Prometheus v3.15.0 released on 25-09-2026
Prometheus v3.15.0 is out now. Operators and query authors will see a collection of PromQL correctness fixes and tracing additions, plus TSDB, scraping and service-discovery improvements that target stability and performance.
See the GitHub release notes for full details and upgrade instructions.
What’s in this release
- PromQL: Fixes for range-subquery evaluation that prevented subqueries reading past a parent’s final step (which could inflate peakSamples), avoidance of unnecessary start-timestamp resets between identical consecutive samples, and a raft of edge-case fixes (info()/enrichment, matching, regex, anchored/smoothed range-selector and subquery issues). Added more tracing spans for queries and rule evaluations.
- TSDB and storage: Stabilised XOR2 float chunk encoding (the old –enable-feature=xor2-encoding flag is deprecated and replaced by storage.tsdb.chunk_encoding.floats: xor2), auto-enablement of XOR2 + ST-capable histogram encoding when st-storage is enabled, a fast path for XOR chunk decompression, new TSDB metrics, and fixes for chunk ID overflow/wrapping plus several GC/WAL/eviction corner cases.
- Scraping and discovery: Support for Unix Domain Socket targets, OM2.0 scrape format and optional zstd-compressed scrape responses (behind feature flag zstd-scrape), scrape appender v2 improvements (start-time synthesis for summary _count/_sum) and faster shutdown by stopping pools in parallel. Docker SD gains extra image labels and many AWS/IONOS/Kubernetes discovery robustness and performance fixes.
Upgrade notes
- The –log.level flag is deprecated; use runtime.log_level in configuration (and you can change process log level on configuration reload via runtime.log_level).
- XOR2 float chunk encoding is now stable and the old feature flag is deprecated—switch to storage.tsdb.chunk_encoding.floats: xor2 only after confirming any software that reads the TSDB directly (for example Thanos sidecar) supports XOR2. The Agent now ignores unknown WAL record types to help users rolling back when appropriate.
Share your experience or any issues on the project’s GitHub issue tracker so others can benefit from what you saw.
