Grafana Mimir vs Keboola

Almost every entry in this window is an identical bot-generated Helm chart release — byte-for-byte the same body across seven tags. The one substantive release is the 3.2.0 RC, whose notes run past the feed's truncation point: continued Mimir Query Engine optimisation, native histogram support in more operators, per-tenant cost attribution trackers, and a tenant-fair compute worker pool on the ingester. Buried in its changelog are changes operators must act on — the embedded per-tenant Alertmanager web UI is gone, credential-bearing headers are now rejected at startup, and querier/store-gateway traffic requires coming from 3.1.

Mimir is deepening the query path rather than widening the product: most of 3.2 is MQE work, subquery spin-off, caching and elimination passes, largely behind experimental flags. The operational surface is growing in parallel — Kafka producer controls, mimirtool and kafkatool subcommands, SigV4 for shared API clients. The chart is versioned and released independently on a weekly bot cadence, which is why the feed reads as empty even in an active month.

The experimental MQE flags that landed in 3.2 are the ones to watch for promotion to default in 3.3, particularly subquery spin-off and scalar common subexpression elimination. Whether a 3.2.0 final follows this RC is not visible in the feed — only the RC has been published here.