Prometheus v3.12.0 released on 28-05-2026
Prometheus v3.12.0 is out now. It delivers security fixes and request hardening for remote ingestion and endpoints, together with PromQL, TSDB, discovery and UI updates that are relevant for operators and those running remote-write or OTLP pipelines.
See the project’s GitHub release notes for full details, including credits to external reporters and the complete list of fixes, features and flags.
What’s in this release
- Security and request hardening: rejects snappy-compressed remote-write requests whose declared decoded length exceeds 32MB; limits decompressed body size for gzip-encoded OTLP write requests; fixes plaintext secret exposure in STACKIT service discovery (GHSA-39j6-789q-qxvh); and rejects concurrent fgprof profiles.
- PromQL, API and tooling: new experimental functions start(), end(), range(), step(); warnings when sort/sort_by_label variants are used inside range (matrix) queries; start-timestamp-aware resets()/rate()/irate()/increase() behind the use-start-timestamps feature flag; /api/v1/status/self_metrics endpoint; and promtool query-instant –header added.
- TSDB, agent and performance: start timestamp field for in-memory WAL histogram samples when st-storage is enabled; agent.DB CheckpointFromInMemorySeries and an st-synthesis flag for synthesising start timestamps; major TSDB performance work including constant-time head chunk lookup and mmap optimisations; plus multiple correctness fixes for WAL, histograms and retention handling.
Upgrade notes
- Feature flags and compatibility: the start-timestamp behaviour is behind the use-start-timestamps flag and does not work with extended range selectors such as anchored and smoothed. Test rate/irate/increase changes in staging before enabling the flag in production.
- Rollback: if you encounter issues with the new request limits or feature-flag behaviour, revert to your previous release and follow your standard rollback procedures while you investigate.
Share your experience after upgrading — report issues on the project’s issue tracker or discuss behaviour and edge cases with the community.
