VictoriaMetrics/docs/victoriametrics/vmagent.md

weight, menu, title, tags, aliases

weight	menu	title	tags	aliases
3	docs parent weight victoriametrics 3	vmagent	metrics	/vmagent.html /vmagent/index.html /vmagent/

vmagent is a tiny agent that helps you collect metrics from various sources, relabel and filter the collected metrics and store them in VictoriaMetrics or any other storage systems via Prometheus remote_write protocol or via VictoriaMetrics remote_write protocol.

See Quick Start for details.

{width="700"}

Motivation

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 into VictoriaMetrics. Also, we found that our user's infrastructure are like snowflakes in that no two are alike. Therefore, we decided to add more flexibility to vmagent such as the ability to accept metrics via popular push protocols and to discover Prometheus-compatible targets and scrape metrics from them.

Features

• Can be used as a drop-in replacement for Prometheus for discovering and scraping targets such as node_exporter. Note that single-node VictoriaMetrics can also discover and scrape Prometheus-compatible targets in the same way vmagent does - see these docs.
• Can add, remove and modify labels (aka tags) via Prometheus relabeling and filter data before sending it to remote storage. See these docs for details.
• Can accept data via all the ingestion protocols supported by VictoriaMetrics - see these docs.
• Can aggregate incoming samples by time and by labels before sending them to remote storage - see these docs.
• Can replicate collected metrics simultaneously to multiple Prometheus-compatible remote storage systems - see these docs.
• Can save egress network bandwidth usage costs when VictoriaMetrics remote write protocol is used for sending the data to VictoriaMetrics.
• Works smoothly 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 the connection to the remote storage is repaired. The maximum disk usage for the buffer can be limited with -remoteWrite.maxDiskUsagePerURL.
• Uses much lower amounts of RAM, CPU, disk IO and network bandwidth than Prometheus. The RAM usage and CPU usage can be reduced further if needed according to these docs.
• Scrape targets can be spread among multiple vmagent instances when large number of targets must be scraped. See these docs.
• Can load scrape configs from multiple files. See these docs.
• Can efficiently scrape targets that expose millions of time series such as /federate endpoint in Prometheus. See these docs.
• Can deal with high cardinality and high churn rate issues by limiting the number of unique time series at scrape time and before sending them to remote storage systems. See these docs.
• Can write collected metrics to multiple tenants. See these docs.
• Can read and write data from / to Kafka. See these docs.
• Can read and write data from / to Google PubSub. See these docs.

Quick Start

Please download vmutils-* archive from releases page ( vmagent is also available in docker images Docker Hub and Quay), unpack it and pass the following flags to the vmagent binary in order to start scraping Prometheus-compatible targets and sending the data to the Prometheus-compatible remote storage:

• -promscrape.config with the path to Prometheus config file (usually located at /etc/prometheus/prometheus.yml). The path can point either to local file or to http url. See scrape config examples. vmagent doesn't support some sections of Prometheus config file, so you may need either to delete these sections or to run vmagent with -promscrape.config.strictParse=false command-line flag. In this case vmagent ignores unsupported sections. See the list of unsupported sections.
• -remoteWrite.url of a Prometheus-compatible remote storage endpoint (e.g., VictoriaMetrics) to send data to. The -remoteWrite.url may refer to a DNS SRV address. See these docs for details.

Example command for writing the data received via supported push-based protocols to single-node VictoriaMetrics located at victoria-metrics-host:8428:

/path/to/vmagent -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write

See these docs if you need to write data to VictoriaMetrics cluster.

Example command for scraping Prometheus targets and writing the data to single-node VictoriaMetrics:

/path/to/vmagent -promscrape.config=/path/to/prometheus.yml -remoteWrite.url=https://victoria-metrics-host:8428/api/v1/write

See how to scrape Prometheus-compatible targets for more details.

If you use single-node VictoriaMetrics, then you can discover and scrape Prometheus-compatible targets directly from VictoriaMetrics without the need to use vmagent - see these docs.

vmagent can save network bandwidth usage costs under high load when VictoriaMetrics remote write protocol is used.

Common vmagent issues are covered in the troubleshooting docs.

See various use cases for vmagent.

Pass -help to vmagent in order to see the full list of supported command-line flags with their descriptions.

Use Cases

IoT and Edge monitoring

vmagent can run and collect metrics in IoT environments and industrial networks with unreliable or scheduled connections to remote storage. It buffers the collected data in local files until the connection to remote storage becomes available and then sends the buffered data to the remote storage. It re-tries sending the data to remote storage until errors are resolved. The maximum on-disk size for the buffered metrics can be limited with -remoteWrite.maxDiskUsagePerURL.

vmagent works on several common architectures used in IoT environments - 32-bit arm, 64-bit arm, ppc64, 386, amd64.

vmagent can save on network bandwidth usage costs by using VictoriaMetrics remote write protocol.

See how to optimize index size at VictoriaMetrics for IoT and industrial monitoring.

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 Prometheus. Typically, vmagent requires lower amounts of RAM, CPU and network bandwidth compared with Prometheus. See these docs for details.

Statsd alternative

vmagent can be used as an alternative to statsd when stream aggregation is enabled. See these docs for details.

Flexible metrics relay

vmagent can accept metrics in various popular data ingestion protocols, apply relabeling to the accepted metrics (for example, change metric names/labels or drop unneeded metrics) and then forward the relabeled metrics to other remote storage systems, that support Prometheus remote_write protocol (including other vmagent instances).

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 is temporarily out of service, then the collected data remains available in the other remote storage instances. vmagent buffers the collected data in files at -remoteWrite.tmpDataPath until the remote storage becomes available again, and then it sends the buffered data to the remote storage in order to prevent data gaps.

VictoriaMetrics cluster already supports replication, so there is no need to specify multiple -remoteWrite.url flags when writing data to the same cluster. See these docs.

Sharding among remote storages

By default vmagent replicates data to remote storage systems via the -remoteWrite.url command-line flag. If the -remoteWrite.shardByURL command-line flag is set, then vmagent spreads the outgoing time series evenly among all the remote storage systems listed in -remoteWrite.url.

It is possible to replicate samples among remote storage systems by passing -remoteWrite.shardByURLReplicas=N command-line flag to vmagent additionally to -remoteWrite.shardByURL command-line flag. This instructs vmagent to write every outgoing sample to N number of distinct remote storage systems listed in -remoteWrite.url in addition to sharding.

Samples for the same time series are routed to the same remote storage system if -remoteWrite.shardByURL flag is specified. This allows building scalable data processing pipelines when a single remote storage cannot keep up with the data ingestion workload. For example, this allows horizontal scaling with stream aggregation by routing outgoing samples for the same time series like counter and histogram types from top-level vmagent instances to the same second-level vmagent instance, so they are aggregated properly.

If -remoteWrite.shardByURL command-line flag is set, then all the metric labels are used for even sharding among remote storage systems specified in -remoteWrite.url.

The -remoteWrite.shardByURL may not work as expected when SRV URLs are in use.

An SRV record might resolve to multiple addresses, one address is chosen randomly for all subsequent logic, including sharding. It will make sharding inconsistent. Samples of the same time series always go to the same remote write URL/SRV record, but they may reach different addresses randomly based on the DNS resolution.

For example, if you set -remoteWrite.url=srv+foo and it's resolved to three addresses (192.168.1.1, 192.168.1.2, 192.168.1.3), vmagent will only choose one randomly every time it (re-)creates the connection. In contrast, specifying the addresses manually (-remoteWrite.url=192.168.1.1 -remoteWrite.url=192.168.1.2 -remoteWrite.url=192.168.1.3) will shard samples across all three URLs.

Sometimes it may be needed to use only a particular set of labels for sharding. For example, it may be necessary to route all the metrics with the same instance label to the same -remoteWrite.url. In this case you can specify comma-separated list of these labels in the -remoteWrite.shardByURL.labels command-line flag. For example, -remoteWrite.shardByURL.labels=instance,__name__ would shard metrics with the same name and instance label to the same -remoteWrite.url.

Sometimes, it may be necessary to ignore some labels when sharding samples across multiple -remoteWrite.url backends. For example, if all the raw samples with the same set of labels except for the labels instance and pod must be routed to the same backend. In this case the list of ignored labels must be passed to -remoteWrite.shardByURL.ignoreLabels command-line flag: -remoteWrite.shardByURL.ignoreLabels=instance,pod.

See also how to scrape large number of targets.

Relabeling and filtering

vmagent can add, remove or update labels on the collected data before sending it to the remote storage. Additionally, it can remove unwanted samples via Prometheus-like relabeling before sending the collected data to remote storage. Please see Relabeling cookbook for details.

Splitting data streams among multiple systems

vmagent supports splitting the collected data between multiple destinations with the help of -remoteWrite.urlRelabelConfig, it is applied independently for each configured -remoteWrite.url destination. For example, it is possible to replicate or split data among long-term remote storage, short-term remote storage and a real-time analytical system built on top of Kafka. Note that each destination can receive its own subset of the collected data due to per-destination relabeling via -remoteWrite.urlRelabelConfig.

For example, let's assume that all the metrics scraped or received by vmagent have the label env with the value of dev or prod. To route metrics with the label env=dev to dev and metrics with the label env=prod to prod apply the following config:

1. Create a relabeling config file relabelDev.yml to drop all metrics that don't have label env=dev:

- action: keep
  source_labels: [env]
  regex: "dev"

2. Create a relabeling config file relabelProd.yml to drop all metrics that don't have label env=prod:

- action: keep
  source_labels: [env]
  regex: "prod"

3. Configure vmagent with 2 -remoteWrite.url flags pointing to destinations dev and prod with corresponding -remoteWrite.urlRelabelConfig config files:

./vmagent \
  -remoteWrite.url=http://<dev-url> -remoteWrite.urlRelabelConfig=relabelDev.yml \
  -remoteWrite.url=http://<prod-url> -remoteWrite.urlRelabelConfig=relabelProd.yml

With this configuration vmagent will only forward metrics with the label env=dev to http://<dev-url> and metrics with the label env=prod to http://<prod-url>.

Please note, the order of flags is important: the first -remoteWrite.urlRelabelConfig will be applied to the first -remoteWrite.url, and so on.

Prometheus remote_write proxy

vmagent can be used as a proxy for Prometheus data sent via Prometheus remote_write protocol. It can accept data via the remote_write API at the/api/v1/write endpoint. It then applies relabeling and filtering then proxies it to another remote_write system. vmagent can also be configured to encrypt the incoming remote_write requests with -tls* command-line flags. Basic Auth can be enabled for the incoming remote_write requests with -httpAuth.* command-line flags.

remote_write for clustered version

While vmagent can accept data in several supported protocols (OpenTSDB, Influx, Prometheus, Graphite) and scrape data from various targets, writes are always performed in Prometheus remote_write protocol. Therefore, for the clustered version, the -remoteWrite.url command-line flag should be configured as <schema>://<vminsert-host>:8480/insert/<accountID>/prometheus/api/v1/write according to these docs. There is also support for multitenant writes. See these docs.

Flexible deduplication

Deduplication at stream aggregation allows setting up arbitrary complex de-duplication schemes for the collected samples. Examples:

• The following command instructs vmagent to send only the last sample for each time series every 60 seconds:

  ./vmagent -remoteWrite.url=http://remote-storage/api/v1/write -streamAggr.dedupInterval=60s
  

• The following command instructs vmagent to merge time series with different replica label values and then to send only the last sample for each merged series every 60 seconds:

  ./vmagent -remoteWrite.url=http://remote-storage/api/v1/write -streamAggr.dropInputLabels=replica -streamAggr.dedupInterval=60s
  

Life of a sample

vmagent supports limiting, relabeling, deduplication and stream aggregation for all metric samples, scraped or pushed. The received data is then forwarded to the specified -remoteWrite.url destinations. The pipeline is as follows:

%%{init: { "themeCSS": ".nodeLabel, .edgeLabel { white-space: nowrap; word-break: normal; overflow-wrap: normal; }" }}%%
flowchart TB
    A[Pushed or Scraped samples] --> B[Ingestion rate limiting<br><b>-maxIngestionRate</b>]
    B --> C[Global <a href="https://docs.victoriametrics.com/victoriametrics/relabeling/">relabeling</a><br><b>-remoteWrite.relabelConfig</b>]
    C --> D[complexity limiting<br><b>-maxLabelsPerTimeseries</b><br><b>-maxLabelNameLen</b><br><b>-maxLabelValueLen</b>]
    D --> E[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#cardinality-limiter">cardinality limiting</a><br><b>-remoteWrite.maxHourlySeries</b><br><b>-remoteWrite.maxDailySeries</b>]
    E --> F[Global <a href="https://docs.victoriametrics.com/victoriametrics/stream-aggregation">aggregation</a><br><b>-streamAggr.config</b><br><b>-streamAggr.dedupInterval</b>]
    F --> G[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#replication-and-high-availability">replicate</a> to each <b>-remoteWrite.url</b><br/>or <a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#sharding-among-remote-storages">shard</a> if <b>-remoteWrite.shardByURL</b> is set]

      %% Left branch
      G --> H1[per-url <a href="https://docs.victoriametrics.com/victoriametrics/relabeling/">relabeling</a><br><b>-remoteWrite.urlRelabelConfig</b>]
      H1 --> H2[per-url <a href="https://docs.victoriametrics.com/victoriametrics/stream-aggregation">aggregation</a><br><b>-remoteWrite.streamAggr.config</b><br><b>-remoteWrite.streamAggr.dedupInterval</b>]
      H2 --> H3["per-url <a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#calculating-disk-space-for-persistence-queue">queue</a> (default: enabled)<br><b>-remoteWrite.disableOnDiskQueue</b>"]
      H3 --> H4[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#adding-labels-to-metrics">add extra labels</a><br><b>-remoteWrite.label</b>]
      H4 --> H5[[push to <b>-remoteWrite.url</b>]]

      %% Right branch
      G --> R1[per-url <a href="https://docs.victoriametrics.com/victoriametrics/relabeling/">relabeling</a><br><b>-remoteWrite.urlRelabelConfig</b>]
      R1 --> R2[per-url <a href="https://docs.victoriametrics.com/victoriametrics/stream-aggregation">aggregation</a><br><b>-remoteWrite.streamAggr.config</b><br><b>-remoteWrite.streamAggr.dedupInterval</b>]
      R2 --> R3["per-url <a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#calculating-disk-space-for-persistence-queue">queue</a> (default: enabled)<br><b>-remoteWrite.disableOnDiskQueue</b>"]
      R3 --> R4[<a href="https://docs.victoriametrics.com/victoriametrics/vmagent/#adding-labels-to-metrics">add extra labels</a><br><b>-remoteWrite.label</b>]
      R4 --> R5[[push to <b>-remoteWrite.url</b>]]

Scraping has additional settings that can be applied before samples are pushed to the processing pipeline above:

1. Service Discovery relabeling
2. Scraping relabeling
3. sample_limit, series_limit, label_limit in scrape_configs.

How to push data to vmagent

vmagent supports the same set of push-based data ingestion protocols as VictoriaMetrics does in addition to the pull-based Prometheus-compatible targets' scraping:

• DataDog "submit metrics" API. See these docs.
• InfluxDB line protocol via http://<vmagent>:8429/write. See these docs.
• Graphite plaintext protocol if -graphiteListenAddr command-line flag is set. See these docs.
• OpenTelemetry http API. See these docs.
• NewRelic API. See these docs.
• OpenTSDB telnet and http protocols if -opentsdbListenAddr command-line flag is set. See these docs.
• Zabbix Connector streaming protocol. See these docs.
• Prometheus remote write protocol via http://<vmagent>:8429/api/v1/write.
• JSON lines import protocol via http://<vmagent>:8429/api/v1/import. See these docs.
• Native data import protocol via http://<vmagent>:8429/api/v1/import/native. See these docs.
• Prometheus exposition format via http://<vmagent>:8429/api/v1/import/prometheus. See these docs for details.
• Arbitrary CSV data via http://<vmagent>:8429/api/v1/import/csv. See these docs.

How to collect metrics in Prometheus format

Specify the path to prometheus.yml file via -promscrape.config command-line flag. vmagent takes into account the following sections from Prometheus config file:

• global
• scrape_configs

All other sections are ignored, including the remote_write section. Use -remoteWrite.* command-line flag instead for configuring remote write settings. See the list of unsupported config sections.

The file pointed to by -promscrape.config may contain %{ENV_VAR} placeholders that are substituted by the corresponding ENV_VAR environment variable values.

See also:

• scrape config examples
• the list of supported service discovery types for Prometheus scrape targets.

scrape_config enhancements

vmagent supports the following additional options in scrape_configs section:

• headers - a list of HTTP headers to send to a scrape target with each scrape request. This can be used when the scrape target needs custom authorization and authentication. For example:

scrape_configs:
- job_name: custom_headers
  headers:
  - "TenantID: abc"
  - "My-Auth: TopSecret"

• disable_compression: true for disabling response compression on a per-job basis. By default, vmagent requests compressed responses from scrape targets for saving network bandwidth.
• disable_keepalive: true for disabling HTTP keep-alive connections on a per-job basis. By default, vmagent uses keep-alive connections to scrape targets for reducing overhead on connection re-establishing.
• series_limit: N for limiting the number of unique time series a single scrape target can expose. See these docs.
• stream_parse: true for scraping targets in a streaming manner. This may be useful when targets export large number of metrics. See these docs.
• scrape_align_interval: duration for aligning scrapes to the given interval instead of using random offset in the range [0 ... scrape_interval] for scraping each target. The random offset helps to spread scrapes evenly in time.
• scrape_offset: duration for specifying the exact offset for scraping instead of using random offset in the range [0 ... scrape_interval].

See scrape_configs docs for more details on all the supported options.

Loading scrape configs from multiple files

vmagent supports loading scrape configs from multiple files specified in the scrape_config_files section of -promscrape.config file. For example, the following -promscrape.config instructs vmagent to load scrape configs from all the *.yml files under the configs directory, from the local file single_scrape_config.yml and from the URL https://config-server/scrape_config.yml:

scrape_config_files:
- configs/*.yml
- single_scrape_config.yml
- https://config-server/scrape_config.yml

Every referred file can contain arbitrary number of supported scrape configs. There is no need to specify a top-level scrape_configs section in these files. For example:

- job_name: foo
  static_configs:
  - targets: ["vmagent:8429"]
- job_name: bar
  kubernetes_sd_configs:
  - role: pod

vmagent is able to dynamically reload these files - see these docs.

Unsupported Prometheus config sections

vmagent doesn't support the following sections in Prometheus config file passed to -promscrape.config command-line flag:

• remote_write. This section is substituted with various -remoteWrite* command-line flags. See the full list of flags. The remote_write section isn't supported in order to reduce possible confusion when vmagent is used for accepting incoming metrics via supported push protocols. In this case the -promscrape.config file isn't needed.
• remote_read. This section isn't supported at all, since vmagent doesn't provide Prometheus querying API. It is expected that the querying API is provided by the remote storage specified via -remoteWrite.url such as VictoriaMetrics. See Prometheus querying API docs for VictoriaMetrics.
• rule_files and alerting. These sections are supported by vmalert.

The list of supported service discovery types is available in how-to-collect-metrics-in-prometheus-format.

Additionally, vmagent doesn't support refresh_interval option at service discovery sections. This option is substituted with -promscrape.*CheckInterval command-line flags, that are specific per each service discovery type. See the full list of command-line flags for vmagent.

Configuration update

vmagent should be restarted in order to update config options set via command-line args. vmagent supports multiple approaches for reloading configs from updated config files such as -promscrape.config, -remoteWrite.relabelConfig, -remoteWrite.urlRelabelConfig, -streamAggr.config and -remoteWrite.streamAggr.config:

• Sending SIGHUP signal to vmagent process:

  kill -SIGHUP `pidof vmagent`
  

• Sending HTTP request to http://vmagent:8429/-/reload endpoint. This endpoint can be protected with -reloadAuthKey command-line flag.

There is also -promscrape.configCheckInterval command-line flag, that can be used for automatic reloading configs from updated -promscrape.config file.

SRV URLs

If vmagent encounters URLs with srv+ prefix in hostname (such as http://srv+some-addr/some/path), then it resolves some-addr DNS SRV record into TCP address with hostname and TCP port, and then uses the resulting url when it needs connecting to it.

SRV URLs are supported in the following places:

• In -remoteWrite.url command-line flag. For example, if victoria-metrics DNS SRV record contains victoria-metrics-host:8428 TCP address, then -remoteWrite.url=http://srv+victoria-metrics/api/v1/write is automatically resolved into -remoteWrite.url=http://victoria-metrics-host:8428/api/v1/write. If the DNS SRV record is resolved into multiple TCP addresses, then vmagent uses randomly chosen address per each connection it establishes to the remote storage.

• In scrape target addresses aka __address__ label - see these docs for details.

• In urls used for service discovery.

SRV urls are useful when HTTP services run on different TCP ports or when they can change TCP ports over time (for instance, after the restart).

VictoriaMetrics remote write protocol

vmagent supports sending data to the configured -remoteWrite.url either via Prometheus remote write protocol or via VictoriaMetrics remote write protocol.

VictoriaMetrics remote write protocol provides the following benefits comparing to Prometheus remote write protocol:

• Reduced network bandwidth usage by 2x-5x. This allows saving network bandwidth usage costs when vmagent and the configured remote storage systems are located in different datacenters, availability zones or regions.

• Reduced disk read/write IO and disk space usage at vmagent when the remote storage is temporarily unavailable. In this case vmagent buffers the incoming data to disk using the VictoriaMetrics remote write format. This reduces disk read/write IO and disk space usage by 2x-5x comparing to Prometheus remote write format.

vmagent uses VictoriaMetrics remote write protocol by default {{% available_from "v1.116.0" %}} when it sends data to VictoriaMetrics components such as other vmagent instances, single-node VictoriaMetrics or vminsert at cluster version. If needed, It can automatically downgrade to a Prometheus protocol at runtime. It is possible to force switch to VictoriaMetrics remote write protocol by specifying -remoteWrite.forceVMProto command-line flag for the corresponding -remoteWrite.url. It is possible to tune the compression level for VictoriaMetrics remote write protocol with -remoteWrite.vmProtoCompressLevel command-line flag. Bigger values reduce network usage at the cost of higher CPU usage. Negative values reduce CPU usage at the cost of higher network usage. The default value for the compression level is 0, the minimum value is -22 and the maximum value is 22. The default value works optimally in most cases, so it isn't recommended changing it.

vmagent automatically switches to Prometheus remote write protocol when it sends data to old versions of VictoriaMetrics components or to other Prometheus-compatible remote storage systems. It is possible to force switch to Prometheus remote write protocol by specifying -remoteWrite.forcePromProto command-line flag for the corresponding -remoteWrite.url.

Multitenancy

By default vmagent collects the data without tenant identifiers and routes it to the remote storage specified via -remoteWrite.url command-line flag. The -remoteWrite.url can point to /insert/<tenant_id>/prometheus/api/v1/write path at vminsert according to these docs. In this case all the metrics are written to the given <tenant_id> tenant.

The easiest way to write data to multiple distinct tenants is to specify the needed tenants via vm_account_id and vm_project_id labels and then to push metrics with these labels to multitenant url at VictoriaMetrics cluster. The vm_account_id and vm_project_id labels can be specified via relabeling before sending the metrics to -remoteWrite.url.

For example, the following relabeling rule instructs sending metrics to <account_id>:0 tenant defined in the prometheus.io/account_id annotation of Kubernetes pod deployment:

scrape_configs:
- kubernetes_sd_configs:
  - role: pod
  relabel_configs:
  - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_account_id]
    target_label: vm_account_id

vmagent can accept data via the same multitenant endpoints (/insert/<accountID>/<suffix>) as vminsert at VictoriaMetrics cluster does according to these docs if -enableMultitenantHandlers command-line flag is set. In this case, vmagent automatically converts tenant identifiers from the URL to vm_account_id and vm_project_id labels. These tenant labels are added before applying relabeling specified via -remoteWrite.relabelConfig and -remoteWrite.urlRelabelConfig command-line flags. Metrics with vm_account_id and vm_project_id labels can be routed to the corresponding tenants when specifying -remoteWrite.url to multitenant url at VictoriaMetrics cluster.

Adding labels to metrics

Extra labels can be added to metrics collected by vmagent via the following mechanisms:

• The global -> external_labels section in -promscrape.config file. These labels are added only to metrics scraped from targets configured in the -promscrape.config file. They aren't added to metrics collected via other data ingestion protocols.

• The -remoteWrite.label command-line flag. These labels are added to all the collected metrics before sending them to all configured -remoteWrite.url. For example, the following command starts vmagent, which adds {datacenter="foobar"} label to all the metrics pushed to all the configured -remoteWrite.url destinations:

  /path/to/vmagent -remoteWrite.label=datacenter=foobar ...
  

• Via relabeling. Relabeling can be applied globally and per each configured -remoteWrite.url destination. See Relabeling Cookbook.

• Add extra_label GET param to -remoteWrite.url address (only works when sending data to VictoriaMetrics components):

  /path/to/vmagent -remoteWrite.url=http://127.0.0.1:8428/api/v1/write?extra_label="env=prod"
  

Automatically generated metrics

vmagent automatically generates the following metrics per each scrape of every Prometheus-compatible target and attaches instance, job and other target-specific labels to these metrics:

• up - this metric exposes 1 value on successful scrape and 0 value on unsuccessful scrape. This allows monitoring failing scrapes with the following MetricsQL query:

  up == 0
  

• scrape_duration_seconds - the duration of the scrape for the given target. This allows monitoring slow scrapes. For example, the following MetricsQL query returns scrapes, which take more than 1.5 seconds to complete:

  scrape_duration_seconds > 1.5
  

• scrape_timeout_seconds - the configured timeout for the current scrape target (aka scrape_timeout). This allows detecting targets with scrape durations close to the configured scrape timeout. For example, the following MetricsQL query returns targets (identified by instance label), which take more than 80% of the configured scrape_timeout during scrapes:

  scrape_duration_seconds / scrape_timeout_seconds > 0.8
  

• scrape_response_size_bytes - response size in bytes for the given target. This allows to monitor amount of data scraped and to adjust max_scrape_size option for scraped targets. For example, the following MetricsQL query returns targets with scrape response bigger than 10MiB:

  scrape_response_size_bytes > 10MiB
  

• scrape_samples_scraped - the number of samples parsed per each scrape. This allows detecting targets, which expose too many series. For example, the following MetricsQL query returns targets, which expose more than 10000 metrics:

  scrape_samples_scraped > 10000
  

• scrape_samples_limit - the configured limit on the number of samples the given target can expose. The limit can be set via sample_limit option at scrape_configs. This metric is exposed only if the sample_limit is set. This allows detecting targets, which expose too many metrics compared to the configured sample_limit. For example, the following query returns targets (identified by instance label), which expose more than 80% metrics compared to the configured sample_limit:

  scrape_samples_scraped / scrape_samples_limit > 0.8
  

• scrape_samples_post_metric_relabeling - the number of samples left after applying metric-level relabeling from metric_relabel_configs section (see relabeling docs for more details). This allows detecting targets with too many series after the relabeling. For example, the following MetricsQL query returns targets with more than 10000 metrics after the relabeling:

  scrape_samples_post_metric_relabeling > 10000
  

• scrape_labels_limit - the configured limit on the number of labels the given target can expose per sample. The limit can be set via label_limit option at scrape_configs. This metric is exposed only if the label_limit is set.

• scrape_series_added - an approximate number of new series the given target generates during the current scrape. This metric allows detecting targets (identified by instance label), which lead to high churn rate. For example, the following MetricsQL query returns targets, which generate more than 1000 new series during the last hour:

  sum_over_time(scrape_series_added[1h]) > 1000
  

  vmagent sets scrape_series_added to zero when it runs with -promscrape.noStaleMarkers command-line flag or when it scrapes target with no_stale_markers: true option, e.g. when staleness markers are disabled.

• scrape_series_limit - the limit on the number of unique series the given target can expose according to these docs. This metric is exposed only if the series limit is set.

• scrape_series_current - the number of unique series the given target exposed so far. This metric is exposed only if the series limit is set according to these docs. This metric allows alerting when the number of exposed series by the given target reaches the limit. For example, the following query would alert when the target exposes more than 90% of unique series compared to the configured limit.

  scrape_series_current / scrape_series_limit > 0.9
  

• scrape_series_limit_samples_dropped - exposes the number of dropped samples during the scrape because of the exceeded limit on the number of unique series. This metric is exposed only if the series limit is set according to these docs. This metric allows alerting when scraped samples are dropped because of the exceeded limit. For example, the following query alerts when at least a single sample is dropped because of the exceeded limit during the last hour:

  sum_over_time(scrape_series_limit_samples_dropped[1h]) > 0
  

If the target exports metrics with names clashing with the automatically generated metric names, then vmagent automatically adds exported_ prefix to these metric names, so they don't clash with automatically generated metric names.

Relabeling defined in relabel_configs or metric_relabel_configs of scrape config isn't applied to automatically generated metrics. But they still can be relabeled via -remoteWrite.relabelConfig before sending metrics to remote address.

Prometheus staleness markers

vmagent sends Prometheus staleness markers to -remoteWrite.url in the following cases:

• If they are passed to vmagent via Prometheus remote_write protocol.
• If the metric disappears from the list of scraped metrics, then stale marker is sent to this particular metric.
• If the scrape target becomes temporarily unavailable, then stale markers are sent for all the metrics scraped from this target.
• If the scrape target is removed from the list of targets, then stale markers are sent for all the metrics scraped from this target.

Prometheus staleness markers' tracking needs additional memory, since it must store the previous response body per each scrape target in order to compare it to the current response body. The memory usage may be reduced by disabling staleness tracking in the following ways:

• By passing -promscrape.noStaleMarkers command-line flag to vmagent. This disables staleness tracking across all the targets.
• By specifying no_stale_markers: true option in the scrape_config for the corresponding target.

When staleness tracking is disabled, then vmagent doesn't track the number of new time series per each scrape, e.g. it sets scrape_series_added metric to zero. See these docs for details.

Metric metadata

vmagent accepts{{% available_from "#" %}} metric metadata exposed by scrape targets in Prometheus exposition format, received via Prometheus remote write v1 or OpenTelemetry protocol by default. Set -enableMetadata=false to disable metadata processing{{% available_from "v1.125.1" %}}. During processing, metadata won't be dropped or modified by relabeling or streaming aggregation.

When -enableMultitenantHandlers is enabled, vmagent adds tenant info to metadata received via the multitenant endpoints (/insert/<accountID>/<suffix>). However, if vm_account_id or vm_project_id labels are added directly to metrics before reaching vmagent, and vmagent writes to the vminsert multitenant endpoints, the tenant info won't be attached and the metadata will be stored under the default tenant of VictoriaMetrics cluster.

Metadata requires extra memory, disk space, and network traffic.

Stream parsing mode

By default, vmagent parses the full response from the scrape target, applies relabeling and then pushes the resulting metrics to the configured -remoteWrite.url in one go. This mode works great for the majority of cases when the scrape target exposes small number of metrics (e.g. less than 10K). But this mode may take large amounts of memory when the scrape target exposes a large number of metrics (for example, when vmagent scrapes kube-state-metrics in large Kubernetes cluster). It is recommended to enable stream parsing for such targets. When stream parsing is enabled, vmagent processes the response from the scrape target in chunks. This saves memory when scraping targets that expose millions of metrics.

Stream parsing is automatically enabled for scrape targets returning response bodies with sizes larger than the -promscrape.minResponseSizeForStreamParse command-line flag value. Additionally, stream parsing can be explicitly enabled in the following places:

• Via -promscrape.streamParse command-line flag. In this case all the scrape targets defined in the file pointed by -promscrape.config are scraped in stream parsing mode.
• Via stream_parse: true option at scrape_configs section. In this case all the scrape targets defined in this section are scraped in stream parsing mode.
• Via __stream_parse__=true label, which can be set via relabeling at relabel_configs section. In this case stream parsing mode is enabled for the corresponding scrape targets. Typical use case: to set the label via Kubernetes annotations for targets exposing large number of metrics.

Examples:

scrape_configs:
- job_name: 'big-federate'
  stream_parse: true
  static_configs:
  - targets:
    - big-prometheus1
    - big-prometheus2
  honor_labels: true
  metrics_path: /federate
  params:
    'match[]': ['{__name__!=""}']

Note that vmagent in stream parsing mode stores up to sample_limit samples to the configured -remoteStorage.url instead of dropping all the samples read from the target, because the parsed data is sent to the remote storage as soon as it is parsed in stream parsing mode.

Scraping big number of targets

A single vmagent instance can scrape tens of thousands of scrape targets. Sometimes this isn't enough due to limitations on CPU, network, RAM, etc. In this case scrape targets can be split among multiple vmagent instances (aka vmagent horizontal scaling, sharding and clustering). The number of vmagent instances in the cluster must be passed to -promscrape.cluster.membersCount command-line flag. Each vmagent instance in the cluster must use identical -promscrape.config files with distinct -promscrape.cluster.memberNum values in the range 0 ... N-1, where N is the number of vmagent instances in the cluster specified via -promscrape.cluster.membersCount. For example, the following commands spread scrape targets among a cluster of two vmagent instances:

/path/to/vmagent -promscrape.cluster.membersCount=2 -promscrape.cluster.memberNum=0 -promscrape.config=/path/to/config.yml ...
/path/to/vmagent -promscrape.cluster.membersCount=2 -promscrape.cluster.memberNum=1 -promscrape.config=/path/to/config.yml ...

The -promscrape.cluster.memberNum can be set to a StatefulSet pod name when vmagent runs in Kubernetes. The pod name must end with a number in the range 0 ... promscrape.cluster.membersCount-1. For example, -promscrape.cluster.memberNum=vmagent-0.

By default, each scrape target is scraped only by a single vmagent instance in the cluster. If there is a need for replicating scrape targets among multiple vmagent instances, then -promscrape.cluster.replicationFactor command-line flag must be set to the desired number of replicas. For example, the following commands start a cluster of three vmagent instances, where each target is scraped by two vmagent instances:

/path/to/vmagent -promscrape.cluster.membersCount=3 -promscrape.cluster.replicationFactor=2 -promscrape.cluster.memberNum=0 -promscrape.config=/path/to/config.yml ...
/path/to/vmagent -promscrape.cluster.membersCount=3 -promscrape.cluster.replicationFactor=2 -promscrape.cluster.memberNum=1 -promscrape.config=/path/to/config.yml ...
/path/to/vmagent -promscrape.cluster.membersCount=3 -promscrape.cluster.replicationFactor=2 -promscrape.cluster.memberNum=2 -promscrape.config=/path/to/config.yml ...

Every vmagent in the cluster exposes all the discovered targets at http://vmagent:8429/service-discovery page. Each discovered target on this page contains its status (UP, DOWN or DROPPED with the reason why the target has been dropped). If the target is dropped because of sharding to other vmagent instances in the cluster, then the status column contains -promscrape.cluster.memberNum values for vmagent instances where the given target is scraped.

The /service-discovery page provides links to the corresponding vmagent instances if -promscrape.cluster.memberURLTemplate command-line flag is set. Every occurrence of %d inside the -promscrape.cluster.memberURLTemplate is substituted with the -promscrape.cluster.memberNum for the corresponding vmagent instance. For example, -promscrape.cluster.memberURLTemplate='http://vmagent-instance-%d:8429/targets' generates http://vmagent-instance-42:8429/targets url for vmagent instance, which runs with -promscrape.cluster.memberNum=42.

Note that vmagent shows up to -promscrape.maxDroppedTargets dropped targets on the /service-discovery page. Increase the -promscrape.maxDroppedTargets command-line flag value if the /service-discovery page misses some dropped targets.

If each target is scraped by multiple vmagent instances, then data deduplication must be enabled at remote storage pointed by -remoteWrite.url. The -dedup.minScrapeInterval must be set to the scrape_interval configured at -promscrape.config. See these docs for details.

The -promscrape.cluster.memberLabel command-line flag allows specifying a name for member num label to add to all the scraped metrics. The value of the member num label is set to -promscrape.cluster.memberNum. For example, the following config instructs adding vmagent_instance="0" label to all the metrics scraped by the given vmagent instance:

/path/to/vmagent -promscrape.cluster.membersCount=2 -promscrape.cluster.memberNum=0 -promscrape.cluster.memberLabel=vmagent_instance

See also how to shard data among multiple remote storage systems.

High availability

It is possible to run multiple identically configured vmagent instances or vmagent clusters, so they scrape the same set of targets and push the collected data to the same set of VictoriaMetrics remote storage systems. Two identically configured vmagent instances or clusters is usually called an HA pair.

When running HA pairs, deduplication must be configured on the VictoriaMetrics side in order to de-duplicate received samples. See these docs for details.

It is also recommended to pass different values to -promscrape.cluster.name command-line flag per vmagent instance or per vmagent cluster in a HA setup. This is needed for proper data de-duplication. See this issue for details.

Scraping targets via a proxy

vmagent supports scraping targets via http, https and socks5 proxies. Proxy address must be specified in proxy_url option. For example, the following scrape config instructs target scraping via https proxy at https://proxy-addr:1234:

scrape_configs:
- job_name: foo
  proxy_url: https://proxy-addr:1234

Proxy can be configured with the following optional settings:

• proxy_authorization for generic token authorization. See these docs.
• proxy_basic_auth for Basic authorization. See these docs.
• proxy_bearer_token and proxy_bearer_token_file for Bearer token authorization
• proxy_oauth2 for OAuth2 config. See these docs.
• proxy_tls_config for TLS config. See these docs.
• proxy_headers for passing additional HTTP headers in requests to proxy.

For example:

scrape_configs:
- job_name: foo
  proxy_url: https://proxy-addr:1234
  proxy_basic_auth:
    username: foobar
    password: secret
  proxy_tls_config:
    insecure_skip_verify: true
    cert_file: /path/to/cert
    key_file: /path/to/key
    ca_file: /path/to/ca
    server_name: real-server-name
  proxy_headers:
  - "Proxy-Auth: top-secret"

On-disk persistence

vmagent stores pending data that cannot be sent to the configured remote storage systems in a timely manner. By default, vmagent writes all the pending data to folder configured via -remoteWrite.tmpDataPath cmd-line flag until this data is sent to the configured -remoteWrite.url systems or until the folder becomes full. The maximum data size that can be saved to -remoteWrite.tmpDataPath per every configured -remoteWrite.url can be limited via -remoteWrite.maxDiskUsagePerURL command-line flag. When this limit is reached, vmagent drops the oldest data from disk in order to save newly ingested data.

The folder structure of persisted data is as follows:

<remoteWrite.tmpDataPath>
└── persistent-queue
    └── 1_B9EB7BE220B91E9D

Each remote write URL corresponds to a folder similar to 1_B9EB7BE220B91E9D.

It's generated based on the following information:

1. The sequence order of the remote write URL cmd-line flags, starting from 1.
2. The hash result of the remote write URL itself, excluding query parameters and fragments.

For example, for the remote write configs:

-remoteWrite.url=http://example-1:8428/prometheus/api/v1/write?foo=bar#baz
-remoteWrite.url=http://user:pass@example-2:8428/prometheus/api/v1/write?qux=quux#quuz

vmagent will generate the following persistent queue folders:

# 1_<hash(http://example-1:8428/prometheus/api/v1/write)>, query parameters foo=bar and fragment baz are removed.
1_BA6E4303DCFA0D45
# 2_<hash(http://user:pass@example-2:8428/prometheus/api/v1/write)>, query parameters qux=quux and fragment quuz are removed.
2_0AAFDF53E314A72A

Disabling On-disk persistence

There are cases when it is better to disable on-disk persistence for pending data on the vmagent side:

• When the persistent disk performance isn't enough for the given data processing rate.
• When it is better to buffer pending data at the client side instead of buffering it at vmagent side in the -remoteWrite.tmpDataPath folder.
• When the data is already buffered at Kafka or Google PubSub.
• When it is better to drop pending data instead of buffering it.

In this case -remoteWrite.disableOnDiskQueue command-line flag can be passed to vmagent per each configured -remoteWrite.url. vmagent works in the following way if the corresponding remote storage system at -remoteWrite.url cannot keep up with the data ingestion rate and the -remoteWrite.disableOnDiskQueue command-line flag is set:

• It returns 429 Too Many Requests HTTP error to clients, which send data to vmagent via supported HTTP endpoints. If -remoteWrite.dropSamplesOnOverload command-line flag is set or if multiple -remoteWrite.url command-line flags are set, then the ingested samples are silently dropped instead of returning the error to clients.
• It suspends consuming data from Kafka or Google PubSub until the remote storage becomes available. If -remoteWrite.dropSamplesOnOverload command-line flag is set or if multiple -remoteWrite.disableOnDiskQueue command-line flags are set for different -remoteWrite.url options, then the fetched samples are silently dropped instead of suspending data consumption from Kafka or Google PubSub.
• It drops samples pushed to vmagent via non-HTTP protocols and logs the error. Pass -remoteWrite.dropSamplesOnOverload command-line flag in order to suppress error messages in this case.
• It drops samples scraped from Prometheus-compatible targets, because it is better from operations perspective to drop samples instead of blocking the scrape process.
• It drops stream aggregation output samples, because it is better from operations perspective to drop output samples instead of blocking the stream aggregation process.

The number of dropped samples because of overloaded remote storage can be monitored via vmagent_remotewrite_samples_dropped_total metric. The number of unsuccessful attempts to send data to overloaded remote storage can be monitored via vmagent_remotewrite_push_failures_total metric.

Running vmagent on hosts with more RAM or increasing the value for -memory.allowedPercent may reduce the number of unsuccessful attempts or dropped samples on spiky workloads, since vmagent may buffer more data in memory before returning the error or dropping data.

vmagent still may write pending in-memory data to -remoteWrite.tmpDataPath on graceful shutdown if -remoteWrite.disableOnDiskQueue command-line flag is specified. It may also read buffered data from -remoteWrite.tmpDataPath on startup.

When -remoteWrite.disableOnDiskQueue command-line flag is set, vmagent may send the same samples multiple times to the configured remote storage if it cannot keep up with the data ingestion rate. In this case deduplication must be enabled on all the configured remote storage systems.

Cardinality limiter

By default, vmagent doesn't limit the number of time series each scrape target can expose. The limit can be enforced in the following places:

• Via -promscrape.seriesLimitPerTarget command-line flag. This limit is applied individually to all the scrape targets defined in the file pointed by -promscrape.config.
• Via series_limit config option at scrape_config section. The series_limit allows overriding the -promscrape.seriesLimitPerTarget on a per-scrape_config basis. If series_limit is set to 0 or to negative value, then it isn't applied to the given scrape_config, even if -promscrape.seriesLimitPerTarget command-line flag is set.
• Via __series_limit__ label, which can be set with relabeling at relabel_configs section. The __series_limit__ allows overriding the series_limit on a per-target basis. If __series_limit__ is set to 0 or to negative value, then it isn't applied to the given target. Typical use case: to set the limit via Kubernetes annotations for targets, which may expose too high number of time series.

Scraped metrics are dropped for time series exceeding the given limit on the time window of 24h. vmagent creates the following additional per-target metrics for targets with non-zero series limit:

• scrape_series_limit_samples_dropped - the number of dropped samples during the scrape when the unique series limit is exceeded.
• scrape_series_limit - the series limit for the given target.
• scrape_series_current - the current number of series for the given target.

These metrics are automatically sent to the configured -remoteWrite.url alongside with the scraped per-target metrics.

These metrics allow building the following alerting rules:

• scrape_series_current / scrape_series_limit > 0.9 - alerts when the number of series exposed by the target reaches 90% of the limit.
• sum_over_time(scrape_series_limit_samples_dropped[1h]) > 0 - alerts when some samples are dropped because the series limit on a particular target is reached.

See also sample_limit option at scrape_config section.

By default, vmagent doesn't limit the number of time series written to remote storage systems specified at -remoteWrite.url. The limit can be enforced by setting the following command-line flags:

• -remoteWrite.maxHourlySeries - limits the number of unique time series vmagent can write to remote storage systems during the last hour. Useful for limiting the number of active time series.
• -remoteWrite.maxDailySeries - limits the number of unique time series vmagent can write to remote storage systems during the last day. Useful for limiting daily churn rate.

Both limits can be set simultaneously. If any of these limits is reached, then samples for new time series are dropped instead of sending them to remote storage systems. A sample of dropped series is put in the log with WARNING level.

vmagent exposes the following metrics at http://vmagent:8429/metrics page (see monitoring docs for details):

• vmagent_hourly_series_limit_rows_dropped_total - the number of metrics dropped due to exceeded hourly limit on the number of unique time series.
• vmagent_hourly_series_limit_max_series - the hourly series limit set via -remoteWrite.maxHourlySeries.
• vmagent_hourly_series_limit_current_series - the current number of unique series registered during the last hour.
• vmagent_daily_series_limit_rows_dropped_total - the number of metrics dropped due to exceeded daily limit on the number of unique time series.
• vmagent_daily_series_limit_max_series - the daily series limit set via -remoteWrite.maxDailySeries.
• vmagent_daily_series_limit_current_series - the current number of unique series registered during the last day.

These limits are approximate, so vmagent can underflow/overflow the limit by a small percentage (usually less than 1%).

See also cardinality explorer docs.

Monitoring

vmagent exports various metrics in Prometheus exposition format at http://vmagent-host:8429/metrics page. We recommend setting up regular scraping of this page either through vmagent itself or by Prometheus-compatible scraper, so that the exported metrics may be analyzed later.

If you use Google Cloud Managed Prometheus for scraping metrics from VictoriaMetrics components, then pass -metrics.exposeMetadata command-line to them, so they add TYPE and HELP comments per each exposed metric at /metrics page. See these docs for details.

Use official Grafana dashboard for vmagent state overview. Graphs on this dashboard contain useful hints - hover the i icon at the top left corner of each graph in order to read it. If you have suggestions for improvements or have found a bug - please open an issue on github or add a review to the dashboard.

vmagent also exports the status for various targets at the following pages:

• http://vmagent-host:8429/targets. This pages shows the current status for every active target.
• http://vmagent-host:8429/service-discovery. This pages shows the list of discovered targets with the discovered __meta_* labels according to these docs. This page may help debugging target relabeling.
• http://vmagent-host:8429/api/v1/targets. This handler returns JSON response compatible with the corresponding page from Prometheus API.
• http://vmagent-host:8429/ready. This handler returns http 200 status code when vmagent finishes its initialization for all the service_discovery configs. It may be useful to perform vmagent rolling update without any scrape loss.

Troubleshooting

• It is recommended to set up the official Grafana dashboard in order to monitor the state of vmagent.

• It is recommended to increase the maximum number of open files in the system (ulimit -n) when scraping a large number of targets, as vmagent establishes at least one TCP connection per target.

• If vmagent uses too much RAM or CPU, then follow these recommendations.

• When vmagent scrapes many unreliable targets, it can flood the error log with scrape errors. It is recommended to investigate and fix these errors. If it is unfeasible to fix all the reported errors, then they 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 and http://vmagent-host:8429/api/v1/targets.

• The http://vmagent-host:8429/config page shows current active -promscrape.config configuration. Access to endpoint can be protected via -configAuthKey command-line flag.

• Pages http://vmagent-host:8429/remotewrite-relabel-config and http://vmagent-host:8429/remotewrite-url-relabel-config {{% available_from "v1.129.0" %}} show current active -remoteWrite.relabelConfig and -remoteWrite.urlRelabelConfig configuration correspondingly. Access to endpoints can be protected via -configAuthKey command-line flag.

• The http://vmagent-host:8429/service-discovery page could be useful for debugging the relabeling process for scrape targets. This page contains original labels for targets dropped during relabeling. By default, the -promscrape.maxDroppedTargets targets are shown here. If your setup drops more targets during relabeling, then increase -promscrape.maxDroppedTargets command-line flag value to see all the dropped targets. Note that tracking each dropped target requires up to 10Kb of RAM. Therefore, big values for -promscrape.maxDroppedTargets may result in increased memory usage if a large number of scrape targets are dropped during relabeling.

• It is recommended to increase -remoteWrite.queues if vmagent_remotewrite_pending_data_bytes metric grows constantly. It is also recommended to increase -remoteWrite.maxBlockSize and -remoteWrite.maxRowsPerBlock command-line flags in this case. This can improve data ingestion performance to the configured remote storage systems at the cost of higher memory usage.

• If you see gaps in the data pushed by vmagent to remote storage when -remoteWrite.maxDiskUsagePerURL is set, try increasing -remoteWrite.queues. Such gaps may appear because vmagent cannot keep up with sending the collected data to remote storage. Therefore, it starts dropping the buffered data if the on-disk buffer size exceeds -remoteWrite.maxDiskUsagePerURL.

• vmagent drops data blocks if remote storage replies with 400 Bad Request or 409 Conflict HTTP responses. The number of dropped blocks can be monitored via vmagent_remotewrite_packets_dropped_total metric exported at /metrics page.

• Use -remoteWrite.queues=1 when -remoteWrite.url points to remote storage, which doesn't accept out-of-order samples (aka data backfilling). Such storage systems include Prometheus, Mimir, Cortex and Thanos, which typically emit out of order sample errors. The best solution is to use remote storage with backfilling support such as VictoriaMetrics.

• vmagent buffers scraped data at the -remoteWrite.tmpDataPath directory until it is sent to -remoteWrite.url. The directory can grow large when remote storage is unavailable for extended periods of time and if the maximum directory size isn't limited with -remoteWrite.maxDiskUsagePerURL command-line flag. If you don't want to send all the buffered data from the directory to remote storage then simply stop vmagent and delete the directory.

• If vmagent runs on a host with slow persistent storage, which cannot keep up with the volume of processed samples, then it is possible to disable the persistent storage with -remoteWrite.disableOnDiskQueue command-line flag. See these docs for more details.

• By default vmagent masks -remoteWrite.url with secret-url values in logs and at /metrics page because the url may contain sensitive information such as auth tokens or passwords. Pass -remoteWrite.showURL command-line flag when starting vmagent in order to see all the valid urls.

• By default vmagent evenly spreads scrape load in time. If a particular scrape target must be scraped at the beginning of some interval, then scrape_align_interval option must be used. For example, the following config aligns hourly scrapes to the beginning of hour:

  scrape_configs:
  - job_name: foo
    scrape_interval: 1h
    scrape_align_interval: 1h
  

• By default vmagent evenly spreads scrape load in time. If a particular scrape target must be scraped at specific offset, then scrape_offset option must be used. For example, the following config instructs vmagent to scrape the target at 10 seconds of every minute:

  scrape_configs:
  - job_name: foo
    scrape_interval: 1m
    scrape_offset: 10s
  

• If you see skipping duplicate scrape target with identical labels errors when scraping Kubernetes pods, then it is likely these pods listen to multiple ports or they use an init container. These errors can either be fixed or suppressed with the -promscrape.suppressDuplicateScrapeTargetErrors command-line flag. See the available options below if you prefer fixing the root cause of the error:

  The following relabeling rule may be added to relabel_configs section in order to filter out pods with unneeded ports:

  - action: keep_if_equal
    source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_port, __meta_kubernetes_pod_container_port_number]
  

  The following relabeling rule may be added to relabel_configs section in order to filter out init container pods:

  - action: drop
    source_labels: [__meta_kubernetes_pod_container_init]
    regex: true
  

See also:

• Relabel Debugging
• General Troubleshooting

Calculating disk space for persistence queue

vmagent buffers collected metrics on disk at the directory specified via -remoteWrite.tmpDataPath command-line flag until the metrics are sent to remote storage configured via -remoteWrite.url command-line flag. The -remoteWrite.tmpDataPath directory can grow large when remote storage is unavailable for extended periods of time and if the maximum directory size isn't limited with -remoteWrite.maxDiskUsagePerURL command-line flag.

To estimate the allocated disk size for persistent queue, or to estimate -remoteWrite.maxDiskUsagePerURL command-line flag value, take into account the following attributes:

1. The size in bytes of data stream sent by vmagent:

   Run the query sum(rate(vmagent_remotewrite_bytes_sent_total[1h])) by(instance,url) in vmui or Grafana to get the amount of bytes sent by each vmagent instance per second.

2. The amount of time a persistent queue should keep the data before starting to drop it.

   For example, if vmagent should be able to buffer the data for at least 6 hours, then the following query can be used for estimating the needed amounts of disk space in gigabytes:

   sum(rate(vmagent_remotewrite_bytes_sent_total[1h])) by(instance,url) * 6h / 1Gi
   

Additional notes:

1. Ensure that vmagent monitoring is configured properly.
2. Re-evaluate the estimation each time when:
   • there is an increase in the vmagent's workload
   • there is a change in relabeling rules which could increase the amount metrics to send
   • there is a change in number of configured -remoteWrite.url addresses
3. The minimum disk size to allocate for the persistent queue is 500Mi per each -remoteWrite.url.
4. On-disk persistent queue can be disabled if needed. See these docs.

Security

See general recommendations regarding security.

vmagent's /remotewrite-relabel-config and /remotewrite-url-relabel-config endpoints {{% available_from "v1.129.0" %}} can be protected via -configAuthKey command-line flag.

mTLS protection

By default vmagent accepts http requests at port 8429 (this port can be changed via -httpListenAddr command-line flags), it is expected to run in an isolated trusted network. Enterprise version of vmagent supports the ability to accept mTLS requests at this port, by specifying -tls and -mtls command-line flags. For example, the following command runs vmagent, which accepts only mTLS requests at port 8429:

./vmagent -tls -mtls -remoteWrite.url=...

By default, the system-wide TLS Root CA is used for verifying client certificates if -mtls command-line flag is specified. It is possible to specify custom TLS Root CA via -mtlsCAFile command-line flag.

Performance optimizations

vmagent is optimized for low CPU usage and low RAM usage without the need to tune any configs. Sometimes it is needed to optimize the CPU / RAM usage of vmagent even more. For example, if vmagent needs to scrape thousands of targets in resource-constrained environments, then the following options may help to reduce CPU and RAM usage:

• Set GOGC environment variable to 100. This reduces CPU usage at the cost of higher RAM usage.

• Set GOMAXPROCS environment variable to the value slightly bigger than the number of CPU cores used by vmagent. Another option is to set CPU limit in Kubernetes / Docker to the integer value bigger than the number of CPU cores used by vmagent. This reduces RAM and CPU usage when vmagent runs in an environment with a large number of available CPU cores. Note that it may be necessary to increase the -remoteWrite.queues command-line flag to a larger value if GOMAXPROCS is set to too small of a value, since by default -remoteWrite.queues is proportional to GOMAXPROCS.

• Avoid increasing the value for the -maxConcurrentRequests command-line flag. The default value is optimized for the best performance in most cases, even if many clients send data to vmagent via many concurrent connections and the number of these connections significantly exceeds the default value for the -maxConcurrentRequests command-line flag. vmagent puts incoming requests into a wait queue if the number of concurrently executed requests exceeds -maxConcurrentRequests. The pending requests at the wait queue do not consume CPU and do not consume significant amounts of RAM, so it is OK to have thousands of pending requests in the wait queue. Pending requests in the wait queue are canceled if they wait for their exection for longer than the duration specified in the -insert.maxQueueDuration command-line flag. Canceled requests can be monitored via vm_concurrent_insert_limit_timeout_total metric.

• Disable response compression at scrape targets via -promscrape.disableCompression command-line flag or via disable_compression: true option in the scrape_config. This reduces CPU usage at the cost of higher network bandwidth usage between vmagent and scrape targets.

• Disable tracking of original labels for the discovered targets via -promscrape.dropOriginalLabels command-line flag. This helps to reduce RAM usage when vmagent discovers a large number of scrape targets and the majority of these targets are dropped. This is a typical case when vmagent discovers Kubernetes targets. The downside of using -promscrape.dropOriginalLabels command-line flag is the reduced debuggability for improperly configured per-target relabeling.

• Disable staleness markers via -promscrape.noStaleMarkers command-line flag or via no_stale_markers: true option in the scrape_config. This reduces RAM and CPU usage. Note that disabling staleness markers may result in unexpected query results when scrape targets are frequently rotated (this is a typical case in Kubernetes).

• Set -memory.allowedBytes command-line flag to a value close to the actual memory usage of vmagent. Another option is to set memory limit in Kubernetes / Docker to the value 50% larger than the actual memory usage of vmagent. This should reduce memory usage spikes for vmagent running in the environment with a large amount of available memory and when the remote storage cannot keep up with the data ingestion rate. Increasing -remoteWrite.queues command-line flag value may help in this case too.

• In extreme cases it may be useful to set -promscrape.disableKeepAlive command-line flag in order to save RAM on HTTP keep-alive connections to thousands of scrape targets.

• Increase the scrape_interval option in the global section of the -promscrape.config and/or at every scrape_config to reduce CPU usage. For example, increasing the scrape_interval from 10s to 30s across all the targets decreases CPU usage at vmagent by up to 3x.

Example command, which runs vmagent in an optimized mode:

GOGC=100 GOMAXPROCS=1 ./vmagent -promscrape.disableCompression -promscrape.dropOriginalLabels -promscrape.noStaleMarkers -memory.allowedBytes=1GiB -promscrape.disableKeepAlive ...

How to build from sources

We recommend using the official binary releases - vmagent is located in the vmutils-... archives.

It may be needed to build vmagent from source code when developing or testing new feature or bugfix.

Development build

1. Install Go.
2. Run make vmagent from the root folder of the repository. It builds the vmagent binary and puts it into the bin folder.

Production build

1. Install docker.
2. Run make vmagent-prod from the root folder of the repository. It builds vmagent-prod binary and puts it into the bin folder.

Building docker images

Run make package-vmagent. It builds victoriametrics/vmagent:<PKG_TAG> docker image locally. <PKG_TAG> is an 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-vmagent.

The base docker image is alpine but it is possible to use 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:

ROOT_IMAGE=scratch make package-vmagent

ARM build

ARM build may run on Raspberry Pi or on energy-efficient ARM servers.

Development ARM build

1. Install Go.
2. Run make vmagent-linux-arm or make vmagent-linux-arm64 from the root folder of the repository It builds vmagent-linux-arm or vmagent-linux-arm64 binary respectively and puts it into the bin folder.

Production ARM build

1. Install docker.
2. Run make vmagent-linux-arm-prod or make vmagent-linux-arm64-prod from the root folder of the repository. It builds vmagent-linux-arm-prod or vmagent-linux-arm64-prod binary respectively and puts it into the bin folder.

Profiling

vmagent provides handlers for collecting the following Go profiles:

• Memory profile can be collected with the following command (replace 0.0.0.0 with hostname if needed):

curl http://0.0.0.0:8429/debug/pprof/heap > mem.pprof

• CPU profile can be collected with the following command (replace 0.0.0.0 with hostname if needed):

curl http://0.0.0.0:8429/debug/pprof/profile > cpu.pprof

The command for collecting CPU profile waits for 30 seconds before returning.

The collected profiles may be analyzed with go tool pprof.

It is safe sharing the collected profiles from security point of view, since they do not contain sensitive information.

Advanced usage

vmagent can be fine-tuned with various command-line flags. Run ./vmagent -help in order to see the full list of these flags with their descriptions and default value.

Common flags

These flags are available in both VictoriaMetrics OSS and VictoriaMetrics Enterprise. {{% content "vmagent_common_flags.md" %}}

Enterprise flags

These flags are available only in VictoriaMetrics enterprise. {{% content "vmagent_enterprise_flags.md" %}}

---

Section below contains backward-compatible anchors for links that were moved or renamed.

Relabeling

Moved to Relabeling Cookbook.

Relabeling enhancements

Moved to Relabeling Cookbook.

Graphite relabeling

Moved to Relabeling Cookbook.

Relabel debug

Moved to Relabeling Cookbook.

Debugging scrape targets

Moved to Relabeling Cookbook.

Google PubSub integration

Moved to integrations/pubsub.

Reading metrics from PubSub

Moved to integrations/pubsub/#reading-metrics.

Consume metrics from multiple topics

Moved to integrations/pubsub/#multiple-topics.

Command-line flags for PubSub consumer

Moved to integrations/pubsub/#consumer-command-line-flags.

Writing metrics to PubSub

Moved to integrations/pubsub/#writing-metrics.

Command-line flags for PubSub producer

Moved to integrations/pubsub/#producer-command-line-flags.

Kafka integration

Moved to integrations/kafka.

Reading metrics from Kafka

Moved to integrations/kafka/#reading-metrics.

Command-line flags for Kafka consumer

Moved to integrations/kafka/#consumer-command-line-flags.

Writing metrics to Kafka

Moved to integrations/kafka/#writing-metrics.

Estimating message size and rate

Moved to integrations/kafka/#estimating-message-size-and-rate.

Kafka broker authorization and authentication

Moved to integrations/kafka/#kafka-broker-authorization-and-authentication.