It is recommended to use the [single-node version](https://github.com/VictoriaMetrics/VictoriaMetrics) instead of the cluster version
for ingestion rates lower than a million data points per second.
The single-node version [scales perfectly](https://medium.com/@valyala/measuring-vertical-scalability-for-time-series-databases-in-google-cloud-92550d78d8ae)
by a separate service sitting in front of VictoriaMetrics cluster such as [vmauth](https://docs.victoriametrics.com/vmauth.html) or [vmgateway](https://docs.victoriametrics.com/vmgateway.html). [Contact us](mailto:info@victoriametrics.com) if you need assistance with such service.
- The database performance and resource usage doesn't depend on the number of tenants. It depends mostly on the total number of active time series in all the tenants. A time series is considered active if it received at least a single sample during the last hour or it has been touched by queries during the last hour.
Compiled binaries for the cluster version are available in the `assets` section of the [releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases).
It is recommended to run at least two nodes for each service for high availability purposes. In this case the cluster continues working when a single node is temporarily unavailable and the remaining nodes can handle the increased workload. The node may be temporarily unavailable when the underlying hardware breaks, during software upgrades, migration or other maintenance tasks.
It is preferred to run many small `vmstorage` nodes over a few big `vmstorage` nodes, since this reduces the workload increase on the remaining `vmstorage` nodes when some of `vmstorage` nodes become temporarily unavailable.
An http load balancer such as [vmauth](https://docs.victoriametrics.com/vmauth.html) or `nginx` must be put in front of `vminsert` and `vmselect` nodes. It must contain the following routing configs according to [the url format](#url-format):
- [An example docker-compose config for VictoriaMetrics cluster](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/deployment/docker/docker-compose.yml)
- [Helm charts for VictoriaMetrics](https://github.com/VictoriaMetrics/helm-charts)
- [Kubernetes operator for VictoriaMetrics](https://github.com/VictoriaMetrics/operator)
It is possible manualy setting up a toy cluster on a single host. In this case every cluster component - `vminsert`, `vmselect` and `vmstorage` - must have distinct values for `-httpListenAddr` command-line flag. This flag specifies http address for accepting http requests for [monitoring](#monitoring) and [profiling](#profiling). `vmstorage` node must have distinct values for the following additional command-line flags in order to prevent resource usage clash:
- Each `.` in flag names must be substituted by `_` (for example `-insert.maxQueueDuration <duration>` will translate to `insert_maxQueueDuration=<duration>`)
- For repeating flags, an alternative syntax can be used by joining the different values into one using `,` as separator (for example `-storageNode <nodeA> -storageNode <nodeB>` will translate to `storageNode=<nodeA>,<nodeB>`)
- It is possible setting prefix for environment vars with `-envflag.prefix`. For instance, if `-envflag.prefix=VM_`, then env vars must be prepended with `VM_`
By default `vminsert` and `vmselect` nodes use unencrypted connections to `vmstorage` nodes, since it is assumed that all the cluster components run in a protected environment. [Enterprise version of VictoriaMetrics](https://victoriametrics.com/products/enterprise/) provides optional support for [mTLS connections](https://en.wikipedia.org/wiki/Mutual_authentication#mTLS) between cluster components. Pass `-cluster.tls=true` command-line flag to `vminsert`, `vmselect` and `vmstorage` nodes in order to enable mTLS protection. Additionally, `vminsert`, `vmselect` and `vmstorage` must be configured with mTLS certificates via `-cluster.tlsCertFile`, `-cluster.tlsKeyFile` command-line options. These certificates are mutually verified when `vminsert` and `vmselect` dial `vmstorage`.
-`-cluster.tlsInsecureSkipVerify` can be set at `vminsert`, `vmselect` and `vmstorage` in order to disable peer certificate verification. Note that this breaks security.
-`-cluster.tlsCAFile` can be set at `vminsert`, `vmselect` and `vmstorage` for verifying peer certificates issued with custom [certificate authority](https://en.wikipedia.org/wiki/Certificate_authority). By default system-wide certificate authority is used for peer certificate verification.
-`-cluster.tlsCipherSuites` can be set to the list of supported TLS cipher suites at `vmstorage`. See [the list of supported TLS cipher suites](https://pkg.go.dev/crypto/tls#pkg-constants).
When `vmselect` runs with `-clusternativeListenAddr` command-line option, then it can be configured with `-clusternative.tls*` options similar to `-cluster.tls*` for accepting `mTLS` connections from top-level `vmselect` nodes in [multi-level cluster setup](#multi-level-cluster-setup).
[Enterprise version of VictoriaMetrics](https://victoriametrics.com/products/enterprise/) can be downloaded and evaluated for free from [the releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases).
All the cluster components expose various metrics in Prometheus-compatible format at `/metrics` page on the TCP port set in `-httpListenAddr` command-line flag.
or [an alternative dashboard for VictoriaMetrics cluster](https://grafana.com/grafana/dashboards/11831). Graphs on these dashboards contain useful hints - hover the `i` icon at the top left corner of each graph in order to read it.
It is recommended setting up alerts in [vmalert](https://docs.victoriametrics.com/vmalert.html) or in Prometheus from [this config](https://github.com/VictoriaMetrics/VictoriaMetrics/blob/cluster/deployment/docker/alerts.yml).
`vmstorage` nodes automatically switch to readonly mode when the directory pointed by `-storageDataPath` contains less than `-storage.minFreeDiskSpaceBytes` of free space. `vminsert` nodes stop sending data to such nodes and start re-routing the data to the remaining `vmstorage` nodes.
-`<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`.
-`prometheus` and `prometheus/api/v1/write` - for inserting data with [Prometheus remote write API](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#remote_write).
-`datadog/api/v1/series` - for inserting data with [DataDog submit metrics API](https://docs.datadoghq.com/api/latest/metrics/#submit-metrics). See [these docs](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html#how-to-send-data-from-datadog-agent) for details.
-`influx/write` and `influx/api/v2/write` - for inserting data with [InfluxDB line protocol](https://docs.influxdata.com/influxdb/v1.7/write_protocols/line_protocol_tutorial/). See [these docs](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html#how-to-send-data-from-influxdb-compatible-agents-such-as-telegraf) for details.
-`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://docs.victoriametrics.com/Single-server-VictoriaMetrics.html#sending-opentsdb-data-via-http-apiput-requests) for details.
-`prometheus/api/v1/import` - for importing data obtained via `api/v1/export` at `vmselect` (see below).
-`prometheus/api/v1/import/native` - for importing data obtained via `api/v1/export/native` on `vmselect` (see below).
-`prometheus/api/v1/import/csv` - for importing arbitrary CSV data. See [these docs](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html#how-to-import-csv-data) for details.
-`prometheus/api/v1/import/prometheus` - for importing data in [Prometheus text exposition format](https://github.com/prometheus/docs/blob/master/content/docs/instrumenting/exposition_formats.md#text-based-format) and in [OpenMetrics format](https://github.com/OpenObservability/OpenMetrics/blob/master/specification/OpenMetrics.md). See [these docs](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html#how-to-import-data-in-prometheus-exposition-format) for details.
- URLs for [Prometheus querying API](https://prometheus.io/docs/prometheus/latest/querying/api/): `http://<vmselect>:8481/select/<accountID>/prometheus/<suffix>`, where:
-`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).
-`api/v1/export` - exports raw data in JSON line format. See [this article](https://medium.com/@valyala/analyzing-prometheus-data-with-external-tools-5f3e5e147639) for details.
-`api/v1/export/native` - exports raw data in native binary format. It may be imported into another VictoriaMetrics via `api/v1/import/native` (see above).
-`render` - implements Graphite Render API. See [these docs](https://graphite.readthedocs.io/en/stable/render_api.html). This functionality is available in [Enterprise package](https://victoriametrics.com/products/enterprise/). Enterprise binaries can be downloaded and evaluated for free from [the releases page](https://github.com/VictoriaMetrics/VictoriaMetrics/releases).
-`metrics/find` - searches Graphite metrics. See [these docs](https://graphite-api.readthedocs.io/en/latest/api.html#metrics-find).
-`metrics/expand` - expands Graphite metrics. See [these docs](https://graphite-api.readthedocs.io/en/latest/api.html#metrics-expand).
-`metrics/index.json` - returns all the metric names. See [these docs](https://graphite-api.readthedocs.io/en/latest/api.html#metrics-index-json).
-`tags/tagSeries` - registers time series. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#adding-series-to-the-tagdb).
-`tags/tagMultiSeries` - register multiple time series. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#adding-series-to-the-tagdb).
-`tags` - returns tag names. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#exploring-tags).
-`tags/<tag_name>` - returns tag values for the given `<tag_name>`. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#exploring-tags).
-`tags/findSeries` - returns series matching the given `expr`. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#exploring-tags).
-`tags/autoComplete/tags` - returns tags matching the given `tagPrefix` and/or `expr`. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#auto-complete-support).
-`tags/autoComplete/values` - returns tag values matching the given `valuePrefix` and/or `expr`. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#auto-complete-support).
-`tags/delSeries` - deletes series matching the given `path`. See [these docs](https://graphite.readthedocs.io/en/stable/tags.html#removing-series-from-the-tagdb).
- URL with basic Web UI: `http://<vmselect>:8481/select/<accountID>/vmui/`.
- URL for query stats across all tenants: `http://<vmselect>:8481/api/v1/status/top_queries`. It lists with the most frequently executed queries and queries taking the most duration.
- URL for time series deletion: `http://<vmselect>:8481/delete/<accountID>/prometheus/api/v1/admin/tsdb/delete_series?match[]=<timeseries_selector_for_delete>`.
Note that the `delete_series` handler should be used only in exceptional cases such as deletion of accidentally ingested incorrect time series. It shouldn't
be used on a regular basis, since it carries non-zero overhead.
which can be used for backups in background. Snapshots are created in `<storageDataPath>/snapshots` folder, where `<storageDataPath>` is the corresponding
command-line flag value.
-`/snapshot/list` - list available snasphots.
-`/snapshot/delete?snapshot=<id>` - delete the given snapshot.
-`/snapshot/delete_all` - delete all the snapshots.
Snapshots may be created independently on each `vmstorage` node. There is no need in synchronizing snapshots' creation
- Adding more CPU and RAM to existing `vmselect` nodes improves the performance for heavy queries, which process big number of time series with big number of raw samples. See [this article on how to detect and optimize heavy queries](https://valyala.medium.com/how-to-optimize-promql-and-metricsql-queries-85a1b75bf986).
- Adding more `vmstorage` nodes increases the number of [active time series](https://docs.victoriametrics.com/FAQ.html#what-is-an-active-time-series) the cluster can handle. This also increases query performance over time series with [high churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate). The cluster stability is also improved with the number of `vmstorage` nodes, since active `vmstorage` nodes need to handle lower additional workload when some of `vmstorage` nodes become unavailable.
- Adding more CPU and RAM to existing `vmstorage` nodes increases the number of [active time series](https://docs.victoriametrics.com/FAQ.html#what-is-an-active-time-series) the cluster can handle. It is preferred to add more `vmstorage` nodes over adding more CPU and RAM to existing `vmstorage` nodes, since higher number of `vmstorage` nodes increases cluster stability and improves query performance over time series with [high churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate).
- Adding more `vminsert` nodes increases the maximum possible data ingestion speed, since the ingested data may be split among bigger number of `vminsert` nodes.
- Adding more `vmselect` nodes increases the maximum possible queries rate, since the incoming concurrent requests may be split among bigger number of `vmselect` nodes.
-`vmselect` continues serving partial responses if at least a single `vmstorage` node is available. If consistency over availability is preferred, then either pass `-search.denyPartialResponse` command-line flag to `vmselect` or pass `deny_partial_response=1` query arg in requests to `vmselect`.
`vmselect` doesn't serve partial responses for API handlers returning raw datapoints - [`/api/v1/export*` endpoints](https://docs.victoriametrics.com/#how-to-export-time-series), since users usually expect this data is always complete.
VictoriaMetrics uses lower amounts of CPU, RAM and storage space on production workloads compared to competing solutions (Prometheus, Thanos, Cortex, TimescaleDB, InfluxDB, QuestDB, M3DB) according to [our case studies](https://docs.victoriametrics.com/CaseStudies.html).
Each node type - `vminsert`, `vmselect` and `vmstorage` - can run on the most suitable hardware. Cluster capacity scales linearly with the available resources. The needed amounts of CPU and RAM per each node type highly depends on the workload - the number of [active time series](https://docs.victoriametrics.com/FAQ.html#what-is-an-active-time-series), [series churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate), query types, query qps, etc. It is recommended setting up a test VictoriaMetrics cluster for your production workload and iteratively scaling per-node resources and the number of nodes per node type until the cluster becomes stable. It is recommended setting up [monitoring for the cluster](#monitoring). It helps determining bottlenecks in cluster setup. It is also recommended following [the troubleshooting docs](https://docs.victoriametrics.com/#troubleshooting).
The needed storage space for the given retention (the retention is set via `-retentionPeriod` command-line flag at `vmstorage`) can be extrapolated from disk space usage in a test run. For example, if the storage space usage is 10GB after a day-long test run on a production workload, then it will need at least `10GB*100=1TB` of disk space for `-retentionPeriod=100d` (100-days retention period). Storage space usage can be monitored with [the official Grafana dashboard for VictoriaMetrics cluster](#monitoring).
- 50% of free RAM across all the node types for reducing the probability of OOM (out of memory) crashes and slowdowns during temporary spikes in workload.
- 50% of spare CPU across all the node types for reducing the probability of slowdowns during temporary spikes in workload.
- At least 20% of free storage space at the directory pointed by `-storageDataPath` command-line flag at `vmstorage` nodes. See also `-storage.minFreeDiskSpaceBytes` command-line flag [description for vmstorage](#list-of-command-line-flags-for-vmstorage).
- The [replication](#replication-and-data-safety) increases the amounts of needed resources for the cluster by up to `N` times where `N` is replication factor. This is because `vminsert` stores `N` copies of every ingested sample on distinct `vmstorage` nodes. These copies are de-duplicated by `vmselect` during querying. The most cost-efficient and performant solution for data durability is to rely on replicated durable persistent disks such as [Google Compute persistent disks](https://cloud.google.com/compute/docs/disks#pdspecs) instead of using the [replication at VictoriaMetrics level](#replication-and-data-safety).
- It is recommended to run a cluster with big number of small `vmstorage` nodes instead of a cluster with small number of big `vmstorage` nodes. This increases chances that the cluster remains available and stable when some of `vmstorage` nodes are temporarily unavailable during maintenance events such as upgrades, configuration changes or migrations. For example, when a cluster contains 10 `vmstorage` nodes and a single node becomes temporarily unavailable, then the workload on the remaining 9 nodes increases by `1/9=11%`. When a cluster contains 3 `vmstorage` nodes and a single node becomes temporarily unavailable, then the workload on the remaining 2 nodes increases by `1/2=50%`. The remaining `vmstorage` nodes may have no enough free capacity for handling the increased workload. In this case the cluster may become overloaded, which may result to decreased availability and stability.
- Cluster capacity for [active time series](https://docs.victoriametrics.com/FAQ.html#what-is-an-active-time-series) can be increased by increasing RAM and CPU resources per each `vmstorage` node or by adding new `vmstorage` nodes.
- Query latency can be reduced by increasing CPU resources per each `vmselect` node, since each incoming query is processed by a single `vmselect` node. Performance for heavy queries scales with the number of available CPU cores at `vmselect` node, since `vmselect` processes time series referred by the query on all the available CPU cores.
- If the cluster needs to process incoming queries at a high rate, then its capacity can be increased by adding more `vmselect` nodes, so incoming queries could be spread among bigger number of `vmselect` nodes.
- By default `vminsert` compresses the data it sends to `vmstorage` in order to reduce network bandwidth usage. The compression takes additional CPU resources at `vminsert`. If `vminsert` nodes have limited CPU, then the compression can be disabled by passing `-rpc.disableCompression` command-line flag at `vminsert` nodes.
- By default `vmstorage` compresses the data it sends to `vmselect` during queries in order to reduce network bandwidth usage. The compression takes additional CPU resources at `vmstorage`. If `vmstorage` nodes have limited CPU, then the compression can be disabled by passing `-rpc.disableCompression` command-line flag at `vmstorage` nodes.
See also [resource usage limits docs](#resource-usage-limits).
## Resource usage limits
By default cluster components of VictoriaMetrics are tuned for an optimal resource usage under typical workloads. Some workloads may need fine-grained resource usage limits. In these cases the following command-line flags may be useful:
-`-memory.allowedPercent` and `-search.allowedBytes` limit the amounts of memory, which may be used for various internal caches at all the cluster components of VictoriaMetrics - `vminsert`, `vmselect` and `vmstorage`. Note that VictoriaMetrics components may use more memory, since these flags don't limit additional memory, which may be needed on a per-query basis.
-`-search.maxUniqueTimeseries` at `vmselect` component limits the number of unique time series a single query can find and process. `vmselect` passes the limit to `vmstorage` component, which keeps in memory some metainformation about the time series located by each query and spends some CPU time for processing the found time series. This means that the maximum memory usage and CPU usage a single query can use at `vmstorage` is proportional to `-search.maxUniqueTimeseries`.
-`-search.maxQueryDuration` at `vmselect` limits the duration of a single query. If the query takes longer than the given duration, then it is canceled. This allows saving CPU and RAM at `vmselect` and `vmstorage` when executing unexpected heavy queries.
-`-search.maxConcurrentRequests` at `vmselect` limits the number of concurrent requests a single `vmselect` node can process. Bigger number of concurrent requests usually means bigger memory usage at both `vmselect` and `vmstorage`. For example, if a single query needs 100 MiB of additional memory during its execution, then 100 concurrent queries may need `100 * 100 MiB = 10 GiB` of additional memory. So it is better to limit the number of concurrent queries, while suspending additional incoming queries if the concurrency limit is reached. `vmselect` provides `-search.maxQueueDuration` command-line flag for limiting the max wait time for suspended queries.
-`-search.maxSamplesPerSeries` at `vmselect` limits the number of raw samples the query can process per each time series. `vmselect` sequentially processes raw samples per each found time series during the query. It unpacks raw samples on the selected time range per each time series into memory and then applies the given [rollup function](https://docs.victoriametrics.com/MetricsQL.html#rollup-functions). The `-search.maxSamplesPerSeries` command-line flag allows limiting memory usage at `vmselect` in the case when the query is executed on a time range, which contains hundreds of millions of raw samples per each located time series.
-`-search.maxSamplesPerQuery` at `vmselect` limits the number of raw samples a single query can process. This allows limiting CPU usage at `vmselect` for heavy queries.
-`-search.maxSeries` at `vmselect` limits the number of time series, which may be returned from [/api/v1/series](https://prometheus.io/docs/prometheus/latest/querying/api/#finding-series-by-label-matchers). This endpoint is used mostly by Grafana for auto-completion of metric names, label names and label values. Queries to this endpoint may take big amounts of CPU time and memory at `vmstorage` and `vmselect` when the database contains big number of unique time series because of [high churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate). In this case it might be useful to set the `-search.maxSeries` to quite low value in order limit CPU and memory usage.
-`-search.maxTagKeys` at `vmselect` limits the number of items, which may be returned from [/api/v1/labels](https://prometheus.io/docs/prometheus/latest/querying/api/#getting-label-names). This endpoint is used mostly by Grafana for auto-completion of label names. Queries to this endpoint may take big amounts of CPU time and memory at `vmstorage` and `vmselect` when the database contains big number of unique time series because of [high churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate). In this case it might be useful to set the `-search.maxTagKeys` to quite low value in order to limit CPU and memory usage.
-`-search.maxTagValues` at `vmselect` limits the number of items, which may be returned from [/api/v1/label/.../values](https://prometheus.io/docs/prometheus/latest/querying/api/#querying-label-values). This endpoint is used mostly by Grafana for auto-completion of label values. Queries to this endpoint may take big amounts of CPU time and memory at `vmstorage` and `vmselect` when the database contains big number of unique time series because of [high churn rate](https://docs.victoriametrics.com/FAQ.html#what-is-high-churn-rate). In this case it might be useful to set the `-search.maxTagValues` to quite low value in order to limit CPU and memory usage.
-`-storage.maxDailySeries` at `vmstorage` can be used for limiting the number of time series seen per day.
-`-storage.maxHourlySeries` at `vmstorage` can be used for limiting the number of [active time series](https://docs.victoriametrics.com/FAQ.html#what-is-an-active-time-series).
See also [capacity planning docs](#capacity-planning) and [cardinality limiter in vmagent](https://docs.victoriametrics.com/vmagent.html#cardinality-limiter).
The database is considered highly available if it continues accepting new data and processing incoming queries when some of its components are temporarily unavailable.
VictoriaMetrics cluster is highly available according to this definition - see [cluster availability docs](#cluster-availability).
Then an additional `vmselect` nodes can be configured for reading the data from multiple clusters according to [these docs](#multi-level-cluster-setup).
`vmselect` nodes can be queried by other `vmselect` nodes if they run with `-clusternativeListenAddr` command-line flag. For example, if `vmselect` is started with `-clusternativeListenAddr=:8401`, then it can accept queries from another `vmselect` nodes at TCP port 8401 in the same way as `vmstorage` nodes do. This allows chaining `vmselect` nodes and building multi-level cluster topologies. For example, the top-level `vmselect` node can query second-level `vmselect` nodes in different availability zones (AZ), while the second-level `vmselect` nodes can query `vmstorage` nodes in local AZ.
`vminsert` nodes can accept data from another `vminsert` nodes if they run with `-clusternativeListenAddr` command-line flag. For example, if `vminsert` is started with `-clusternativeListenAddr=:8400`, then it can accept data from another `vminsert` nodes at TCP port 8400 in the same way as `vmstorage` nodes do. This allows chaining `vminsert` nodes and building multi-level cluster topologies. For example, the top-level `vminsert` node can replicate data among the second level of `vminsert` nodes located in distinct availability zones (AZ), while the second-level `vminsert` nodes can spread the data among `vmstorage` nodes in local AZ.
*`vminsert` nodes at top level re-route incoming data to the remaining AZs when some AZs are temporariliy unavailable. This results in data gaps at AZs which were temporarily unavailable.
These issues are addressed by [vmagent](https://docs.victoriametrics.com/vmagent.html) when it runs in [multitenancy mode](https://docs.victoriametrics.com/vmagent.html#multitenancy). `vmagent` buffers data, which must be sent to a particular AZ, when this AZ is temporarily unavailable. The buffer is stored on disk. The buffered data is sent to AZ as soon as it becomes available.
By default VictoriaMetrics offloads replication to the underlying storage pointed by `-storageDataPath` such as [Google compute persistent disk](https://cloud.google.com/compute/docs/disks#pdspecs), which guarantees data durability. VictoriaMetrics supports application-level replication if replicated durable persistent disks cannot be used for some reason.
The replication can be enabled by passing `-replicationFactor=N` command-line flag to `vminsert`. This instructs `vminsert` to store `N` copies for every ingested sample on `N` distinct `vmstorage` nodes. This guarantees that all the stored data remains available for querying if up to `N-1``vmstorage` nodes are unavailable.
The cluster must contain at least `2*N-1``vmstorage` nodes, where `N` is replication factor, in order to maintain the given replication factor for newly ingested data when `N-1` of storage nodes are unavailable.
When the replication is enabled, `-dedup.minScrapeInterval=1ms` command-line flag must be passed to `vmselect` nodes, so they could de-duplicate replicated samples obtained from distinct `vmstorage` nodes during querying. If duplicate data is pushed to VictoriaMetrics from identically configured [vmagent](https://docs.victoriametrics.com/vmagent.html) instances or Prometheus instances, then the `-dedup.minScrapeInterval` must be set to bigger values according to [deduplication docs](#deduplication).
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.
Note that the replication increases resource usage - CPU, RAM, disk space, network bandwidth - by up to `-replicationFactor=N` times, because `vminsert` stores `N` copies of incoming data to distinct `vmstorage` nodes and `vmselect` needs to de-duplicate the replicated data obtained from `vmstorage` nodes during querying. So it is more cost-effective to offload the replication to underlying replicated durable storage pointed by `-storageDataPath` such as [Google Compute Engine persistent disk](https://cloud.google.com/compute/docs/disks/#pdspecs), which is protected from data loss and data corruption. It also provides consistently high performance and [may be resized](https://cloud.google.com/compute/docs/disks/add-persistent-disk) without downtime. HDD-based persistent disks should be enough for the majority of use cases. It is recommended using durable replicated persistent volumes in Kubernetes.
Cluster version of VictoriaMetrics supports data deduplication in the same way as single-node version do. See [these docs](https://docs.victoriametrics.com/#deduplication) for details. The only difference is that the same `-dedup.minScrapeInterval` command-line flag value must be passed to both `vmselect` and `vmstorage` nodes because of the following aspects:
By default `vminsert` tries to route all the samples for a single time series to a single `vmstorage` node. But samples for a single time series can be spread among multiple `vmstorage` nodes under certain conditions:
* when adding/removing `vmstorage` nodes. Then new samples for a part of time series will be routed to another `vmstorage` nodes;
* when `vmstorage` nodes are temporarily unavailable (for instance, during their restart). Then new samples are re-routed to the remaining available `vmstorage` nodes;
* when `vmstorage` node has no enough capacity for processing incoming data stream. Then `vminsert` re-routes new samples to other `vmstorage` nodes.
It is recommended performing periodical backups from [instant snapshots](https://medium.com/@valyala/how-victoriametrics-makes-instant-snapshots-for-multi-terabyte-time-series-data-e1f3fb0e0282)
for protecting from user errors such as accidental data deletion.
The following steps must be performed for each `vmstorage` node for creating a backup:
1. Create an instant snapshot by navigating to `/snapshot/create` HTTP handler. It will create snapshot and return its name.
2. Archive the created snapshot from `<-storageDataPath>/snapshots/<snapshot_name>` folder using [vmbackup](https://docs.victoriametrics.com/vmbackup.html).
Downsampling is available in [enterprise version of VictoriaMetrics](https://victoriametrics.com/products/enterprise/). It is configured with `-downsampling.period` command-line flag. The same flag value must be passed to both `vmstorage` and `vmselect` nodes. See [these docs](https://docs.victoriametrics.com/#downsampling) for details.
- Fragile gossip protocols. See [failed attempt in Thanos](https://github.com/improbable-eng/thanos/blob/030bc345c12c446962225221795f4973848caab5/docs/proposals/completed/201809_gossip-removal.md).
Path to TLS CA file to use for verifying certificates provided by -storageNode if -cluster.tls flag is set. By default system CA is used. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Path to client-side TLS certificate file to use when connecting to -storageNode if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Whether to skip verification of TLS certificates provided by -storageNode nodes if -cluster.tls flag is set. Note that disabled TLS certificate verification breaks security
Path to client-side TLS key file to use when connecting to -storageNode if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
TCP address to listen for data from other vminsert nodes in multi-level cluster setup. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#multi-level-cluster-setup . Usually :8400 must be set. Doesn't work if empty
Trim timestamps when importing csv data to this duration. Minimum practical duration is 1ms. Higher duration (i.e. 1s) may be used for reducing disk space usage for timestamp data (default 1ms)
Whether to disable re-routing when some of vmstorage nodes accept incoming data at slower speed compared to other storage nodes. Disabled re-routing limits the ingestion rate by the slowest vmstorage node. On the other side, disabled re-routing minimizes the number of active time series in the cluster during rolling restarts and during spikes in series churn rate. See also -dropSamplesOnOverload (default true)
Whether to drop incoming samples if the destination vmstorage node is overloaded and/or unavailable. This prioritizes cluster availability over consistency, e.g. the cluster continues accepting all the ingested samples, but some of them may be dropped if vmstorage nodes are temporarily unavailable and/or overloaded
Whether to enable reading flags from environment variables additionally to command line. Command line flag values have priority over values from environment vars. Flags are read only from command line if this flag isn't set. See https://docs.victoriametrics.com/#environment-variables for more details
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 read data files bigger than 2^32 bytes in memory. mmap() is usually faster for reading small data chunks than pread()
Trim timestamps for Graphite data to this duration. Minimum practical duration is 1s. Higher duration (i.e. 1m) may be used for reducing disk space usage for timestamp data (default 1s)
Incoming http connections are closed after the configured timeout. This may help to spread the incoming load among a cluster of services behind a load balancer. Please note that the real timeout may be bigger by up to 10% as a protection against the thundering herd problem (default 2m0s)
The maximum duration for a graceful shutdown of the HTTP server. A highly loaded server may require increased value for a graceful shutdown (default 7s)
An optional prefix to add to all the paths handled by http server. For example, if '-http.pathPrefix=/foo/bar' is set, then all the http requests will be handled on '/foo/bar/*' paths. This may be useful for proxied requests. See https://www.robustperception.io/using-external-urls-and-proxies-with-prometheus
Optional delay before http server shutdown. During this delay, the server returns non-OK responses from /health page, so load balancers can route new requests to other servers
The maximum length in bytes of a single line accepted by /api/v1/import; the line length can be limited with 'max_rows_per_line' query arg passed to /api/v1/export
Supports the following optional suffixes for size values: KB, MB, GB, KiB, MiB, GiB (default 104857600)
Comma-separated list of database names to return from /query and /influx/query API. This can be needed for accepting data from Telegraf plugins such as https://github.com/fangli/fluent-plugin-influxdb
Supports an array of values separated by comma or specified via multiple flags.
TCP and UDP address to listen for InfluxDB line protocol data. Usually :8089 must be set. Doesn't work if empty. This flag isn't needed when ingesting data over HTTP - just send it to http://<victoriametrics>:8428/write
Trim timestamps for InfluxDB line protocol data to this duration. Minimum practical duration is 1ms. Higher duration (i.e. 1s) may be used for reducing disk space usage for timestamp data (default 1ms)
Per-second limit on the number of ERROR messages. If more than the given number of errors are emitted per second, the remaining errors are suppressed. Zero values disable the rate limit
Timezone to use for timestamps in logs. Timezone must be a valid IANA Time Zone. For example: America/New_York, Europe/Berlin, Etc/GMT+3 or Local (default "UTC")
Per-second limit on the number of WARN messages. If more than the given number of warns are emitted per second, then the remaining warns are suppressed. Zero values disable the rate limit
The maximum number of concurrent inserts. Default value should work for most cases, since it minimizes the overhead for concurrent inserts. This option is tigthly coupled with -insert.maxQueueDuration (default 16)
The maximum length of label values in the accepted time series. Longer label values are truncated. In this case the vm_too_long_label_values_total metric at /metrics page is incremented (default 16384)
The maximum number of labels accepted per time series. Superfluous labels are dropped. In this case the vm_metrics_with_dropped_labels_total metric at /metrics page is incremented (default 30)
Allowed size of system memory VictoriaMetrics caches may occupy. This option overrides -memory.allowedPercent if set to a non-zero value. Too low a value may increase the cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache resulting in higher disk IO usage
Supports the following optional suffixes for size values: KB, MB, GB, KiB, MiB, GiB (default 0)
Allowed percent of system memory VictoriaMetrics caches may occupy. See also -memory.allowedBytes. Too low a value may increase cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache which will result in higher disk IO usage (default 60)
TCP and UDP address to listen for OpentTSDB metrics. Telnet put messages and HTTP /api/put messages are simultaneously served on TCP port. Usually :4242 must be set. Doesn't work if empty
Trim timestamps for OpenTSDB 'telnet put' data to this duration. Minimum practical duration is 1s. Higher duration (i.e. 1m) may be used for reducing disk space usage for timestamp data (default 1s)
Trim timestamps for OpenTSDB HTTP data to this duration. Minimum practical duration is 1ms. Higher duration (i.e. 1s) may be used for reducing disk space usage for timestamp data (default 1ms)
Optional path to a file with relabeling rules, which are applied to all the ingested metrics. The path can point either to local file or to http url. See https://docs.victoriametrics.com/#relabeling for details. The config is reloaded on SIGHUP signal
Whether to log metrics before and after relabeling with -relabelConfig. If the -relabelDebug is enabled, then the metrics aren't sent to storage. This is useful for debugging the relabeling configs
Replication factor for the ingested data, i.e. how many copies to make among distinct -storageNode instances. Note that vmselect must run with -dedup.minScrapeInterval=1ms for data de-duplication when replicationFactor is greater than 1. Higher values for -dedup.minScrapeInterval at vmselect is OK (default 1)
Whether to sort labels for incoming samples before writing them to storage. This may be needed for reducing memory usage at storage when the order of labels in incoming samples is random. For example, if m{k1="v1",k2="v2"} may be sent as m{k2="v2",k1="v1"}. Enabled sorting for labels can slow down ingestion performance a bit
Path to file with TLS certificate if -tls is set. Prefer ECDSA certs instead of RSA certs as RSA certs are slower. The provided certificate file is automatically re-read every second, so it can be dynamically updated
Optional list of TLS cipher suites for incoming requests over HTTPS if -tls is set. See the list of supported cipher suites at https://pkg.go.dev/crypto/tls#pkg-constants
Path to TLS CA file to use for verifying certificates provided by -storageNode if -cluster.tls flag is set. By default system CA is used. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Path to client-side TLS certificate file to use when connecting to -storageNode if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Whether to skip verification of TLS certificates provided by -storageNode nodes if -cluster.tls flag is set. Note that disabled TLS certificate verification breaks security
Path to client-side TLS key file to use when connecting to -storageNode if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Leave only the last sample in every time series per each discrete interval equal to -dedup.minScrapeInterval > 0. See https://docs.victoriametrics.com/#deduplication for details
Comma-separated downsampling periods in the format 'offset:period'. For example, '30d:10m' instructs to leave a single sample per 10 minutes for samples older than 30 days. See https://docs.victoriametrics.com/#downsampling for details
Supports an array of values separated by comma or specified via multiple flags.
Whether to enable reading flags from environment variables additionally to command line. Command line flag values have priority over values from environment vars. Flags are read only from command line if this flag isn't set. See https://docs.victoriametrics.com/#environment-variables for more details
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 read data files bigger than 2^32 bytes in memory. mmap() is usually faster for reading small data chunks than pread()
Trim timestamps for Graphite data to this duration. Minimum practical duration is 1s. Higher duration (i.e. 1m) may be used for reducing disk space usage for timestamp data (default 1s)
Incoming http connections are closed after the configured timeout. This may help to spread the incoming load among a cluster of services behind a load balancer. Please note that the real timeout may be bigger by up to 10% as a protection against the thundering herd problem (default 2m0s)
The maximum duration for a graceful shutdown of the HTTP server. A highly loaded server may require increased value for a graceful shutdown (default 7s)
An optional prefix to add to all the paths handled by http server. For example, if '-http.pathPrefix=/foo/bar' is set, then all the http requests will be handled on '/foo/bar/*' paths. This may be useful for proxied requests. See https://www.robustperception.io/using-external-urls-and-proxies-with-prometheus
Optional delay before http server shutdown. During this delay, the server returns non-OK responses from /health page, so load balancers can route new requests to other servers
Per-second limit on the number of ERROR messages. If more than the given number of errors are emitted per second, the remaining errors are suppressed. Zero values disable the rate limit
Timezone to use for timestamps in logs. Timezone must be a valid IANA Time Zone. For example: America/New_York, Europe/Berlin, Etc/GMT+3 or Local (default "UTC")
Per-second limit on the number of WARN messages. If more than the given number of warns are emitted per second, then the remaining warns are suppressed. Zero values disable the rate limit
Allowed size of system memory VictoriaMetrics caches may occupy. This option overrides -memory.allowedPercent if set to a non-zero value. Too low a value may increase the cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache resulting in higher disk IO usage
Supports the following optional suffixes for size values: KB, MB, GB, KiB, MiB, GiB (default 0)
Allowed percent of system memory VictoriaMetrics caches may occupy. See also -memory.allowedBytes. Too low a value may increase cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache which will result in higher disk IO usage (default 60)
The maximum duration since the current time for response data, which is always queried from the original raw data, without using the response cache. Increase this value if you see gaps in responses due to time synchronization issues between VictoriaMetrics and data sources (default 5m0s)
Whether to deny partial responses if a part of -storageNode instances fail to perform queries; this trades availability over consistency; see also -search.maxQueryDuration
The interval between datapoints stored in the database. It is used at Graphite Render API handler for normalizing the interval between datapoints in case it isn't normalized. It can be overriden by sending 'storage_step' query arg to /render API or by sending the desired interval via 'Storage-Step' http header during querying /render API (default 10s)
The time when data points become visible in query results after the collection. Too small value can result in incomplete last points for query results (default 30s)
The maximum number of concurrent search requests. It shouldn't be high, since a single request can saturate all the CPU cores. See also -search.maxQueueDuration (default 8)
The maximum number of time series, which can be scanned during queries to Graphite Render API. See https://docs.victoriametrics.com/#graphite-render-api-usage (default 300000)
Synonym to -search.lookback-delta from Prometheus. The value is dynamically detected from interval between time series datapoints if not set. It can be overridden on per-query basis via max_lookback arg. See also '-search.maxStalenessInterval' flag, which has the same meaining due to historical reasons
The maximum points per a single timeseries returned from /api/v1/query_range. This option doesn't limit the number of scanned raw samples in the database. The main purpose of this option is to limit the number of per-series points returned to graphing UI such as Grafana. There is no sense in setting this limit to values bigger than the horizontal resolution of the graph (default 30000)
The maximum number of raw samples a single query can process across all time series. This protects from heavy queries, which select unexpectedly high number of raw samples. See also -search.maxSamplesPerSeries (default 1000000000)
The maximum interval for staleness calculations. By default it is automatically calculated from the median interval between samples. This flag could be useful for tuning Prometheus data model closer to Influx-style data model. See https://prometheus.io/docs/prometheus/latest/querying/basics/#staleness for details. See also '-search.setLookbackToStep' flag
The maximum step when /api/v1/query_range handler adjusts points with timestamps closer than -search.latencyOffset to the current time. The adjustment is needed because such points may contain incomplete data (default 1m0s)
The maximum number of time series, which can be processed during the call to /api/v1/status/tsdb. This option allows limiting memory usage (default 10000000)
The maximum number of unique time series, which can be selected during /api/v1/query and /api/v1/query_range queries. This option allows limiting memory usage (default 300000)
The minimum interval for staleness calculations. This flag could be useful for removing gaps on graphs generated from time series with irregular intervals between samples. See also '-search.maxStalenessInterval'
Set this flag to true if the database doesn't contain Prometheus stale markers, so there is no need in spending additional CPU time on its handling. Staleness markers may exist only in data obtained from Prometheus scrape targets
The minimum duration for queries to track in query stats at /api/v1/status/top_queries. Queries with lower duration are ignored in query stats (default 1ms)
Whether to fix lookback interval to 'step' query arg value. If set to true, the query model becomes closer to InfluxDB data model. If set to true, then -search.maxLookback and -search.maxStalenessInterval are ignored
Whether to treat dots as is in regexp label filters used in queries. For example, foo{bar=~"a.b.c"} will be automatically converted to foo{bar=~"a\\.b\\.c"}, i.e. all the dots in regexp filters will be automatically escaped in order to match only dot char instead of matching any char. Dots in ".+", ".*" and ".{n}" regexps aren't escaped. This option is DEPRECATED in favor of {__graphite__="a.*.c"} syntax for selecting metrics matching the given Graphite metrics filter
Path to file with TLS certificate if -tls is set. Prefer ECDSA certs instead of RSA certs as RSA certs are slower. The provided certificate file is automatically re-read every second, so it can be dynamically updated
Optional list of TLS cipher suites for incoming requests over HTTPS if -tls is set. See the list of supported cipher suites at https://pkg.go.dev/crypto/tls#pkg-constants
Optional URL for proxying requests to vmalert. For example, if -vmalert.proxyURL=http://vmalert:8880 , then alerting API requests such as /api/v1/rules from Grafana will be proxied to http://vmalert:8880/api/v1/rules
Whether to use TLS when accepting connections from vminsert and vmselect. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Path to TLS CA file to use for verifying certificates provided by vminsert and vmselect if -cluster.tls flag is set. By default system CA is used. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Path to server-side TLS certificate file to use when accepting connections from vminsert and vmselect if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
-cluster.tlsCipherSuites array
Optional list of TLS cipher suites used for connections from vminsert and vmselect if -cluster.tls flag is set. See the list of supported cipher suites at https://pkg.go.dev/crypto/tls#pkg-constants
Supports an array of values separated by comma or specified via multiple flags.
Whether to skip verification of TLS certificates provided by vminsert and vmselect if -cluster.tls flag is set. Note that disabled TLS certificate verification breaks security
Path to server-side TLS key file to use when accepting connections from vminsert and vmselect if -cluster.tls flag is set. See https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#mtls-protection
Leave only the last sample in every time series per each discrete interval equal to -dedup.minScrapeInterval > 0. See https://docs.victoriametrics.com/#deduplication for details
Whether to deny queries outside of the configured -retentionPeriod. When set, then /api/v1/query_range would return '503 Service Unavailable' error for queries with 'from' value outside -retentionPeriod. This may be useful when multiple data sources with distinct retentions are hidden behind query-tee
Comma-separated downsampling periods in the format 'offset:period'. For example, '30d:10m' instructs to leave a single sample per 10 minutes for samples older than 30 days. See https://docs.victoriametrics.com/#downsampling for details
Supports an array of values separated by comma or specified via multiple flags.
Whether to enable reading flags from environment variables additionally to command line. Command line flag values have priority over values from environment vars. Flags are read only from command line if this flag isn't set. See https://docs.victoriametrics.com/#environment-variables for more details
The delay before starting final merge for per-month partition after no new data is ingested into it. Final merge may require additional disk IO and CPU resources. Final merge may increase query speed and reduce disk space usage in some cases. Zero value disables final merge
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 read data files bigger than 2^32 bytes in memory. mmap() is usually faster for reading small data chunks than pread()
Incoming http connections are closed after the configured timeout. This may help to spread the incoming load among a cluster of services behind a load balancer. Please note that the real timeout may be bigger by up to 10% as a protection against the thundering herd problem (default 2m0s)
The maximum duration for a graceful shutdown of the HTTP server. A highly loaded server may require increased value for a graceful shutdown (default 7s)
An optional prefix to add to all the paths handled by http server. For example, if '-http.pathPrefix=/foo/bar' is set, then all the http requests will be handled on '/foo/bar/*' paths. This may be useful for proxied requests. See https://www.robustperception.io/using-external-urls-and-proxies-with-prometheus
Optional delay before http server shutdown. During this delay, the server returns non-OK responses from /health page, so load balancers can route new requests to other servers
Whether to log new series. This option is for debug purposes only. It can lead to performance issues when big number of new series are ingested into VictoriaMetrics
Per-second limit on the number of ERROR messages. If more than the given number of errors are emitted per second, the remaining errors are suppressed. Zero values disable the rate limit
Timezone to use for timestamps in logs. Timezone must be a valid IANA Time Zone. For example: America/New_York, Europe/Berlin, Etc/GMT+3 or Local (default "UTC")
Per-second limit on the number of WARN messages. If more than the given number of warns are emitted per second, then the remaining warns are suppressed. Zero values disable the rate limit
Allowed size of system memory VictoriaMetrics caches may occupy. This option overrides -memory.allowedPercent if set to a non-zero value. Too low a value may increase the cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache resulting in higher disk IO usage
Supports the following optional suffixes for size values: KB, MB, GB, KiB, MiB, GiB (default 0)
Allowed percent of system memory VictoriaMetrics caches may occupy. See also -memory.allowedBytes. Too low a value may increase cache miss rate usually resulting in higher CPU and disk IO usage. Too high a value may evict too much data from OS page cache which will result in higher disk IO usage (default 60)
Data with timestamps outside the retentionPeriod is automatically deleted
The following optional suffixes are supported: h (hour), d (day), w (week), y (year). If suffix isn't set, then the duration is counted in months (default 1)
The offset for performing indexdb rotation. If set to 0, then the indexdb rotation is performed at 4am UTC time per each -retentionPeriod. If set to 2h, then the indexdb rotation is performed at 4am EET time (the timezone with +2h offset)
The maximum number of unique time series, which can be scanned during every query. This allows protecting against heavy queries, which select unexpectedly high number of series. Zero means 'no limit'. See also -search.max* command-line flags at vmselect
Automatically delete snapshots older than -snapshotsMaxAge if it is set to non-zero duration. Make sure that backup process has enough time to finish the backup before the corresponding snapshot is automatically deleted
The following optional suffixes are supported: h (hour), d (day), w (week), y (year). If suffix isn't set, then the duration is counted in months (default 0)
The maximum number of unique series can be added to the storage during the last 24 hours. Excess series are logged and dropped. This can be useful for limiting series churn rate. See also -storage.maxHourlySeries
The maximum number of unique series can be added to the storage during the last hour. Excess series are logged and dropped. This can be useful for limiting series cardinality. See also -storage.maxDailySeries
Path to file with TLS certificate if -tls is set. Prefer ECDSA certs instead of RSA certs as RSA certs are slower. The provided certificate file is automatically re-read every second, so it can be dynamically updated
Optional list of TLS cipher suites for incoming requests over HTTPS if -tls is set. See the list of supported cipher suites at https://pkg.go.dev/crypto/tls#pkg-constants
