Mimir blocks retention

[Ferrany1]

Hello, unfortunately I didn't manage to find a way to enable retention for blocks, currently I'm running everything on a low disk space machine and until stalling want to enable retention for example older than 3 days.

I've tried to use, but unfortunately it didn't help since as I understand my blocks yet doesn't escape to compactor (±7day series collection)

limits:
  compactor_blocks_retention_period: 72h

mimir.yaml

target: all,alertmanager,overrides-exporter,ruler

multitenancy_enabled: false
no_auth_tenant: "anonymous"

memberlist:
 join_members: [ mimir-1, mimir-2 ]

blocks_storage:
 backend: filesystem
 bucket_store:
   sync_dir: /mimir/tsdb-sync
 filesystem:
   dir: /mimir/data/tsdb
 tsdb:
   dir: /mimir/tsdb

limits:
 compactor_blocks_retention_period: 72h

server:
 log_level: warn
 http_listen_port: 8080

[pstibrany]

Hello,

option compactor_blocks_retention_period: 72h is used to delete old blocks from long-term storage.

You mention running on "low disk space" machine. Mimir in monolithic mode (-target=all) needs disk space for:

1. storing WAL and blocks for each tenant.
2. downloading block indexes by store-gateway component
3. doing compactions -- this requires extra disk space to download blocks and generate compacted blocks
4. storing alertmanager and ruler config, but these are typically small

Re 1: Blocks are kept on disk for -blocks-storage.tsdb.retention-period. Note that changing this requires changes to some other parameters like -querier.query-store-after, -querier.query-ingesters-within and -blocks-storage.bucket-store.ignore-blocks-within.

Re 2, 3, 4: if you could run other components on different machines, that would help to control disk usage. But Mimir would no longer be running in monolithic mode then.

Unfortunately that may prove difficult to optimize Mimir's disk usage in monolithic mode on small disk.