Grafana Mimir operator and user guideRunning Grafana Mimir in productionScaling out Grafana Mimir

Scaling out Grafana Mimir

Grafana Mimir can horizontally scale every component. Scaling out Grafana Mimir means that to respond to increased load you, can increase the number of replicas of each Grafana Mimir component.

We have designed Grafana Mimir to scale up quickly, safely, and with no manual intervention. However, be careful when scaling down some of the stateful components as these action can result in writes and reads failures, or receiving partial query results.

Monolithic mode

When running Grafana Mimir in monolithic mode, you can safely scale up to any number of instances. To scale down the Grafana Mimir cluster, see Scaling down ingesters.

Microservices mode

When running Grafana Mimir in microservices mode, you can safely scale up any component. You can also safely scale down any stateless component.

The following stateful components have limitations when scaling down:

  • Alertmanagers
  • Ingesters
  • Store-gateways

Scaling down Alertmanagers

Scaling down Alertmanagers can result in downtime.

Consider the following guidelines when you scale down Alertmanagers:

  • Scale down no more than two Alertmanagers at the same time.
  • Ensure at least -alertmanager.sharding-ring.replication-factor Alertmanager instances are running (three when running Grafana Mimir with the default configuration).

Note: If you enabled zone-aware replication for Alertmanagers, you can, in parallel, scale down any number of Alertmanager instances within one zone at a time.

Scaling down ingesters

Ingesters store recently received samples in memory. When an ingester shuts down, because of a scale down operation, the samples stored in the ingester cannot be discarded in order to guarantee no data loss.

You might experience the following challenges when you scale down ingesters:

  • By default, when an ingester shuts down, the samples stored in the ingester are not uploaded to the long-term storage, which causes data loss.

    Ingesters expose an API endpoint /ingester/shutdown that flushes in-memory time series data from ingester to the long-term storage and unregisters the ingester from the ring.

    After the /ingester/shutdown API endpoint successfully returns, the ingester does not receive write or read requests, but the process will not exit.

    You can terminate the process by sending a SIGINT or SIGTERM signal after the shutdown endpoint returns.

    To mitigate this challenge, ensure that the ingester blocks are uploaded to the long-term storage before shutting down.

  • When you scale down ingesters, the querier might temporarily return partial results.

    The blocks an ingester uploads to the long-term storage are not immediately available for querying. It takes the queriers and store-gateways some time before a newly uploaded block is available for querying. If you scale down two or more ingesters in a short period of time, queries might return partial results.

Scaling down ingesters deployed in a single zone (default)

Complete the following steps to scale down ingesters deployed in a single zone.

  1. Configure the Grafana Mimir cluster to discover and query new uploaded blocks as quickly as possible.

    a. Configure queriers and rulers to always query the long-term storage and to disable ingesters shuffle sharding on the read path:

    -querier.query-store-after=0s
    -querier.shuffle-sharding-ingesters-lookback-period=87600h
    

    b. Configure the compactors to frequently update the bucket index:

    -compactor.cleanup-interval=5m
    

    c. Configure the store-gateways to frequently refresh the bucket index and to immediately load all blocks:

    -blocks-storage.bucket-store.sync-interval=5m
    -blocks-storage.bucket-store.ignore-blocks-within=0s
    

    d. Configure queriers, rulers and store-gateways with reduced TTLs for the metadata cache:

    -blocks-storage.bucket-store.metadata-cache.bucket-index-content-ttl=1m
    -blocks-storage.bucket-store.metadata-cache.tenants-list-ttl=1m
    -blocks-storage.bucket-store.metadata-cache.tenant-blocks-list-ttl=1m
    -blocks-storage.bucket-store.metadata-cache.metafile-doesnt-exist-ttl=1m
    
  2. Scale down one ingester at a time:

    a. Invoke the /ingester/shutdown API endpoint on the ingester to terminate.

    b. Wait until the API endpoint call has successfully returned and the ingester logged “finished flushing and shipping TSDB blocks”.

    c. Send a SIGINT or SIGTERM signal to the process of the ingester to terminate.

    d. Wait 10 minutes before proceeding with the next ingester. The temporarily configuration applied guarantees newly uploaded blocks are available for querying within 10 minutes.

  3. Wait until the originally configured -querier.query-store-after period of time has elapsed since when all ingesters have been shutdown.

  4. Revert the temporarily configuration changes done at the beginning of the scale down procedure.

Scaling down ingesters deployed in multiple zones

Grafana Mimir can tolerate a full-zone outage when you deploy ingesters in multiple zones. A scale down of ingesters in one zone can be seen as a partial-zone outage. To simplify the scale down process, you can leverage ingesters deployed in multiple zones.

For each zone, complete the following steps:

  1. Invoke the /ingester/shutdown API endpoint on all ingesters that you want to terminate.
  2. Wait until the API endpoint calls have successfully returned and the ingester has logged “finished flushing and shipping TSDB blocks”.
  3. Send a SIGINT or SIGTERM signal to the processes of the ingesters that you want to terminate.
  4. Wait until the blocks uploaded by terminated ingesters are available for querying before proceeding with the next zone.

The required amount of time to wait depends on your configuration and it’s the maximum value for the following settings:

  • The configured -querier.query-store-after
  • Two times the configured -blocks-storage.bucket-store.sync-interval
  • Two times the configured -compactor.cleanup-interval

Scaling down store-gateways

To guarantee no downtime when scaling down store-gateways, complete the following steps:

  1. Scale down no more than two store-gateways at the same time.
  2. Ensure at least -store-gateway.sharding-ring.replication-factor store-gateway instances are running (three when running Grafana Mimir with the default configuration).

Note: If you enabled zone-aware replication for store-gateways, you can in parallel scale down any number of store-gateway instances in one zone at a time.