By default 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.
@@ -285,8 +305,6 @@ 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
 
@@ -326,8 +344,7 @@ Due to `KISS` cluster version of VictoriaMetrics has no the following "features"
 
 - 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).
+- Complex replication schemes, which may go nuts in unforesseen edge cases. See [replication docs](#replication-and-data-safety) for details.
 - 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 :)

docs/FAQ.md

@@ -2,64 +2,64 @@
 
 ### 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/).
+To provide the best monitoring solution.
+
+
+### Who uses VictoriaMetrics?
+
+See [case studies](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies).
 
 
 ### 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/MetricsQL).
-* 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, M3DB, 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.
+See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#prominent-features).
 
 
 ### 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.
+See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Quick-Start).
 
 
-### Is it safe to enable [remote write storage](https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage) in Prometheus?
+### What is the difference between vmagent and Prometheus?
 
-Yes. Prometheus continues writing data to local storage after enabling remote storage write, so all the existing local storage data
+While both [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) and Prometheus may scrape Prometheus targets (aka `/metrics` pages)
+according to the provided Prometheus-compatible [scrape configs](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#scrape_config)
+and send data to multiple remote storage systems, vmagent has the following additional features:
+
+- vmagent usually requires lower amounts of CPU, RAM and disk IO comparing to Prometheus when scraping big number of targets (more than 1000)
+  or targets with big number of exposed metrics.
+- vmagent provides independent disk-backed buffers per each configured remote storage (aka `-remoteWrite.url`). This means that slow or temporarily unavailable storage
+  doesn't prevent from sending data to healthy storage in parallel. Prometheus uses a single shared buffer for all the configured remote storage systems (aka `remote_write->url`)
+  with the hardcoded retention of 2 hours.
+- vmagent may accept, relabel and filter data obtained via multiple data ingestion protocols additionally to data scraped from Prometheus targets.
+  I.e. it supports both `pull` and `push` protocols for data ingestion.
+  See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#features) for details.
+- vmagent may be used in different use cases:
+  - [IoT and edge monitoring](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#iot-and-edge-monitoring)
+  - [Drop-in replacement for Prometheus](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#drop-in-replacement-for-prometheus)
+  - [Replication and High Availability](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#replication-and-high-availability)
+  - [Relabeling and Filtering](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#relabeling-and-filtering)
+  - [Splitting data streams among multiple systems](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#splitting-data-streams-among-multiple-systems)
+  - [Prometheus remote_write proxy](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md#prometheus-remote_write-proxy)
+
+
+### Is it safe to enable [remote write](https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage) in Prometheus?
+
+Yes. Prometheus continues writing data to local storage after enabling remote write, so all the existing local storage data
 and new data is available for querying via Prometheus as usual.
 
+It is recommended using [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) for scraping Prometheus targets
+and writing data to VictoriaMetrics.
+
 
 ### How does VictoriaMetrics compare to other remote storage solutions for Prometheus such as [M3 from Uber](https://eng.uber.com/m3/), [Thanos](https://github.com/thanos-io/thanos), [Cortex](https://github.com/cortexproject/cortex), etc.?
 
-VictoriaMetrics is simpler, faster, more cost-effective and it provides [MetricsQL with useful extensions for PromQL](MetricsQL). The simplicity is twofold:
-- It is simpler to configure and operate. There is no need in configuring third-party [sidecars](https://github.com/thanos-io/thanos/blob/master/docs/components/sidecar.md)
-  or fighting with [gossip protocol](https://github.com/improbable-eng/thanos/blob/030bc345c12c446962225221795f4973848caab5/docs/proposals/completed/201809_gossip-removal.md).
-- VictoriaMetrics has simpler architecture, which means less bugs and more useful features in the long run comparing to competing TSDBs.
+VictoriaMetrics is simpler, faster, more cost-effective and it provides [MetricsQL query language](MetricsQL) based on PromQL. The simplicity is twofold:
+- It is simpler to configure and operate. There is no need in configuring [sidecars](https://github.com/thanos-io/thanos/blob/master/docs/components/sidecar.md),
+  fighting [gossip protocol](https://github.com/improbable-eng/thanos/blob/030bc345c12c446962225221795f4973848caab5/docs/proposals/completed/201809_gossip-removal.md)
+  or setting up third-party systems such as [Consul](https://github.com/cortexproject/cortex/issues/157), [Cassandra](https://cortexmetrics.io/docs/production/cassandra/),
+  [DynamoDB](https://cortexmetrics.io/docs/production/aws/) or [Memcached](https://cortexmetrics.io/docs/production/caching/).
+- VictoriaMetrics has simpler architecture. This means less bugs and more useful features in the long run comparing to competing TSDBs.
 
 See [comparing Thanos to VictoriaMetrics cluster](https://medium.com/@valyala/comparing-thanos-to-victoriametrics-cluster-b193bea1683)
 and [Remote Write Storage Wars](https://promcon.io/2019-munich/talks/remote-write-storage-wars/) talk from [PromCon 2019](https://promcon.io/2019-munich/talks/remote-write-storage-wars/).
@@ -70,55 +70,68 @@ VictoriaMetrics also [uses less RAM than Thanos components](https://github.com/t
 ### What is the difference between VictoriaMetrics and [Cortex](https://github.com/cortexproject/cortex)?
 
 VictoriaMetrics is similar to Cortex in the following aspects:
-- Both systems accept data from Prometheus via standard [remote_write API](https://prometheus.io/docs/practices/remote_write/),
-  i.e. there is no need in running sidecars unlike in [Thanos](https://github.com/thanos-io/thanos) case.
-- Both systems support multi-tenancy out of the box. See [the corresponding docs for VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#url-format).
+- Both systems accept data from [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) or Prometheus
+  via standard [remote_write API](https://prometheus.io/docs/practices/remote_write/), i.e. there is no need in running sidecars
+  unlike in [Thanos](https://github.com/thanos-io/thanos) case.
+- Both systems support multi-tenancy out of the box. See [the corresponding docs for VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#multitenancy).
+- Both systems support data replication. See [replication in Cortex](https://github.com/cortexproject/cortex/blob/fe56f1420099aa1bf1ce09316c186e05bddee879/docs/architecture.md#hashing) and [replication in VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#replication-and-data-safety).
+- Both systems scale horizontally to multiple nodes. See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#cluster-resizing-and-scalability) for details.
+- Both systems support alerting and recording rules via the corresponding tools such as [vmalert](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/README.md).
+
 
 The main differences between Cortex and VictoriaMetrics:
 - Cortex re-uses Prometheus source code, while VictoriaMetrics is written from scratch.
-- Cortex provides [Ruler](https://github.com/cortexproject/cortex/blob/master/docs/architecture.md#ruler) and [Alertmanager](https://github.com/cortexproject/cortex/blob/master/docs/architecture.md#alertmanager) components,
-  which are currently missing in VictoriaMetrics. However, these components can be substituted by [Promxy](https://github.com/jacksontj/promxy#how-do-i-use-alertingrecording-rules-in-promxy).
 - Cortex heavily relies on third-party services such as Consul, Memcache, DynamoDB, BigTable, Cassandra, etc.
   This may increase operational complexity and reduce system reliability comparing to VictoriaMetrics' case,
   which doesn't use any external services. Compare [Cortex Architecture](https://github.com/cortexproject/cortex/blob/master/docs/architecture.md)
   to [VictoriaMetrics architecture](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#architecture-overview).
 - VictoriaMetrics provides [production-ready single-node solution](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md),
   which is much easier to setup and operate than Cortex cluster.
-- Cortex may lose up to 12 hours of recent data on Ingestor failure - see [the corresponding docs](https://github.com/cortexproject/cortex/blob/master/docs/architecture.md#ingesters-failure-and-data-loss).
+- Cortex may lose up to 12 hours of recent data on Ingestor failure - see [the corresponding docs](https://github.com/cortexproject/cortex/blob/fe56f1420099aa1bf1ce09316c186e05bddee879/docs/architecture.md#ingesters-failure-and-data-loss).
   VictoriaMetrics may lose only a few seconds of recent data, which isn't synced to persistent storage yet.
   See [this article for details](https://medium.com/@valyala/wal-usage-looks-broken-in-modern-time-series-databases-b62a627ab704).
-- Cortex is usually slower and requires more CPU and RAM than VictoriaMetrics. See [this talk from Adidas at PromCon 2019](https://promcon.io/2019-munich/talks/remote-write-storage-wars/).
+- Cortex is usually slower and requires more CPU and RAM than VictoriaMetrics. See [this talk from Adidas at PromCon 2019](https://promcon.io/2019-munich/talks/remote-write-storage-wars/) and [other case studies](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies).
+- VictoriaMetrics accepts data in multiple popular data ingestion protocols additionally to Prometheus remote_write protocol - InfluxDB, OpenTSDB, Graphite, CSV.
+  See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-import-time-series-data) for details.
 
 
 ### What is the difference between VictoriaMetrics and [Thanos](https://github.com/thanos-io/thanos)?
 
 - Thanos re-uses Prometheus source code, while VictoriaMetrics is written from scratch.
-- Thanos provides [Ruler component](https://github.com/thanos-io/thanos/blob/master/docs/components/rule.md),
-  while VictoriaMetrics relies on [Promxy for alerting and recording rules](https://github.com/jacksontj/promxy#how-do-i-use-alertingrecording-rules-in-promxy).
 - VictoriaMetrics accepts data via [standard remote_write API for Prometheus](https://prometheus.io/docs/practices/remote_write/),
   while Thanos uses non-standard [Sidecar](https://github.com/thanos-io/thanos/blob/master/docs/components/sidecar.md), which must run alongside each Prometheus instance.
-- Thanos Sidecar requires disabling data compaction in Prometheus, which may hurt Prometheus performance and increase RAM usage.
-- Thanos stores data on object storage (Amazon S3 or Google GCS), while VictoriaMetrics stores data on block storage (GCP persistent disks, Amazon EBS or bare metal HDD).
+- Thanos Sidecar requires disabling data compaction in Prometheus, which may hurt Prometheus performance and increase RAM usage. See [these docs](https://thanos.io/components/sidecar.md/) for more details.
+- Thanos stores data in object storage (Amazon S3 or Google GCS), while VictoriaMetrics stores data in block storage
+  ([GCP persistent disks](https://cloud.google.com/compute/docs/disks#pdspecs), Amazon EBS or bare metal HDD).
+  While object storage is usually less expensive, block storage provides much lower latencies and higher throughput.
+  VictoriaMetrics works perfectly with HDD-based block storage - there is no need in using more expensive SSD or NVMe disks in most cases.
 - Thanos may lose up to 2 hours of recent data, which wasn't uploaded yet to object storage. VictoriaMetrics may lose only a few seconds of recent data,
   which isn't synced to persistent storage yet. See [this article for details](https://medium.com/@valyala/wal-usage-looks-broken-in-modern-time-series-databases-b62a627ab704).
+- VictoriaMetrics provides [production-ready single-node solution](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md),
+  which is much easier to setup and operate than Thanos components.
 - Thanos may be harder to setup and operate comparing to VictoriaMetrics, since it has more moving parts, which can be connected with less reliable networks.
   See [this article for details](https://medium.com/faun/comparing-thanos-to-victoriametrics-cluster-b193bea1683).
 - Thanos is usually slower and requires more CPU and RAM than VictoriaMetrics. See [this talk from Adidas at PromCon 2019](https://promcon.io/2019-munich/talks/remote-write-storage-wars/).
+- VictoriaMetrics accepts data in multiple popular data ingestion protocols additionally to Prometheus remote_write protocol - InfluxDB, OpenTSDB, Graphite, CSV.
+  See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-import-time-series-data) for details.
 
 
 ### 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.
+- 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).
+- VictoriaMetrics provides [better query language](https://medium.com/@valyala/promql-tutorial-for-beginners-9ab455142085) than InfluxQL or Flux.
+- VictoriaMetrics accepts data in multiple popular data ingestion protocols additionally to InfluxDB - Prometheus remote_write, OpenTSDB, Graphite, CSV.
+  See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#how-to-import-time-series-data) for details.
 
 
 ### 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.
+- 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).
+- 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. The gap in storage space usage can be lowered from 70x to 3x if [compression in TimescaleDB is properly configured](https://docs.timescale.com/latest/using-timescaledb/compression) (it isn't an easy task in general case :)).
+- VictoriaMetrics accepts data in multiple popular data ingestion protocols - InfluxDB, OpenTSDB, Graphite, CSV, while TimescaleDB supports only SQL inserts.
 
 
-### Does VictoriaMetrics use Prometheus technologies like other clustered TSDBs built on top of Prometheus such as [Thanos](https://github.com/thanos-io/thanos), [Cortex](https://github.com/cortexproject/cortex)?
+### Does VictoriaMetrics use Prometheus technologies like other clustered TSDBs built on top of Prometheus such as [Thanos](https://github.com/thanos-io/thanos) or [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).
@@ -136,6 +149,8 @@ Yes:
 * [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)
 
+See also [other articles about VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Articles).
+
 
 ### What is the pricing for VictoriaMetrics?
 
@@ -145,11 +160,11 @@ The following versions are open source and free:
 
 We provide commercial support for both versions. [Contact us](mailto:info@victoriametrics.com) for the pricing.
 
-The following versions are commercial:
+The following commercial versions of VictoriaMetrics are planned:
 * Managed cluster in the Cloud.
 * SaaS version.
 
-[Contact us](mailto:info@victoriametrics.com) for the pricing.
+[Contact us](mailto:info@victoriametrics.com) for more information on our plans.
 
 
 ### Why VictoriaMetrics doesn't support [Prometheus remote read API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#%3Cremote_read%3E)?
@@ -168,6 +183,11 @@ or via [Prometheus datasource in Grafana](http://docs.grafana.org/features/datas
 Yes. See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/README.md#deduplication) for details.
 
 
+### Does VictoriaMetrics support replication?
+
+Yes. See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#replication-and-data-safety) for details.
+
+
 ### Where is the source code of VictoriaMetrics?
 
 Source code for the following versions is available in the following places:

docs/MetricsQL.md

@@ -4,6 +4,8 @@ VictoriaMetrics implements MetricsQL - query language inspired by [PromQL](https
 It is backwards compatible with PromQL, so Grafana dashboards backed by Prometheus datasource should work the same after switching from Prometheus to VictoriaMetrics.
 [Standalone MetricsQL package](https://godoc.org/github.com/VictoriaMetrics/metricsql) can be used for parsing MetricsQL in external apps.
 
+If you are unfamiliar with PromQL, then it is suggested reading [this tutorial for beginners](https://medium.com/@valyala/promql-tutorial-for-beginners-9ab455142085).
+
 The following functionality is implemented differently in MetricsQL comparing to PromQL in order to improve user experience:
 * MetricsQL takes into account the previous point before the window in square brackets for range functions such as `rate` and `increase`.
   It also doesn't extrapolate range function results. This addresses [this issue from Prometheus](https://github.com/prometheus/prometheus/issues/3746).
@@ -116,3 +118,7 @@ This functionality can be tried at [an editable Grafana dashboard](http://play-g
   for the given `phi` in the range `[0..1]`.
 - `last_over_time(m[d])` - returns the last value for `m` on the time range `d`.
 - `first_over_time(m[d])` - returns the first value for `m` on the time range `d`.
+- `outliersk(N, m)` - returns up to `N` outlier time series for `m`. Outlier time series have the highest deviation from the `median(m)`.
+  This aggregate function is useful to detect anomalies across groups of similar time series.
+- `ascent_over_time(m[d])` - returns the sum of positive deltas between adjancent data points in `m` over `d`. Useful for tracking height gains in GPS track.
+- `descent_over_time(m[d])` - returns the absolute sum of negative deltas between adjancent data points in `m` over `d`. Useful for tracking height loss in GPS track.

docs/Quick-Start.md

@@ -1,7 +1,7 @@
 # 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/)
+   from [Docker hub](https://hub.docker.com/r/victoriametrics/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
@@ -17,8 +17,10 @@
    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).
+3. Configure [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) or Prometheus to write data to VictoriaMetrics.
+   It is recommended to use `vmagent` instead of Prometheus, since it is more resource efficient. If you still prefer Prometheus, then
+   see [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics#prometheus-setup)
+   for details on how to configure Prometheus.
 
 4. Configure Grafana to query VictoriaMetrics instead of Prometheus.
    See [these instructions](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/Single-server-VictoriaMetrics#grafana-setup).

docs/Single-server-VictoriaMetrics.md

@@ -10,18 +10,25 @@
 
 ## VictoriaMetrics
 
-VictoriaMetrics is fast, cost-effective and scalable time-series database. It can be used as long-term remote storage for Prometheus.
+VictoriaMetrics is fast, cost-effective and scalable time-series database.
+
 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). Just download VictoriaMetrics and see [how to start it](#how-to-start-victoriametrics).
 
 Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaMetrics/tree/cluster).
 
+See our [Wiki](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki) for additional documentation.
+
+[Contact us](mailto:info@victoriametrics.com) if you need paid enterprise support for VictoriaMetrics.
+
+
 ## Case studies and talks
 
 * [Adidas](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#adidas)
 * [CERN](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#cern)
 * [COLOPL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#colopl)
+* [Zerodha](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#zerodha)
 * [Wix.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wixcom)
 * [Wedos.com](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#wedoscom)
 * [Synthesio](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/CaseStudies#synthesio)
@@ -34,6 +41,8 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 
 ## Prominent features
 
+* VictoriaMetrics can be used as long-term storage for Prometheus or for [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md).
+  See [these docs](#prometheus-setup) for details.
 * Supports [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/), so it can be used as Prometheus drop-in replacement in Grafana.
   VictoriaMetrics implements [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/MetricsQL) query language, which is inspired by PromQL.
 * Supports global query view. Multiple Prometheus instances may write data into VictoriaMetrics. Later this data may be used in a single query.
@@ -116,6 +125,8 @@ Cluster version is available [here](https://github.com/VictoriaMetrics/VictoriaM
 * [Monitoring](#monitoring)
 * [Troubleshooting](#troubleshooting)
 * [Backfilling](#backfilling)
+* [Replication](#replication)
+* [Backups](#backups)
 * [Profiling](#profiling)
 * [Integrations](#integrations)
 * [Third-party contributions](#third-party-contributions)
@@ -571,11 +582,11 @@ Run `make package-victoria-metrics`. It builds `victoriametrics/victoria-metrics
 `<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`.
 
-By default the image is built on top of `scratch` image. It is possible to build the package on top of any other base image
-by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds the image on top of `alpine:3.11` image:
+By default the image is built on top of `alpine` image for improved debuggability. It is possible to build the package on top of any other base image
+by setting it via `<ROOT_IMAGE>` environment variable. For example, the following command builds the image on top of `scratch` image:
 
 ```bash
-ROOT_IMAGE=alpine:3.11 make package-victoria-metrics
+ROOT_IMAGE=scratch make package-victoria-metrics
 ```
 
 ### Start with docker-compose
@@ -763,7 +774,13 @@ The required resources for query path:
 ### 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:
+2) Pass addresses of these instances to [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) via `-remoteWrite.url` command-line flag:
+
+```bash
+/path/to/vmagent -remoteWrite.url=http://<victoriametrics-addr-1>:8428/api/v1/write -remoteWrite.url=http://<victoriametrics-addr-2>:8428/api/v1/write
+```
+
+Alternatively these addresses may be passed to `remote_write` section in Prometheus config:
 
 ```yml
 remote_write:
@@ -782,6 +799,8 @@ remote_write:
 kill -HUP `pidof prometheus`
 ```
 
+It is recommended to use [vmagent](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmagent/README.md) instead of Prometheus for highly loaded setups.
+
 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.
@@ -792,6 +811,7 @@ to write data to `victoriametrics-addr-1`, while each `r2` should write data to
 Another option is to write data simultaneously from Prometheus HA pair to a pair of VictoriaMetrics instances
 with the enabled de-duplication. See [this section](#deduplication) for details.
 
+
 ### Deduplication
 
 VictoriaMetrics de-duplicates data points if `-dedup.minScrapeInterval` command-line flag
@@ -809,6 +829,8 @@ Data is split in per-month subdirectories inside `<-storageDataPath>/data/small`
 Directories for months outside the configured retention are deleted on the first day of new month.
 In order to keep data according to `-retentionPeriod` max disk space usage is going to be `-retentionPeriod` + 1 month.
 For example if `-retentionPeriod` is set to 1, data for January is deleted on March 1st.
+It is safe to extend `-retentionPeriod` on existing data. If `-retentionPeriod` is set to lower
+value than before then data outside the configured period will be eventually deleted.
 
 ### Multiple retentions
 
@@ -874,6 +896,10 @@ Consider setting the following command-line flags:
 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`.
 
+Prefer authorizing all the incoming requests from untrusted networks with [vmauth](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmauth/README.md)
+or similar auth proxy.
+
+
 ### Tuning
 
 * There is no need for VictoriaMetrics tuning since it uses reasonable defaults for command-line flags,
@@ -910,6 +936,12 @@ The most interesting metrics are:
 * `sum(rate(vm_rows_inserted_total[5m]))` - ingestion rate, i.e. how many samples are inserted int the database per second.
 * `vm_free_disk_space_bytes` - free space left at `-storageDataPath`.
 * `sum(vm_data_size_bytes)` - the total size of data on disk.
+* `increase(vm_slow_row_inserts_total[5m])` - the number of slow inserts during the last 5 minutes.
+  If this number remains high during extended periods of time, then it is likely more RAM is needed for optimal handling
+  of the current number of active time series.
+* `increase(vm_slow_metric_name_loads_total[5m])` - the number of slow loads of metric names during the last 5 minutes.
+  If this number remains high during extended periods of time, then it is likely more RAM is needed for optimal handling
+  of the current number of active time series.
 
 
 ### Troubleshooting
@@ -922,8 +954,9 @@ The most interesting metrics are:
 
 * 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.
+  VictoriaMetrics [exposes](#monitoring) `vm_slow_*` metrics, which could be used as an indicator of low amounts of RAM.
   It is recommended increasing the amount of RAM on the node with VictoriaMetrics in order to improve
-  ingestion performance.
+  ingestion and query performance in this case.
   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.
 
@@ -970,6 +1003,24 @@ the query cache, which could contain incomplete data cached during the backfilli
 Yet another solution is to increase `-search.cacheTimestampOffset` flag value in order to disable caching
 for data with timestamps close to the current time.
 
+
+### Replication
+
+Single-node VictoriaMetrics doesn't support application-level replication. Use cluster version instead.
+See [these docs](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/README.md#replication-and-data-safety) for details.
+
+Storage-level replication may be offloaded to durable persistent storage such as [Google Cloud disks](https://cloud.google.com/compute/docs/disks#pdspecs).
+
+See also [high availability docs](#high-availability) and [backup docs](#backups).
+
+
+### Backups
+
+VictoriaMetrics supports backups via [vmbackup](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmbackup/README.md)
+and [vmrestore](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmrestore/README.md) tools.
+We also provide provide `vmbackuper` tool for paid enterprise subscribers - see [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/466) for details.
+
+
 ### Profiling
 
 VictoriaMetrics provides handlers for collecting the following [Go profiles](https://blog.golang.org/profiling-go-programs):

docs/vmagent.md

@@ -1,8 +1,8 @@
 ## vmagent
 
-`vmagent` is a tiny but brave agent, which helps you collecting metrics from various sources
-and storing them to [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics)
-or any other Prometheus-compatible storage system that supports `remote_write` protocol.
+`vmagent` is a tiny but brave agent, which helps you collect metrics from various sources
+and stores them in [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics)
+or any other Prometheus-compatible storage system that supports the `remote_write` protocol.
 
 <img alt="vmagent" src="vmagent.png">
 
@@ -11,7 +11,7 @@ or any other Prometheus-compatible storage system that supports `remote_write` p
 
 While VictoriaMetrics provides an efficient solution to store and observe metrics, our users needed something fast
 and RAM friendly to scrape metrics from Prometheus-compatible exporters to VictoriaMetrics.
-Also, we found that users’ infrastructure is like snowflakes - never alike, and we decided to add more flexibility
+Also, we found that users’ infrastructure are snowflakes - no two are alike, and we decided to add more flexibility
 to `vmagent` (like the ability to push metrics instead of pulling them). We did our best and plan to do even more.
 
 
@@ -31,7 +31,7 @@ to `vmagent` (like the ability to push metrics instead of pulling them). We did
 * Works in environments with unstable connections to remote storage. If the remote storage is unavailable, the collected metrics
   are buffered at `-remoteWrite.tmpDataPath`. The buffered metrics are sent to remote storage as soon as connection
   to remote storage is recovered. The maximum disk usage for the buffer can be limited with `-remoteWrite.maxDiskUsagePerURL`.
-* Uses lower amounts of RAM, CPU, disk IO and network bandwidth comparing to Prometheus.
+* Uses lower amounts of RAM, CPU, disk IO and network bandwidth compared to Prometheus.
 
 
 ### Quick Start
@@ -40,8 +40,7 @@ Just download `vmutils-*` archive from [releases page](https://github.com/Victor
 and pass the following flags to `vmagent` binary in order to start scraping Prometheus targets:
 
 * `-promscrape.config` with the path to Prometheus config file (it is usually located at `/etc/prometheus/prometheus.yml`)
-* `-remoteWrite.url` with the remote storage endpoint such as VictoriaMetrics. Multiple `-remoteWrite.url` args can be set in parallel
-  in order to replicate data concurrently to multiple remote storage systems.
+* `-remoteWrite.url` with the remote storage endpoint such as VictoriaMetrics. The `-remoteWrite.url` argument can be specified multiple times in order to replicate data concurrently to an arbitrary amount of remote storage systems.
 
 Example command line:
 
@@ -49,7 +48,7 @@ Example command line:
 /path/to/vmagent -promscrape.config=/path/to/prometheus.yml -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write
 ```
 
-If you need collecting only Influx data, then the following command line would be enough:
+If you only need to collect Influx data, then the following is sufficient:
 
 ```
 /path/to/vmagent -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write
@@ -79,14 +78,14 @@ See [the corresponding Makefile rules](https://github.com/VictoriaMetrics/Victor
 #### Drop-in replacement for Prometheus
 
 If you use Prometheus only for scraping metrics from various targets and forwarding these metrics to remote storage,
-then `vmagent` can replace such Prometheus setup. Usually `vmagent` requires lower amounts of RAM, CPU and network bandwidth comparing to Prometheus for such setup.
+then `vmagent` can replace such Prometheus setup. Usually `vmagent` requires lower amounts of RAM, CPU and network bandwidth comparing to Prometheus for such a setup.
 See [these docs](#how-to-collect-metrics-in-prometheus-format) for details.
 
 
 #### Replication and high availability
 
 `vmagent` replicates the collected metrics among multiple remote storage instances configured via `-remoteWrite.url` args.
-If a single remote storage instance temporarily goes out of service, then the collected data remains available in another remote storage instances.
+If a single remote storage instance temporarily is out of service, then the collected data remains available in another remote storage instances.
 `vmagent` buffers the collected data in files at `-remoteWrite.tmpDataPath` until the remote storage becomes available again.
 Then it sends the buffered data to the remote storage in order to prevent data gaps in the remote storage.
 
@@ -94,13 +93,13 @@ Then it sends the buffered data to the remote storage in order to prevent data g
 #### Relabeling and filtering
 
 `vmagent` can add, remove or update labels on the collected data before sending it to remote storage. Additionally,
-it can remove unneeded samples via Prometheus-like relabeling before sending the collected data to remote storage.
+it can remove unwanted samples via Prometheus-like relabeling before sending the collected data to remote storage.
 See [these docs](#relabeling) for details.
 
 
 #### Splitting data streams among multiple systems
 
-`vmagent` supports splitting of the collected data among muliple destinations with the help of `-remoteWrite.urlRelabelConfig`,
+`vmagent` supports splitting the collected data between muliple destinations with the help of `-remoteWrite.urlRelabelConfig`,
 which is applied independently for each configured `-remoteWrite.url` destination. For instance, it is possible to replicate or split
 data among long-term remote storage, short-term remote storage and real-time analytical system [built on top of Kafka](https://github.com/Telefonica/prometheus-kafka-adapter).
 Note that each destination can receive its own subset of the collected data thanks to per-destination relabeling via `-remoteWrite.urlRelabelConfig`.
@@ -148,6 +147,10 @@ The following scrape types in [scrape_config](https://prometheus.io/docs/prometh
 * `dns_sd_configs` - for scraping targets discovered from DNS records (SRV, A and AAAA).
   See [dns_sd_config](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#dns_sd_config) for details.
 
+Note that `vmagent` doesn't support `refresh_interval` option these scrape configs. Use the corresponding `-promscrape.*CheckInterval`
+command-line flag instead. For example, `-promscrape.consulSDCheckInterval=60s` sets `refresh_interval` for all the `consul_sd_configs`
+entries to 60s. Run `vmagent -help` in order to see default values for `-promscrape.*CheckInterval` flags.
+
 
 File feature requests at [our issue tracker](https://github.com/VictoriaMetrics/VictoriaMetrics/issues) if you need other service discovery mechanisms to be supported by `vmagent`.
 
@@ -200,12 +203,12 @@ either via `vmagent` itself or via Prometheus, so the exported metrics could be
 * When `vmagent` scrapes many unreliable targets, it can flood error log with scrape errors. These errors can be suppressed
   by passing `-promscrape.suppressScrapeErrors` command-line flag to `vmagent`. The most recent scrape error per each target can be observed at `http://vmagent-host:8429/targets`.
 
-* It is recommended increasing `-remoteWrite.queues` if `vmagent` collects more than 100K samples per second
+* It is recommended to increase `-remoteWrite.queues` if `vmagent` collects more than 100K samples per second
   and `vmagent_remotewrite_pending_data_bytes` metric exported at `http://vmagent-host:8429/metrics` page constantly grows.
 
 * `vmagent` buffers scraped data at `-remoteWrite.tmpDataPath` directory until it is sent to `-remoteWrite.url`.
-  The directory can grow big when remote storage is unavailable during extended periods of time and if `-remoteWrite.maxDiskUsagePerURL` isn't set.
-  If you don't want sending all the data from the directory to remote storage, just stop `vmagent` and delete the directory.
+  The directory can grow large when remote storage is unavailable for extended periods of time and if `-remoteWrite.maxDiskUsagePerURL` isn't set.
+  If you don't want to send all the data from the directory to remote storage, simply stop `vmagent` and delete the directory.
 
 
 ### How to build from sources

docs/vmalert.md

@@ -1,20 +1,18 @@
-## VM Alert
+## vmalert
 
-`vmalert` executes a list of given MetricsQL expressions (rules) and
-sends alerts to [Alert Manager](https://github.com/prometheus/alertmanager).   
+`vmalert` executes a list of given [alerting](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/)
+or [recording](https://prometheus.io/docs/prometheus/latest/configuration/recording_rules/)
+rules against configured address.
 
 ### Features:
 * Integration with [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics) TSDB;
 * VictoriaMetrics [MetricsQL](https://github.com/VictoriaMetrics/VictoriaMetrics/wiki/MetricsQL)
- expressions validation;
+ support and expressions validation;
 * Prometheus [alerting rules definition format](https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/#defining-alerting-rules)
  support;
 * Integration with [Alertmanager](https://github.com/prometheus/alertmanager);
 * Lightweight without extra dependencies.
 
-### TODO:
-* Configuration hot reload.
-
 ### QuickStart
 
 To build `vmalert` from sources:
@@ -26,9 +24,9 @@ make vmalert
 The build binary will be placed to `VictoriaMetrics/bin` folder.
 
 To start using `vmalert` you will need the following things:
-* list of alert rules - PromQL/MetricsQL expressions to execute;
+* list of rules - PromQL/MetricsQL expressions to execute;
 * datasource address - reachable VictoriaMetrics instance for rules execution;
-* notifier address - reachable Alertmanager instance for processing, 
+* notifier address - reachable [Alert Manager](https://github.com/prometheus/alertmanager) instance for processing, 
 aggregating alerts and sending notifications.
 
 Then configure `vmalert` accordingly:
@@ -38,21 +36,27 @@ Then configure `vmalert` accordingly:
         -notifier.url=http://localhost:9093
 ```
 
-Example for `.rules` file may be found [here](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/testdata/rules0-good.rules)
+Example for `.rules` file may be found [here](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/master/app/vmalert/testdata).
+
+`vmalert` may be configured with `-remoteWrite` flag to write recording rules and 
+alerts state in form of timeseries via remote write protocol. Alerts state will be written 
+as `ALERTS` timeseries. These timeseries may be used to recover alerts state on `vmalert` 
+restarts if `-remoteRead` is configured.
 
 `vmalert` runs evaluation for every group in a separate goroutine.
 Rules in group evaluated one-by-one sequentially. 
 
+**Important:** while recording rules execution is sequential, writing of timeseries results to remote
+storage is asynchronous. Hence, user shouldn't rely on recording rules chaining when result of previous
+recording rule is reused in next one.
+
 `vmalert` also runs a web-server (`-httpListenAddr`) for serving metrics and alerts endpoints:
+* `http://<vmalert-addr>/api/v1/groups` - list of all loaded groups and rules;
 * `http://<vmalert-addr>/api/v1/alerts` - list of all active alerts;
 * `http://<vmalert-addr>/api/v1/<groupName>/<alertID>/status" ` - get alert status by ID.
 Used as alert source in AlertManager.
 * `http://<vmalert-addr>/metrics` - application metrics.
-
-`vmalert` may be configured with `-remotewrite` flag to write alerts state in form of timeseries
-via remote write protocol. Alerts state will be written as `ALERTS` timeseries. These timeseries
-may be used to recover alerts state on `vmalert` restarts if `-remoteread` is configured.
-
+* `http://<vmalert-addr>/-/reload` - hot configuration reload.
 
 ### Configuration
 
@@ -65,35 +69,35 @@ Usage of vmalert:
         Optional basic auth username for -datasource.url
   -datasource.url string
         Victoria Metrics or VMSelect url. Required parameter. E.g. http://127.0.0.1:8428
-  -enableTCP6
-        Whether to enable IPv6 for listening and dialing. By default only IPv4 TCP is used
   -evaluationInterval duration
-        How often to evaluate the rules. Default 1m (default 1m0s)
+        How often to evaluate the rules (default 1m0s)
   -external.url string
         External URL is used as alert's source for sent alerts to the notifier
-  -http.maxGracefulShutdownDuration duration
-        The maximum duration for graceful shutdown of HTTP server. Highly loaded server may require increased value for graceful shutdown (default 7s)
-  -httpAuth.password string
-        Password for HTTP Basic Auth. The authentication is disabled if -httpAuth.username is empty
-  -httpAuth.username string
-        Username for HTTP Basic Auth. The authentication is disabled if empty. See also -httpAuth.password
   -httpListenAddr string
         Address to listen for http connections (default ":8880")
+  -metricsAuthKey string
+        Auth key for /metrics. It overrides httpAuth settings
   -notifier.url string
         Prometheus alertmanager URL. Required parameter. e.g. http://127.0.0.1:9093
-  -remoteread.basicAuth.password string
-        Optional basic auth password for -remoteread.url
-  -remoteread.basicAuth.username string
-        Optional basic auth username for -remoteread.url
-  -remoteread.lookback duration
+  -remoteRead.basicAuth.password string
+        Optional basic auth password for -remoteRead.url
+  -remoteRead.basicAuth.username string
+        Optional basic auth username for -remoteRead.url
+  -remoteRead.lookback duration
         Lookback defines how far to look into past for alerts timeseries. For example, if lookback=1h then range from now() to now()-1h will be scanned. (default 1h0m0s)
-  -remoteread.url vmalert
-        Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts state. This configuration makes sense only if vmalert was configured with `remotewrite.url` before and has been successfully persisted its state. E.g. http://127.0.0.1:8428
-  -remotewrite.basicAuth.password string
-        Optional basic auth password for -remotewrite.url
-  -remotewrite.basicAuth.username string
-        Optional basic auth username for -remotewrite.url
-  -remotewrite.url string
+  -remoteRead.url vmalert
+        Optional URL to Victoria Metrics or VMSelect that will be used to restore alerts state. This configuration makes sense only if vmalert was configured with `remoteWrite.url` before and has been successfully persisted its state. E.g. http://127.0.0.1:8428
+  -remoteWrite.basicAuth.password string
+        Optional basic auth password for -remoteWrite.url
+  -remoteWrite.basicAuth.username string
+        Optional basic auth username for -remoteWrite.url
+  -remoteWrite.concurrency int
+        Defines number of readers that concurrently write into remote storage (default 1)
+  -remoteWrite.maxBatchSize int
+        Defines defines max number of timeseries to be flushed at once (default 1000)
+  -remoteWrite.maxQueueSize int
+        Defines the max number of pending datapoints to remote write endpoint (default 100000)
+  -remoteWrite.url string
         Optional URL to Victoria Metrics or VMInsert where to persist alerts state in form of timeseries. E.g. http://127.0.0.1:8428
   -rule value
         Path to the file with alert rules. 
@@ -109,8 +113,30 @@ Usage of vmalert:
 Pass `-help` to `vmalert` in order to see the full list of supported 
 command-line flags with their descriptions.
 
+To reload configuration without `vmalert` restart send SIGHUP signal
+or send GET request to `/-/reload` endpoint.
+
 ### Contributing
 
 `vmalert` is mostly designed and built by VictoriaMetrics community.
 Feel free to share your experience and ideas for improving this 
 software. Please keep simplicity as the main priority.
+
+### How to build from sources
+
+It is recommended using 
+[binary releases](https://github.com/VictoriaMetrics/VictoriaMetrics/releases) 
+- `vmalert` is located in `vmutils-*` archives there.
+
+
+#### Development build
+
+1. [Install Go](https://golang.org/doc/install). The minimum supported version is Go 1.13.
+2. Run `make vmalert` from the root folder of the repository.
+   It builds `vmalert` binary and puts it into the `bin` folder.
+
+#### Production build
+
+1. [Install docker](https://docs.docker.com/install/).
+2. Run `make vmalert-prod` from the root folder of the repository.
+   It builds `vmalert-prod` binary and puts it into the `bin` folder.

docs/vmbackup.md

@@ -19,6 +19,9 @@ Backed up data can be restored with [vmrestore](https://github.com/VictoriaMetri
 
 See [this article](https://medium.com/@valyala/speeding-up-backups-for-big-time-series-databases-533c1a927883) for more details.
 
+See also [vmbackuper](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/466) tool built on top of `vmbackup`. This tool simplifies
+creation of hourly, daily, weekly and monthly backups.
+
 
 ### Use cases
 

docs/vmrestore.md

@@ -21,7 +21,8 @@ vmrestore -src=gcs://<bucket>/<path/to/backup> -storageDataPath=<local/path/to/r
 * `<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.
+The original `-storageDataPath` directory may contain old files. They will be susbstituted by the files from backup,
+i.e. the end result would be similar to [rsync --delete](https://askubuntu.com/questions/476041/how-do-i-make-rsync-delete-files-that-have-been-deleted-from-the-source-folder).
 
 
 ### Troubleshooting
@@ -52,7 +53,7 @@ Run `vmrestore -help` in order to see all the available options:
   -envflag.prefix string
     	Prefix for environment variables if -envflag.enable is set
   -fs.disableMmap
-    	Whether to use pread() instead of mmap() for reading data files
+    	Whether to use pread() instead of mmap() for reading data files. By default mmap() is used for 64-bit arches and pread() is used for 32-bit arches, since they cannot data files bigger than 2^32 bytes in memory
   -loggerFormat string
     	Format for logs. Possible values: default, json (default "default")
   -loggerLevel string
@@ -68,8 +69,8 @@ Run `vmrestore -help` in order to see all the available options:
   -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
+    	Destination path where backup must be restored. VictoriaMetrics must be stopped when restoring from backup. -storageDataPath dir can be non-empty. In this case the contents of -storageDataPath dir is synchronized with -src contents, i.e. it works like 'rsync --delete' (default "victoria-metrics-data")
+ -version
     	Show VictoriaMetrics version
 ```
 

go.mod

@@ -1,32 +1,33 @@
 module github.com/VictoriaMetrics/VictoriaMetrics
 
 require (
-	cloud.google.com/go v0.57.0 // indirect
-	cloud.google.com/go/storage v1.7.0
+	cloud.google.com/go/storage v1.8.0
 	github.com/VictoriaMetrics/fastcache v1.5.7
 
 	// Do not use the original github.com/valyala/fasthttp because of issues
 	// like https://github.com/valyala/fasthttp/commit/996610f021ff45fdc98c2ce7884d5fa4e7f9199b
 	github.com/VictoriaMetrics/fasthttp v1.0.1
-	github.com/VictoriaMetrics/metrics v1.11.2
-	github.com/VictoriaMetrics/metricsql v0.2.1
-	github.com/aws/aws-sdk-go v1.30.25
+	github.com/VictoriaMetrics/metrics v1.11.3
+	github.com/VictoriaMetrics/metricsql v0.2.3
+	github.com/aws/aws-sdk-go v1.31.5
 	github.com/cespare/xxhash/v2 v2.1.1
-	github.com/golang/protobuf v1.4.1 // indirect
+	github.com/golang/protobuf v1.4.2 // indirect
 	github.com/golang/snappy v0.0.1
-	github.com/klauspost/compress v1.10.5
+	github.com/klauspost/compress v1.10.7
 	github.com/valyala/fastjson v1.5.1
 	github.com/valyala/fastrand v1.0.0
 	github.com/valyala/gozstd v1.7.0
 	github.com/valyala/histogram v1.0.1
 	github.com/valyala/quicktemplate v1.5.0
-	golang.org/x/net v0.0.0-20200506145744-7e3656a0809f // indirect
+	golang.org/x/mod v0.3.0 // indirect
+	golang.org/x/net v0.0.0-20200520182314-0ba52f642ac2 // indirect
 	golang.org/x/oauth2 v0.0.0-20200107190931-bf48bf16ab8d
-	golang.org/x/sys v0.0.0-20200511232937-7e40ca221e25
-	golang.org/x/tools v0.0.0-20200512001501-aaeff5de670a // indirect
-	google.golang.org/api v0.24.0
-	google.golang.org/genproto v0.0.0-20200511104702-f5ebc3bea380 // indirect
-	gopkg.in/yaml.v2 v2.2.8
+	golang.org/x/sys v0.0.0-20200523222454-059865788121
+	golang.org/x/tools v0.0.0-20200527150044-688b3c5d9fa5 // indirect
+	google.golang.org/api v0.25.0
+	google.golang.org/genproto v0.0.0-20200527145253-8367513e4ece // indirect
+	gopkg.in/yaml.v2 v2.3.0
+	honnef.co/go/tools v0.0.1-2020.1.4 // indirect
 )
 
 go 1.13

go.sum

@@ -18,8 +18,8 @@ cloud.google.com/go/bigquery v1.3.0/go.mod h1:PjpwJnslEMmckchkHFfq+HTD2DmtT67aNF
 cloud.google.com/go/bigquery v1.4.0 h1:xE3CPsOgttP4ACBePh79zTKALtXwn/Edhcr16R5hMWU=
 cloud.google.com/go/bigquery v1.4.0/go.mod h1:S8dzgnTigyfTmLBfrtrhyYhwRxG72rYxvftPBK2Dvzc=
 cloud.google.com/go/bigquery v1.5.0/go.mod h1:snEHRnqQbz117VIFhE8bmtwIDY80NLUZUMb4Nv6dBIg=
-cloud.google.com/go/bigquery v1.6.0 h1:ajp/DjpiCHO71SyIhwb83YsUGAyWuzVvMko+9xCsJLw=
-cloud.google.com/go/bigquery v1.6.0/go.mod h1:hyFDG0qSGdHNz8Q6nDN8rYIkld0q/+5uBZaelxiDLfE=
+cloud.google.com/go/bigquery v1.7.0 h1:a/O/bK/vWrYGOTFtH8di4rBxMZnmkjy+Y5LxpDwo+dA=
+cloud.google.com/go/bigquery v1.7.0/go.mod h1://okPTzCYNXSlb24MZs83e2Do+h+VXtc4gLoIoXIAPc=
 cloud.google.com/go/datastore v1.0.0/go.mod h1:LXYbyblFSglQ5pkeyhO+Qmw7ukd3C+pD7TKLgZqpHYE=
 cloud.google.com/go/datastore v1.1.0 h1:/May9ojXjRkPBNVrq+oWLqmWCkr4OU5uRY29bu0mRyQ=
 cloud.google.com/go/datastore v1.1.0/go.mod h1:umbIZjpQpHh4hmRpGhH4tLFup+FVzqBi1b3c64qFpCk=
@@ -33,8 +33,8 @@ cloud.google.com/go/storage v1.0.0/go.mod h1:IhtSnM/ZTZV8YYJWCY8RULGVqBDmpoyjwiy
 cloud.google.com/go/storage v1.5.0/go.mod h1:tpKbwo567HUNpVclU5sGELwQWBDZ8gh0ZeosJ0Rtdos=
 cloud.google.com/go/storage v1.6.0 h1:UDpwYIwla4jHGzZJaEJYx1tOejbgSoNqsAfHAUYe2r8=
 cloud.google.com/go/storage v1.6.0/go.mod h1:N7U0C8pVQ/+NIKOBQyamJIeKQKkZ+mxpohlUTyfDhBk=
-cloud.google.com/go/storage v1.7.0 h1:DzdLPI8Em+DEk7IzA2a10ivq3mxIEASC9GeNJ6FFt5Q=
-cloud.google.com/go/storage v1.7.0/go.mod h1:jGMIBwF+L/tL6WN/W5InNgYYu4HP0DvGB6rQ1mufWfs=
+cloud.google.com/go/storage v1.8.0 h1:86K1Gel7BQ9/WmNWn7dTKMvTLFzwtBe5FNqYbi9X35g=
+cloud.google.com/go/storage v1.8.0/go.mod h1:Wv1Oy7z6Yz3DshWRJFhqM/UCfaWIRTdp0RXyy7KQOVs=
 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=
@@ -45,12 +45,14 @@ github.com/VictoriaMetrics/fasthttp v1.0.1 h1:I7YdbswTIW63WxoFoUOSNxeOEGB46rdKUL
 github.com/VictoriaMetrics/fasthttp v1.0.1/go.mod h1:BqgsieH90PR7x97c89j+eqZDloKkDhAEQTwhLw6jw/4=
 github.com/VictoriaMetrics/metrics v1.11.2 h1:t/ceLP6SvagUqypCKU7cI7+tQn54+TIV/tGoxihHvx8=
 github.com/VictoriaMetrics/metrics v1.11.2/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
-github.com/VictoriaMetrics/metricsql v0.2.1 h1:OI/W2QCFiQiFULVN3ZiC/iCqZFt25rXp/O7P2NiAwYU=
-github.com/VictoriaMetrics/metricsql v0.2.1/go.mod h1:UIjd9S0W1UnTWlJdM0wLS+2pfuPqjwqKoK8yTos+WyE=
+github.com/VictoriaMetrics/metrics v1.11.3 h1:eSfXc0CrquKa1VTNUvhP+dhNjLUZHQGTFfp19mYCQWE=
+github.com/VictoriaMetrics/metrics v1.11.3/go.mod h1:LU2j9qq7xqZYXz8tF3/RQnB2z2MbZms5TDiIg9/NHiQ=
+github.com/VictoriaMetrics/metricsql v0.2.3 h1:xGscDmLoeIV7+8qX/mdHnOY0vu4m+wHIVGMoy/nBovY=
+github.com/VictoriaMetrics/metricsql v0.2.3/go.mod h1:UIjd9S0W1UnTWlJdM0wLS+2pfuPqjwqKoK8yTos+WyE=
 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/aws/aws-sdk-go v1.30.25 h1:89NXJkfpjnMEnsxkP8MVX+LDsoiLCSqevraLb5y4Mjk=
-github.com/aws/aws-sdk-go v1.30.25/go.mod h1:5zCpMtNQVjRREroY7sYe8lOMRSxkhG6MZveU8YkpAk0=
+github.com/aws/aws-sdk-go v1.31.5 h1:DFA7BzTydO4etqsTja+x7UfkOKQUv1xzEluLvNk81L0=
+github.com/aws/aws-sdk-go v1.31.5/go.mod h1:5zCpMtNQVjRREroY7sYe8lOMRSxkhG6MZveU8YkpAk0=
 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=
@@ -95,6 +97,8 @@ github.com/golang/protobuf v1.4.0 h1:oOuy+ugB+P/kBdUnG5QaMXSIyJ1q38wWSojYCb3z5VQ
 github.com/golang/protobuf v1.4.0/go.mod h1:jodUvKwWbYaEsadDk5Fwe5c77LiNKVO9IDvqG2KuDX0=
 github.com/golang/protobuf v1.4.1 h1:ZFgWrT+bLgsYPirOnRfKLYJLvssAegOj/hgyMFdJZe0=
 github.com/golang/protobuf v1.4.1/go.mod h1:U8fpvMrcmy5pZrNK1lt4xCsGvpyWQ/VVv6QDs8UjoX8=
+github.com/golang/protobuf v1.4.2 h1:+Z5KGCizgyZCbGh1KZqA0fcLLkwbsjIzS4aV2v7wJX0=
+github.com/golang/protobuf v1.4.2/go.mod h1:oDoupMAO8OvCJWAcko0GGGIgR6R6ocIYbsSw735rRwI=
 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=
@@ -128,6 +132,8 @@ github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+o
 github.com/klauspost/compress v1.10.4/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs=
 github.com/klauspost/compress v1.10.5 h1:7q6vHIqubShURwQz8cQK6yIe/xC3IF0Vm7TGfqjewrc=
 github.com/klauspost/compress v1.10.5/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs=
+github.com/klauspost/compress v1.10.7 h1:7rix8v8GpI3ZBb0nSozFRgbtXKv+hOe+qfEpZqybrAg=
+github.com/klauspost/compress v1.10.7/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs=
 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=
@@ -199,6 +205,8 @@ golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzB
 golang.org/x/mod v0.1.1-0.20191107180719-034126e5016b/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
 golang.org/x/mod v0.2.0 h1:KU7oHjnv3XNWfa5COkzUifxZmxp1TyI7ImMXqFxLwvQ=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
+golang.org/x/mod v0.3.0 h1:RM4zey1++hCTbCVQfnWeKs9/IEsaBLA8vTkd0WVtmH4=
+golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 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 v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -221,6 +229,8 @@ golang.org/x/net v0.0.0-20200501053045-e0ff5e5a1de5 h1:WQ8q63x+f/zpC8Ac1s9wLElVo
 golang.org/x/net v0.0.0-20200501053045-e0ff5e5a1de5/go.mod h1:qpuaurCH72eLCgpAm/N6yyVIVM9cpaDIP3A8BGJEC5A=
 golang.org/x/net v0.0.0-20200506145744-7e3656a0809f h1:QBjCr1Fz5kw158VqdE9JfI9cJnl/ymnJWAdMuinqL7Y=
 golang.org/x/net v0.0.0-20200506145744-7e3656a0809f/go.mod h1:qpuaurCH72eLCgpAm/N6yyVIVM9cpaDIP3A8BGJEC5A=
+golang.org/x/net v0.0.0-20200520182314-0ba52f642ac2 h1:eDrdRpKgkcCqKZQwyZRyeFZgfqt37SL7Kv3tok06cKE=
+golang.org/x/net v0.0.0-20200520182314-0ba52f642ac2/go.mod h1:qpuaurCH72eLCgpAm/N6yyVIVM9cpaDIP3A8BGJEC5A=
 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/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
@@ -255,10 +265,11 @@ golang.org/x/sys v0.0.0-20200223170610-d5e6a3e2c0ae/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20200302150141-5c8b2ff67527/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200323222414-85ca7c5b95cd/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200331124033-c3d80250170d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20200409092240-59c9f1ba88fa/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200501052902-10377860bb8e/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200511232937-7e40ca221e25 h1:OKbAoGs4fGM5cPLlVQLZGYkFC8OnOfgo6tt0Smf9XhM=
 golang.org/x/sys v0.0.0-20200511232937-7e40ca221e25/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20200523222454-059865788121 h1:rITEj+UZHYC927n8GT97eC3zrpzXdb/voyeOuVKS46o=
+golang.org/x/sys v0.0.0-20200523222454-059865788121/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/text v0.0.0-20170915032832-14c0d48ead0c/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 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=
@@ -300,10 +311,10 @@ golang.org/x/tools v0.0.0-20200227222343-706bc42d1f0d/go.mod h1:TB2adYChydJhpapK
 golang.org/x/tools v0.0.0-20200304193943-95d2e580d8eb/go.mod h1:o4KQGtdN14AW+yjsvvwRTJJuXz8XRtIHtEnmAXLyFUw=
 golang.org/x/tools v0.0.0-20200312045724-11d5b4c81c7d/go.mod h1:o4KQGtdN14AW+yjsvvwRTJJuXz8XRtIHtEnmAXLyFUw=
 golang.org/x/tools v0.0.0-20200331025713-a30bf2db82d4/go.mod h1:Sl4aGygMT6LrqrWclx+PTx3U+LnKx/seiNR+3G19Ar8=
-golang.org/x/tools v0.0.0-20200409170454-77362c5149f0/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20200501065659-ab2804fb9c9d/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
-golang.org/x/tools v0.0.0-20200512001501-aaeff5de670a h1:vAa2fXRLbiVN3N/xCnodIT36K4QKZQNyQFq3hQJfQ1U=
-golang.org/x/tools v0.0.0-20200512001501-aaeff5de670a/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
+golang.org/x/tools v0.0.0-20200512131952-2bc93b1c0c88/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
+golang.org/x/tools v0.0.0-20200527150044-688b3c5d9fa5 h1:3KBjmg2slvQXATWW9cQJ6tsRc8hj1gsnwWyi1IzYk3o=
+golang.org/x/tools v0.0.0-20200527150044-688b3c5d9fa5/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543 h1:E7g+9GITq07hpfrRu66IVDexMakfv52eLZ2CXBWiKr4=
@@ -319,10 +330,11 @@ google.golang.org/api v0.17.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/
 google.golang.org/api v0.18.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/sfE=
 google.golang.org/api v0.19.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/sfE=
 google.golang.org/api v0.20.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/sfE=
-google.golang.org/api v0.21.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/sfE=
 google.golang.org/api v0.22.0/go.mod h1:BwFmGc8tA3vsd7r/7kR8DY7iEEGSU04BFxCo5jP/sfE=
 google.golang.org/api v0.24.0 h1:cG03eaksBzhfSIk7JRGctfp3lanklcOM/mTGvow7BbQ=
 google.golang.org/api v0.24.0/go.mod h1:lIXQywCXRcnZPGlsd8NbLnOjtAoL6em04bJ9+z0MncE=
+google.golang.org/api v0.25.0 h1:LodzhlzZEUfhXzNUMIfVlf9Gr6Ua5MMtoFWh7+f47qA=
+google.golang.org/api v0.25.0/go.mod h1:lIXQywCXRcnZPGlsd8NbLnOjtAoL6em04bJ9+z0MncE=
 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=
@@ -351,11 +363,13 @@ google.golang.org/genproto v0.0.0-20200228133532-8c2c7df3a383/go.mod h1:55QSHmfG
 google.golang.org/genproto v0.0.0-20200305110556-506484158171/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
 google.golang.org/genproto v0.0.0-20200312145019-da6875a35672/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
 google.golang.org/genproto v0.0.0-20200331122359-1ee6d9798940/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
-google.golang.org/genproto v0.0.0-20200409111301-baae70f3302d/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
 google.golang.org/genproto v0.0.0-20200430143042-b979b6f78d84 h1:pSLkPbrjnPyLDYUO2VM9mDLqo2V6CFBY84lFSZAfoi4=
 google.golang.org/genproto v0.0.0-20200430143042-b979b6f78d84/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
 google.golang.org/genproto v0.0.0-20200511104702-f5ebc3bea380 h1:xriR1EgvKfkKxIoU2uUvrMVl+H26359loFFUleSMXFo=
 google.golang.org/genproto v0.0.0-20200511104702-f5ebc3bea380/go.mod h1:55QSHmfGQM9UVYDPBsyGGes0y52j32PQ3BqQfXhyH3c=
+google.golang.org/genproto v0.0.0-20200526211855-cb27e3aa2013/go.mod h1:NbSheEEYHJ7i3ixzK3sjbqSGDJWnxyFXZblF3eUsNvo=
+google.golang.org/genproto v0.0.0-20200527145253-8367513e4ece h1:1YM0uhfumvoDu9sx8+RyWwTI63zoCQvI23IYFRlvte0=
+google.golang.org/genproto v0.0.0-20200527145253-8367513e4ece/go.mod h1:jDfRM7FcilCzHH/e9qn6dsT145K34l5v+OpcnNgKAAA=
 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=
@@ -365,7 +379,6 @@ google.golang.org/grpc v1.26.0/go.mod h1:qbnxyOmOxrQa7FizSgH+ReBfzJrCY1pSN7KXBS8
 google.golang.org/grpc v1.27.0/go.mod h1:qbnxyOmOxrQa7FizSgH+ReBfzJrCY1pSN7KXBS8abTk=
 google.golang.org/grpc v1.27.1/go.mod h1:qbnxyOmOxrQa7FizSgH+ReBfzJrCY1pSN7KXBS8abTk=
 google.golang.org/grpc v1.28.0/go.mod h1:rpkK4SK4GF4Ach/+MFLZUBavHOvF2JJB5uozKKal+60=
-google.golang.org/grpc v1.28.1/go.mod h1:rpkK4SK4GF4Ach/+MFLZUBavHOvF2JJB5uozKKal+60=
 google.golang.org/grpc v1.29.1 h1:EC2SB8S04d2r73uptxphDSUG+kTKVgjRPF+N3xpxRB4=
 google.golang.org/grpc v1.29.1/go.mod h1:itym6AZVZYACWQqET3MqgPpjcuV5QH3BxFS3IjizoKk=
 google.golang.org/protobuf v0.0.0-20200109180630-ec00e32a8dfd/go.mod h1:DFci5gLYBciE7Vtevhsrf46CRTquxDuWsQurQQe4oz8=
@@ -376,13 +389,18 @@ google.golang.org/protobuf v1.21.0 h1:qdOKuR/EIArgaWNjetjgTzgVTAZ+S/WXVrq9HW9zim
 google.golang.org/protobuf v1.21.0/go.mod h1:47Nbq4nVaFHyn7ilMalzfO3qCViNmqZ2kzikPIcrTAo=
 google.golang.org/protobuf v1.22.0 h1:cJv5/xdbk1NnMPR1VP9+HU6gupuG9MLBoH1r6RHZ2MY=
 google.golang.org/protobuf v1.22.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.23.0 h1:4MY060fB1DLGMB/7MBTLnwQUY6+F09GEiz6SsrNqyzM=
+google.golang.org/protobuf v1.23.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.23.1-0.20200526195155-81db48ad09cc/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
+google.golang.org/protobuf v1.24.0 h1:UhZDfRO8JRQru4/+LlLE0BRKGF8L+PICnvYZmx/fEGA=
+google.golang.org/protobuf v1.24.0/go.mod h1:r/3tXBNzIEhYS9I1OUVjXDlt8tc493IdKGjtUeSXeh4=
 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/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
-gopkg.in/yaml.v2 v2.2.8 h1:obN1ZagJSUGI0Ek/LBmuj4SNLPfIny3KsKFopxRdj10=
-gopkg.in/yaml.v2 v2.2.8/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
+gopkg.in/yaml.v2 v2.3.0 h1:clyUAQHOM3G0M3f5vQj7LuJrETvjVot3Z5el9nffUtU=
+gopkg.in/yaml.v2 v2.3.0/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=
@@ -390,6 +408,8 @@ honnef.co/go/tools v0.0.0-20190523083050-ea95bdfd59fc/go.mod h1:rf3lG4BRIbNafJWh
 honnef.co/go/tools v0.0.1-2019.2.3/go.mod h1:a3bituU0lyd329TUQxRnasdCoJDkEUEAqEt0JzvZhAg=
 honnef.co/go/tools v0.0.1-2020.1.3 h1:sXmLre5bzIR6ypkjXCDI3jHPssRhc8KD/Ome589sc3U=
 honnef.co/go/tools v0.0.1-2020.1.3/go.mod h1:X/FiERA/W4tHapMX5mGpAtMSVEeEUOyHaw9vFzvIQ3k=
+honnef.co/go/tools v0.0.1-2020.1.4 h1:UoveltGrhghAA7ePc+e+QYDHXrBps2PqFZiHkGR/xK8=
+honnef.co/go/tools v0.0.1-2020.1.4/go.mod h1:X/FiERA/W4tHapMX5mGpAtMSVEeEUOyHaw9vFzvIQ3k=
 rsc.io/binaryregexp v0.2.0/go.mod h1:qTv7/COck+e2FymRvadv62gMdZztPaShugOCi3I+8D8=
 rsc.io/quote/v3 v3.1.0/go.mod h1:yEA65RcK8LyAZtP9Kv3t0HmxON59tX3rD+tICJqUlj0=
 rsc.io/sampler v1.3.0/go.mod h1:T1hPZKmBbMNahiBKFy5HrXp6adAjACjK9JXDnKaTXpA=

lib/backup/actions/restore.go

@@ -113,10 +113,16 @@ func (r *Restore) Run() error {
 	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.
+		// Remove only files with the missing part at offset 0.
+		// Assume other files are partially downloaded during the previous Restore.Run call,
+		// so only the last part in them may be incomplete.
+		// The last part for partially downloaded files will be re-downloaded later.
+		// This addresses https://github.com/VictoriaMetrics/VictoriaMetrics/issues/487 .
 		pathsToDelete := make(map[string]bool)
 		for _, p := range partsToDelete {
-			pathsToDelete[p.Path] = true
+			if p.Offset == 0 {
+				pathsToDelete[p.Path] = true
+			}
 		}
 		logger.Infof("deleting %d files from %s", len(pathsToDelete), dst)
 		for path := range pathsToDelete {
@@ -153,7 +159,8 @@ func (r *Restore) Run() error {
 		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.
+			// Sort partsToCopy in order to properly grow file size during downloading
+			// and to properly resume downloading of incomplete files on the next Restore.Run call.
 			common.SortParts(parts)
 			for _, p := range parts {
 				logger.Infof("downloading %s from %s to %s", &p, src, dst)
@@ -169,7 +176,7 @@ func (r *Restore) Run() error {
 					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 fmt.Errorf("cannot close reader from %s from %s: %s", &p, src, err)
 				}
 			}
 			return nil

lib/backup/actions/util.go

@@ -230,6 +230,6 @@ func NewRemoteFS(path string) (common.RemoteFS, error) {
 		}
 		return fs, nil
 	default:
-		return nil, fmt.Errorf("unsupported scheme %q in `-dst`", scheme)
+		return nil, fmt.Errorf("unsupported scheme %q", scheme)
 	}
 }

lib/backup/fsremote/fsremote.go

@@ -210,6 +210,9 @@ func (fs *FS) DeleteFile(filePath string) error {
 // The file is overwritten if it exists.
 func (fs *FS) CreateFile(filePath string, data []byte) error {
 	path := filepath.Join(fs.Dir, filePath)
+	if err := fs.mkdirAll(path); err != nil {
+		return err
+	}
 	if err := ioutil.WriteFile(path, data, 0600); err != nil {
 		return fmt.Errorf("cannot write %d bytes to %q: %s", len(data), path, err)
 	}

lib/bytesutil/bytesutil.go

@@ -2,6 +2,7 @@ package bytesutil
 
 import (
 	"reflect"
+	"runtime"
 	"unsafe"
 )
 
@@ -29,5 +30,7 @@ func ToUnsafeBytes(s string) []byte {
 	slh.Data = sh.Data
 	slh.Len = sh.Len
 	slh.Cap = sh.Len
-	return *(*[]byte)(unsafe.Pointer(&slh))
+	b := *(*[]byte)(unsafe.Pointer(&slh))
+	runtime.KeepAlive(s)
+	return b
 }

lib/fasttime/fasttime.go

@@ -0,0 +1,40 @@
+package fasttime
+
+import (
+	"sync/atomic"
+	"time"
+)
+
+func init() {
+	go func() {
+		ticker := time.NewTicker(time.Second)
+		defer ticker.Stop()
+		for tm := range ticker.C {
+			t := uint64(tm.Unix())
+			atomic.StoreUint64(&currentTimestamp, t)
+		}
+	}()
+}
+
+var currentTimestamp = uint64(time.Now().Unix())
+
+// UnixTimestamp returns the current unix timestamp in seconds.
+//
+// It is faster than time.Now().Unix()
+func UnixTimestamp() uint64 {
+	return atomic.LoadUint64(&currentTimestamp)
+}
+
+// UnixDate returns date from the current unix timestamp.
+//
+// The date is calculated by dividing unix timestamp by (24*3600)
+func UnixDate() uint64 {
+	return UnixTimestamp() / (24 * 3600)
+}
+
+// UnixHour returns hour from the current unix timestamp.
+//
+// The hour is calculated by dividing unix timestamp by 3600
+func UnixHour() uint64 {
+	return UnixTimestamp() / 3600
+}

lib/fasttime/fasttime_test.go

@@ -0,0 +1,30 @@
+package fasttime
+
+import (
+	"testing"
+	"time"
+)
+
+func TestUnixTimestamp(t *testing.T) {
+	tsExpected := uint64(time.Now().Unix())
+	ts := UnixTimestamp()
+	if ts-tsExpected > 1 {
+		t.Fatalf("unexpected UnixTimestamp; got %d; want %d", ts, tsExpected)
+	}
+}
+
+func TestUnixDate(t *testing.T) {
+	dateExpected := uint64(time.Now().Unix() / (24 * 3600))
+	date := UnixDate()
+	if date-dateExpected > 1 {
+		t.Fatalf("unexpected UnixDate; got %d; want %d", date, dateExpected)
+	}
+}
+
+func TestUnixHour(t *testing.T) {
+	hourExpected := uint64(time.Now().Unix() / 3600)
+	hour := UnixHour()
+	if hour-hourExpected > 1 {
+		t.Fatalf("unexpected UnixHour; got %d; want %d", hour, hourExpected)
+	}
+}

lib/fasttime/fasttime_timing_test.go

@@ -0,0 +1,32 @@
+package fasttime
+
+import (
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func BenchmarkUnixTimestamp(b *testing.B) {
+	b.ReportAllocs()
+	b.RunParallel(func(pb *testing.PB) {
+		var ts uint64
+		for pb.Next() {
+			ts += UnixTimestamp()
+		}
+		atomic.StoreUint64(&Sink, ts)
+	})
+}
+
+func BenchmarkTimeNowUnix(b *testing.B) {
+	b.ReportAllocs()
+	b.RunParallel(func(pb *testing.PB) {
+		var ts uint64
+		for pb.Next() {
+			ts += uint64(time.Now().Unix())
+		}
+		atomic.StoreUint64(&Sink, ts)
+	})
+}
+
+// Sink should prevent from code elimination by optimizing compiler
+var Sink uint64

lib/fs/fs.go

@@ -6,8 +6,10 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"sync"
 	"sync/atomic"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/filestream"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"golang.org/x/sys/unix"
@@ -296,6 +298,35 @@ func CreateFlockFile(dir string) (*os.File, error) {
 
 // MustGetFreeSpace returns free space for the given directory path.
 func MustGetFreeSpace(path string) uint64 {
+	// Try obtaining cached value at first.
+	freeSpaceMapLock.Lock()
+	defer freeSpaceMapLock.Unlock()
+
+	e, ok := freeSpaceMap[path]
+	if ok && fasttime.UnixTimestamp()-e.updateTime < 2 {
+		// Fast path - the entry is fresh.
+		return e.freeSpace
+	}
+
+	// Slow path.
+	// Determine the amount of free space at path.
+	e.freeSpace = mustGetFreeSpace(path)
+	e.updateTime = fasttime.UnixTimestamp()
+	freeSpaceMap[path] = e
+	return e.freeSpace
+}
+
+var (
+	freeSpaceMap     = make(map[string]freeSpaceEntry)
+	freeSpaceMapLock sync.Mutex
+)
+
+type freeSpaceEntry struct {
+	updateTime uint64
+	freeSpace  uint64
+}
+
+func mustGetFreeSpace(path string) uint64 {
 	d, err := os.Open(path)
 	if err != nil {
 		logger.Panicf("FATAL: cannot determine free disk space on %q: %s", path, err)

lib/fs/reader_at.go

@@ -4,6 +4,9 @@ import (
 	"flag"
 	"fmt"
 	"os"
+	"sync"
+	"sync/atomic"
+	"time"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
 	"github.com/VictoriaMetrics/metrics"
@@ -28,6 +31,18 @@ type MustReadAtCloser interface {
 type ReaderAt struct {
 	f        *os.File
 	mmapData []byte
+
+	// pageCacheBitmap holds a bitmap for recently touched pages in mmapData.
+	// This bitmap allows using simple copy() instead of copyMmap() for reading recently touched pages,
+	// which is up to 4x faster when reading small chunks of data via MustReadAt.
+	pageCacheBitmap   atomic.Value
+	pageCacheBitmapWG sync.WaitGroup
+
+	stopCh chan struct{}
+}
+
+type pageCacheBitmap struct {
+	m []uint64
 }
 
 // MustReadAt reads len(p) bytes at off from r.
@@ -35,7 +50,11 @@ func (r *ReaderAt) MustReadAt(p []byte, off int64) {
 	if len(p) == 0 {
 		return
 	}
-	if len(r.mmapData) == 0 || len(p) > 8*1024 {
+	if off < 0 {
+		logger.Panicf("off=%d cannot be negative", off)
+	}
+	end := off + int64(len(p))
+	if len(r.mmapData) == 0 || (len(p) > 8*1024 && !r.isInPageCache(off, end)) {
 		// Read big blocks directly from file.
 		// This could be faster than reading these blocks from mmap,
 		// since it triggers less page faults.
@@ -46,18 +65,72 @@ func (r *ReaderAt) MustReadAt(p []byte, off int64) {
 		if n != len(p) {
 			logger.Panicf("FATAL: unexpected number of bytes read; got %d; want %d", n, len(p))
 		}
+		if len(r.mmapData) > 0 {
+			r.markInPageCache(off, end)
+		}
 	} else {
-		if off < 0 || off > int64(len(r.mmapData)-len(p)) {
+		if off > int64(len(r.mmapData)-len(p)) {
 			logger.Panicf("off=%d is out of allowed range [0...%d] for len(p)=%d", off, len(r.mmapData)-len(p), len(p))
 		}
-		copyMmap(p, r.mmapData[off:])
+		src := r.mmapData[off:]
+		if r.isInPageCache(off, end) {
+			// It is safe copying the data with copy(), since it is likely it is in the page cache.
+			// This is up to 4x faster than copyMmap() below.
+			copy(p, src)
+		} else {
+			// The data may be missing in the page cache, so it is better to copy it via cgo trick
+			// in order to avoid P stalls in Go runtime.
+			// See https://medium.com/@valyala/mmap-in-go-considered-harmful-d92a25cb161d for details.
+			copyMmap(p, src)
+			r.markInPageCache(off, end)
+		}
 	}
 	readCalls.Inc()
 	readBytes.Add(len(p))
 }
 
+func (r *ReaderAt) isInPageCache(start, end int64) bool {
+	startBit := uint64(start) / pageSize
+	endBit := uint64(end) / pageSize
+	m := r.pageCacheBitmap.Load().(*pageCacheBitmap).m
+	for startBit <= endBit {
+		idx := startBit / 64
+		off := startBit % 64
+		if idx >= uint64(len(m)) {
+			return true
+		}
+		n := atomic.LoadUint64(&m[idx])
+		if (n>>off)&1 != 1 {
+			return false
+		}
+		startBit++
+	}
+	return true
+}
+
+func (r *ReaderAt) markInPageCache(start, end int64) {
+	startBit := uint64(start) / pageSize
+	endBit := uint64(end) / pageSize
+	m := r.pageCacheBitmap.Load().(*pageCacheBitmap).m
+	for startBit <= endBit {
+		idx := startBit / 64
+		off := startBit % 64
+		n := atomic.LoadUint64(&m[idx])
+		n |= 1 << off
+		// It is OK if multiple concurrent goroutines store the same m[idx].
+		atomic.StoreUint64(&m[idx], n)
+		startBit++
+	}
+}
+
+// Assume page size is 4KB
+const pageSize = 4 * 1024
+
 // MustClose closes r.
 func (r *ReaderAt) MustClose() {
+	close(r.stopCh)
+	r.pageCacheBitmapWG.Wait()
+
 	fname := r.f.Name()
 	if len(r.mmapData) > 0 {
 		if err := unix.Munmap(r.mmapData); err != nil {
@@ -80,8 +153,24 @@ func OpenReaderAt(path string) (*ReaderAt, error) {
 	}
 	var r ReaderAt
 	r.f = f
+	r.stopCh = make(chan struct{})
 	if !*disableMmap {
-		data, err := mmapFile(f)
+		fi, err := f.Stat()
+		if err != nil {
+			return nil, fmt.Errorf("error in stat: %s", err)
+		}
+		size := fi.Size()
+		bm := &pageCacheBitmap{
+			m: make([]uint64, 1+size/pageSize/64),
+		}
+		r.pageCacheBitmap.Store(bm)
+		r.pageCacheBitmapWG.Add(1)
+		go func() {
+			defer r.pageCacheBitmapWG.Done()
+			pageCacheBitmapCleaner(&r.pageCacheBitmap, r.stopCh)
+		}()
+
+		data, err := mmapFile(f, size)
 		if err != nil {
 			MustClose(f)
 			return nil, fmt.Errorf("cannot init reader for %q: %s", path, err)
@@ -92,18 +181,30 @@ func OpenReaderAt(path string) (*ReaderAt, error) {
 	return &r, nil
 }
 
+func pageCacheBitmapCleaner(pcbm *atomic.Value, stopCh <-chan struct{}) {
+	t := time.NewTimer(time.Minute)
+	for {
+		select {
+		case <-stopCh:
+			t.Stop()
+			return
+		case <-t.C:
+		}
+		bmOld := pcbm.Load().(*pageCacheBitmap)
+		bm := &pageCacheBitmap{
+			m: make([]uint64, len(bmOld.m)),
+		}
+		pcbm.Store(bm)
+	}
+}
+
 var (
 	readCalls    = metrics.NewCounter(`vm_fs_read_calls_total`)
 	readBytes    = metrics.NewCounter(`vm_fs_read_bytes_total`)
 	readersCount = metrics.NewCounter(`vm_fs_readers`)
 )
 
-func mmapFile(f *os.File) ([]byte, error) {
-	fi, err := f.Stat()
-	if err != nil {
-		return nil, fmt.Errorf("error in stat: %s", err)
-	}
-	size := fi.Size()
+func mmapFile(f *os.File, size int64) ([]byte, error) {
 	if size == 0 {
 		return nil, nil
 	}

lib/fs/reader_at_test.go

@@ -0,0 +1,37 @@
+package fs
+
+import (
+	"fmt"
+	"io/ioutil"
+	"testing"
+)
+
+func TestReaderAt(t *testing.T) {
+	for _, bufSize := range []int{1, 1e1, 1e2, 1e3, 1e4, 1e5} {
+		t.Run(fmt.Sprintf("%d", bufSize), func(t *testing.T) {
+			testReaderAt(t, bufSize)
+		})
+	}
+}
+
+func testReaderAt(t *testing.T, bufSize int) {
+	path := "TestReaderAt"
+	const fileSize = 8 * 1024 * 1024
+	data := make([]byte, fileSize)
+	if err := ioutil.WriteFile(path, data, 0600); err != nil {
+		t.Fatalf("cannot create %q: %s", path, err)
+	}
+	defer MustRemoveAll(path)
+	r, err := OpenReaderAt(path)
+	if err != nil {
+		t.Fatalf("error in OpenReaderAt(%q): %s", path, err)
+	}
+	defer r.MustClose()
+
+	buf := make([]byte, bufSize)
+	for i := 0; i < fileSize-bufSize; i += bufSize {
+		offset := int64(i)
+		r.MustReadAt(buf[:0], offset)
+		r.MustReadAt(buf, offset)
+	}
+}

lib/fs/reader_at_timing_test.go

@@ -0,0 +1,56 @@
+package fs
+
+import (
+	"fmt"
+	"io/ioutil"
+	"testing"
+)
+
+func BenchmarkReaderAtMustReadAt(b *testing.B) {
+	b.Run("mmap_on", func(b *testing.B) {
+		benchmarkReaderAtMustReadAt(b, true)
+	})
+	b.Run("mmap_off", func(b *testing.B) {
+		benchmarkReaderAtMustReadAt(b, false)
+	})
+}
+
+func benchmarkReaderAtMustReadAt(b *testing.B, isMmap bool) {
+	prevDisableMmap := *disableMmap
+	*disableMmap = !isMmap
+	defer func() {
+		*disableMmap = prevDisableMmap
+	}()
+
+	path := "BenchmarkReaderAtMustReadAt"
+	const fileSize = 8 * 1024 * 1024
+	data := make([]byte, fileSize)
+	if err := ioutil.WriteFile(path, data, 0600); err != nil {
+		b.Fatalf("cannot create %q: %s", path, err)
+	}
+	defer MustRemoveAll(path)
+	r, err := OpenReaderAt(path)
+	if err != nil {
+		b.Fatalf("error in OpenReaderAt(%q): %s", path, err)
+	}
+	defer r.MustClose()
+
+	b.ResetTimer()
+	for _, bufSize := range []int{1, 1e1, 1e2, 1e3, 1e4, 1e5} {
+		b.Run(fmt.Sprintf("%d", bufSize), func(b *testing.B) {
+			b.ReportAllocs()
+			b.SetBytes(int64(bufSize))
+			b.RunParallel(func(pb *testing.PB) {
+				buf := make([]byte, bufSize)
+				var offset int64
+				for pb.Next() {
+					if len(buf)+int(offset) > fileSize {
+						offset = 0
+					}
+					r.MustReadAt(buf, offset)
+					offset += int64(len(buf))
+				}
+			})
+		})
+	}
+}

lib/httpserver/httpserver.go

@@ -283,10 +283,10 @@ func maybeGzipResponseWriter(w http.ResponseWriter, r *http.Request) http.Respon
 	ae = strings.ToLower(ae)
 	n := strings.Index(ae, "gzip")
 	if n < 0 {
+		// Do not apply gzip encoding to the response.
 		return w
 	}
-	h := w.Header()
-	h.Set("Content-Encoding", "gzip")
+	// Apply gzip encoding to the response.
 	zw := getGzipWriter(w)
 	bw := getBufioWriter(zw)
 	zrw := &gzipResponseWriter{
@@ -301,7 +301,14 @@ func maybeGzipResponseWriter(w http.ResponseWriter, r *http.Request) http.Respon
 //
 // The function must be called before the first w.Write* call.
 func DisableResponseCompression(w http.ResponseWriter) {
-	w.Header().Del("Content-Encoding")
+	zrw, ok := w.(*gzipResponseWriter)
+	if !ok {
+		return
+	}
+	if zrw.firstWriteDone {
+		logger.Panicf("BUG: DisableResponseCompression must be called before sending the response")
+	}
+	zrw.disableCompression = true
 }
 
 // EnableCORS enables https://developer.mozilla.org/en-US/docs/Web/HTTP/CORS
@@ -343,20 +350,24 @@ type gzipResponseWriter struct {
 func (zrw *gzipResponseWriter) Write(p []byte) (int, error) {
 	if !zrw.firstWriteDone {
 		h := zrw.Header()
-		if h.Get("Content-Encoding") != "gzip" {
-			// The request handler disabled gzip encoding.
-			// Send uncompressed response body.
+		if zrw.statusCode == http.StatusNoContent {
 			zrw.disableCompression = true
-		} else if h.Get("Content-Type") == "" {
-			// Disable auto-detection of content-type, since it
-			// is incorrectly detected after the compression.
-			h.Set("Content-Type", "text/html")
 		}
+		if h.Get("Content-Encoding") != "" {
+			zrw.disableCompression = true
+		}
+		if !zrw.disableCompression {
+			h.Set("Content-Encoding", "gzip")
+			h.Del("Content-Length")
+			if h.Get("Content-Type") == "" {
+				// Disable auto-detection of content-type, since it
+				// is incorrectly detected after the compression.
+				h.Set("Content-Type", "text/html")
+			}
+		}
+		zrw.writeHeader()
 		zrw.firstWriteDone = true
 	}
-	if zrw.statusCode == 0 {
-		zrw.WriteHeader(http.StatusOK)
-	}
 	if zrw.disableCompression {
 		return zrw.ResponseWriter.Write(p)
 	}
@@ -364,23 +375,25 @@ func (zrw *gzipResponseWriter) Write(p []byte) (int, error) {
 }
 
 func (zrw *gzipResponseWriter) WriteHeader(statusCode int) {
-	if zrw.statusCode != 0 {
-		return
-	}
-	if statusCode == http.StatusNoContent {
-		DisableResponseCompression(zrw.ResponseWriter)
-	}
-	zrw.ResponseWriter.WriteHeader(statusCode)
 	zrw.statusCode = statusCode
 }
 
+func (zrw *gzipResponseWriter) writeHeader() {
+	if zrw.statusCode == 0 {
+		zrw.statusCode = http.StatusOK
+	}
+	zrw.ResponseWriter.WriteHeader(zrw.statusCode)
+}
+
 // Implements http.Flusher
 func (zrw *gzipResponseWriter) Flush() {
-	if err := zrw.bw.Flush(); err != nil && !isTrivialNetworkError(err) {
-		logger.Warnf("gzipResponseWriter.Flush (buffer): %s", err)
-	}
-	if err := zrw.zw.Flush(); err != nil && !isTrivialNetworkError(err) {
-		logger.Warnf("gzipResponseWriter.Flush (gzip): %s", err)
+	if !zrw.disableCompression {
+		if err := zrw.bw.Flush(); err != nil && !isTrivialNetworkError(err) {
+			logger.Warnf("gzipResponseWriter.Flush (buffer): %s", err)
+		}
+		if err := zrw.zw.Flush(); err != nil && !isTrivialNetworkError(err) {
+			logger.Warnf("gzipResponseWriter.Flush (gzip): %s", err)
+		}
 	}
 	if fw, ok := zrw.ResponseWriter.(http.Flusher); ok {
 		fw.Flush()
@@ -389,11 +402,14 @@ func (zrw *gzipResponseWriter) Flush() {
 
 func (zrw *gzipResponseWriter) Close() error {
 	if !zrw.firstWriteDone {
-		zrw.Header().Del("Content-Encoding")
+		zrw.writeHeader()
 		return nil
 	}
 	zrw.Flush()
-	err := zrw.zw.Close()
+	var err error
+	if !zrw.disableCompression {
+		err = zrw.zw.Close()
+	}
 	putGzipWriter(zrw.zw)
 	zrw.zw = nil
 	putBufioWriter(zrw.bw)

lib/memory/memory.go

@@ -24,8 +24,8 @@ func initOnce() {
 		// Do not use logger.Panicf here, since logger may be uninitialized yet.
 		panic(fmt.Errorf("BUG: memory.Allowed must be called only after flag.Parse call"))
 	}
-	if *allowedMemPercent < 10 || *allowedMemPercent > 200 {
-		logger.Panicf("FATAL: -memory.allowedPercent must be in the range [10...200]; got %f", *allowedMemPercent)
+	if *allowedMemPercent < 1 || *allowedMemPercent > 200 {
+		logger.Panicf("FATAL: -memory.allowedPercent must be in the range [1...200]; got %f", *allowedMemPercent)
 	}
 	percent := *allowedMemPercent / 100
 

lib/mergeset/block_header.go

@@ -148,18 +148,18 @@ func unmarshalBlockHeaders(dst []blockHeader, src []byte, blockHeadersCount int)
 	for i := 0; i < blockHeadersCount; i++ {
 		tail, err := dst[dstLen+i].Unmarshal(src)
 		if err != nil {
-			return nil, fmt.Errorf("cannot unmarshal block header: %s", err)
+			return dst, fmt.Errorf("cannot unmarshal block header: %s", err)
 		}
 		src = tail
 	}
 	if len(src) > 0 {
-		return nil, fmt.Errorf("unexpected non-zero tail left after unmarshaling %d block headers; len(tail)=%d", blockHeadersCount, len(src))
+		return dst, fmt.Errorf("unexpected non-zero tail left after unmarshaling %d block headers; len(tail)=%d", blockHeadersCount, len(src))
 	}
 	newBHS := dst[dstLen:]
 
 	// Verify that block headers are sorted by firstItem.
 	if !sort.SliceIsSorted(newBHS, func(i, j int) bool { return string(newBHS[i].firstItem) < string(newBHS[j].firstItem) }) {
-		return nil, fmt.Errorf("block headers must be sorted by firstItem; unmarshaled unsorted block headers: %#v", newBHS)
+		return dst, fmt.Errorf("block headers must be sorted by firstItem; unmarshaled unsorted block headers: %#v", newBHS)
 	}
 
 	return dst, nil

lib/mergeset/block_stream_writer.go

@@ -63,9 +63,12 @@ func (bsw *blockStreamWriter) reset() {
 	bsw.mrFirstItemCaught = false
 }
 
-func (bsw *blockStreamWriter) InitFromInmemoryPart(ip *inmemoryPart, compressLevel int) {
+func (bsw *blockStreamWriter) InitFromInmemoryPart(ip *inmemoryPart) {
 	bsw.reset()
-	bsw.compressLevel = compressLevel
+
+	// Use the minimum compression level for in-memory blocks,
+	// since they are going to be re-compressed during the merge into file-based blocks.
+	bsw.compressLevel = -5 // See https://github.com/facebook/zstd/releases/tag/v1.3.4
 
 	bsw.metaindexWriter = &ip.metaindexData
 	bsw.indexWriter = &ip.indexData

lib/mergeset/encoding.go

@@ -151,7 +151,7 @@ var isInTest = func() bool {
 	return strings.HasSuffix(os.Args[0], ".test")
 }()
 
-// MarshalUnsortedData marshals sorted items from ib to sb.
+// MarshalSortedData marshals sorted items from ib to sb.
 //
 // It also:
 // - appends first item to firstItemDst and returns the result.

lib/mergeset/merge_test.go

@@ -29,14 +29,14 @@ func TestMultilevelMerge(t *testing.T) {
 	// First level merge
 	var dstIP1 inmemoryPart
 	var bsw1 blockStreamWriter
-	bsw1.InitFromInmemoryPart(&dstIP1, 0)
+	bsw1.InitFromInmemoryPart(&dstIP1)
 	if err := mergeBlockStreams(&dstIP1.ph, &bsw1, bsrs[:5], nil, nil, &itemsMerged); err != nil {
 		t.Fatalf("cannot merge first level part 1: %s", err)
 	}
 
 	var dstIP2 inmemoryPart
 	var bsw2 blockStreamWriter
-	bsw2.InitFromInmemoryPart(&dstIP2, 0)
+	bsw2.InitFromInmemoryPart(&dstIP2)
 	if err := mergeBlockStreams(&dstIP2.ph, &bsw2, bsrs[5:], nil, nil, &itemsMerged); err != nil {
 		t.Fatalf("cannot merge first level part 2: %s", err)
 	}
@@ -53,7 +53,7 @@ func TestMultilevelMerge(t *testing.T) {
 		newTestBlockStreamReader(&dstIP1),
 		newTestBlockStreamReader(&dstIP2),
 	}
-	bsw.InitFromInmemoryPart(&dstIP, 0)
+	bsw.InitFromInmemoryPart(&dstIP)
 	if err := mergeBlockStreams(&dstIP.ph, &bsw, bsrsTop, nil, nil, &itemsMerged); err != nil {
 		t.Fatalf("cannot merge second level: %s", err)
 	}
@@ -72,7 +72,7 @@ func TestMergeForciblyStop(t *testing.T) {
 	bsrs, _ := newTestInmemoryBlockStreamReaders(20, 4000)
 	var dstIP inmemoryPart
 	var bsw blockStreamWriter
-	bsw.InitFromInmemoryPart(&dstIP, 0)
+	bsw.InitFromInmemoryPart(&dstIP)
 	ch := make(chan struct{})
 	var itemsMerged uint64
 	close(ch)
@@ -119,7 +119,7 @@ func testMergeBlockStreamsSerial(blocksToMerge, maxItemsPerBlock int) error {
 	var itemsMerged uint64
 	var dstIP inmemoryPart
 	var bsw blockStreamWriter
-	bsw.InitFromInmemoryPart(&dstIP, 0)
+	bsw.InitFromInmemoryPart(&dstIP)
 	if err := mergeBlockStreams(&dstIP.ph, &bsw, bsrs, nil, nil, &itemsMerged); err != nil {
 		return fmt.Errorf("cannot merge block streams: %s", err)
 	}

lib/mergeset/part.go

@@ -7,6 +7,7 @@ import (
 	"sync/atomic"
 	"time"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/filestream"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fs"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/memory"
@@ -227,7 +228,7 @@ func (idxbc *indexBlockCache) cleaner() {
 }
 
 func (idxbc *indexBlockCache) cleanByTimeout() {
-	currentTime := atomic.LoadUint64(&currentTimestamp)
+	currentTime := fasttime.UnixTimestamp()
 	idxbc.mu.Lock()
 	for k, idxbe := range idxbc.m {
 		// Delete items accessed more than 10 minutes ago.
@@ -245,7 +246,7 @@ func (idxbc *indexBlockCache) Get(k uint64) *indexBlock {
 	idxbc.mu.RUnlock()
 
 	if idxbe != nil {
-		currentTime := atomic.LoadUint64(&currentTimestamp)
+		currentTime := fasttime.UnixTimestamp()
 		if atomic.LoadUint64(&idxbe.lastAccessTime) != currentTime {
 			atomic.StoreUint64(&idxbe.lastAccessTime, currentTime)
 		}
@@ -256,9 +257,7 @@ func (idxbc *indexBlockCache) Get(k uint64) *indexBlock {
 }
 
 // Put puts idxb under the key k into idxbc.
-//
-// Returns true if the idxb has been put into idxbc.
-func (idxbc *indexBlockCache) Put(k uint64, idxb *indexBlock) bool {
+func (idxbc *indexBlockCache) Put(k uint64, idxb *indexBlock) {
 	idxbc.mu.Lock()
 
 	// Remove superflouos entries.
@@ -278,12 +277,11 @@ func (idxbc *indexBlockCache) Put(k uint64, idxb *indexBlock) bool {
 
 	// Store idxb in the cache.
 	idxbe := &indexBlockCacheEntry{
-		lastAccessTime: atomic.LoadUint64(&currentTimestamp),
+		lastAccessTime: fasttime.UnixTimestamp(),
 		idxb:           idxb,
 	}
 	idxbc.m[k] = idxbe
 	idxbc.mu.Unlock()
-	return true
 }
 
 func (idxbc *indexBlockCache) Len() uint64 {
@@ -377,7 +375,7 @@ func (ibc *inmemoryBlockCache) cleaner() {
 }
 
 func (ibc *inmemoryBlockCache) cleanByTimeout() {
-	currentTime := atomic.LoadUint64(&currentTimestamp)
+	currentTime := fasttime.UnixTimestamp()
 	ibc.mu.Lock()
 	for k, ibe := range ibc.m {
 		// Delete items accessed more than 10 minutes ago.
@@ -396,7 +394,7 @@ func (ibc *inmemoryBlockCache) Get(k inmemoryBlockCacheKey) *inmemoryBlock {
 	ibc.mu.RUnlock()
 
 	if ibe != nil {
-		currentTime := atomic.LoadUint64(&currentTimestamp)
+		currentTime := fasttime.UnixTimestamp()
 		if atomic.LoadUint64(&ibe.lastAccessTime) != currentTime {
 			atomic.StoreUint64(&ibe.lastAccessTime, currentTime)
 		}
@@ -407,9 +405,7 @@ func (ibc *inmemoryBlockCache) Get(k inmemoryBlockCacheKey) *inmemoryBlock {
 }
 
 // Put puts ib under key k into ibc.
-//
-// Returns true if ib was put into ibc.
-func (ibc *inmemoryBlockCache) Put(k inmemoryBlockCacheKey, ib *inmemoryBlock) bool {
+func (ibc *inmemoryBlockCache) Put(k inmemoryBlockCacheKey, ib *inmemoryBlock) {
 	ibc.mu.Lock()
 
 	// Clean superflouos entries in cache.
@@ -429,12 +425,11 @@ func (ibc *inmemoryBlockCache) Put(k inmemoryBlockCacheKey, ib *inmemoryBlock) b
 
 	// Store ib in the cache.
 	ibe := &inmemoryBlockCacheEntry{
-		lastAccessTime: atomic.LoadUint64(&currentTimestamp),
+		lastAccessTime: fasttime.UnixTimestamp(),
 		ib:             ib,
 	}
 	ibc.m[k] = ibe
 	ibc.mu.Unlock()
-	return true
 }
 
 func (ibc *inmemoryBlockCache) Len() uint64 {
@@ -451,16 +446,3 @@ func (ibc *inmemoryBlockCache) Requests() uint64 {
 func (ibc *inmemoryBlockCache) Misses() uint64 {
 	return atomic.LoadUint64(&ibc.misses)
 }
-
-func init() {
-	go func() {
-		ticker := time.NewTicker(time.Second)
-		defer ticker.Stop()
-		for tm := range ticker.C {
-			t := uint64(tm.Unix())
-			atomic.StoreUint64(&currentTimestamp, t)
-		}
-	}()
-}
-
-var currentTimestamp uint64

lib/mergeset/part_search.go

@@ -25,9 +25,6 @@ type partSearch struct {
 	// The remaining block headers to scan in the current metaindexRow.
 	bhs []blockHeader
 
-	// Pointer to index block, which may be reused.
-	indexBlockReuse *indexBlock
-
 	// Pointer to inmemory block, which may be reused.
 	inmemoryBlockReuse *inmemoryBlock
 
@@ -53,10 +50,6 @@ func (ps *partSearch) reset() {
 	ps.p = nil
 	ps.mrs = nil
 	ps.bhs = nil
-	if ps.indexBlockReuse != nil {
-		putIndexBlock(ps.indexBlockReuse)
-		ps.indexBlockReuse = nil
-	}
 	if ps.inmemoryBlockReuse != nil {
 		putInmemoryBlock(ps.inmemoryBlockReuse)
 		ps.inmemoryBlockReuse = nil
@@ -275,40 +268,25 @@ func (ps *partSearch) nextBlock() error {
 }
 
 func (ps *partSearch) nextBHS() error {
-	if ps.indexBlockReuse != nil {
-		putIndexBlock(ps.indexBlockReuse)
-		ps.indexBlockReuse = nil
-	}
 	if len(ps.mrs) == 0 {
 		return io.EOF
 	}
 	mr := &ps.mrs[0]
 	ps.mrs = ps.mrs[1:]
-	idxb, mayReuseIndexBlock, err := ps.getIndexBlock(mr)
-	if err != nil {
-		return fmt.Errorf("cannot get index block: %s", err)
-	}
-	if mayReuseIndexBlock {
-		ps.indexBlockReuse = idxb
+	idxbKey := mr.indexBlockOffset
+	idxb := ps.idxbCache.Get(idxbKey)
+	if idxb == nil {
+		var err error
+		idxb, err = ps.readIndexBlock(mr)
+		if err != nil {
+			return fmt.Errorf("cannot read index block: %s", err)
+		}
+		ps.idxbCache.Put(idxbKey, idxb)
 	}
 	ps.bhs = idxb.bhs
 	return nil
 }
 
-func (ps *partSearch) getIndexBlock(mr *metaindexRow) (*indexBlock, bool, error) {
-	idxbKey := mr.indexBlockOffset
-	idxb := ps.idxbCache.Get(idxbKey)
-	if idxb != nil {
-		return idxb, false, nil
-	}
-	idxb, err := ps.readIndexBlock(mr)
-	if err != nil {
-		return nil, false, err
-	}
-	ok := ps.idxbCache.Put(idxbKey, idxb)
-	return idxb, !ok, nil
-}
-
 func (ps *partSearch) readIndexBlock(mr *metaindexRow) (*indexBlock, error) {
 	ps.compressedIndexBuf = bytesutil.Resize(ps.compressedIndexBuf, int(mr.indexBlockSize))
 	ps.p.indexFile.MustReadAt(ps.compressedIndexBuf, int64(mr.indexBlockOffset))
@@ -347,8 +325,8 @@ func (ps *partSearch) getInmemoryBlock(bh *blockHeader) (*inmemoryBlock, bool, e
 	if err != nil {
 		return nil, false, err
 	}
-	ok := ps.ibCache.Put(ibKey, ib)
-	return ib, !ok, nil
+	ps.ibCache.Put(ibKey, ib)
+	return ib, false, nil
 }
 
 func (ps *partSearch) readInmemoryBlock(bh *blockHeader) (*inmemoryBlock, error) {