What prometheus/prometheus shipped

Prometheus 3.16.0 is out: search batch_size clamped to 1000, https external-url default under TLS web config, subquery samples counted in query stats. Also this week: promtool tsdb analyze reports disk usage per metric name, and scrape caches share metric metadata strings.

Prometheus 3.16.0 is released. The experimental search endpoints now clamp batch_size to 1000, --web.external-url defaults to https when the web config file enables TLS, and subquery samples are included in query TotalSamples and TotalSamplesPerStep statistics. Other changes this week: promtool tsdb analyze with the extended flag now breaks down disk usage by metric name, the scrape cache interns metric metadata strings shared across targets, and scrape loops acquire the storage appender only after reading the HTTP response. On the TSDB side, samples dropped at Commit for duplicate timestamps are now counted, and the agent no longer hangs on shutdown when a truncate tick races with Close.

What shipped this week: PromQL bitwise operators (and, or, xor), unquoted UTF-8 names on by default, an experimental stateless remote write mode, and fixes for queries over evicted head series, relabel-aware target metadata, and zero-bucket histogram rendering.

This week in the project: PromQL adds the bitwise operators and, or and xor, and unquoted UTF-8 metric names are now enabled by default. Remote write gains an experimental mode field for stateless and stateless-st sending. Fixes landed for queries that read series evicted from the head, for target metadata after metric relabeling, and for remote write type and unit labels falling back to WAL metadata. Linode discovery now converts transfer allowances from GiB correctly, the query UI renders histograms with only a zero bucket without errors, and Windows file discovery no longer deadlocks during refresh.

Prometheus 3.15.1 and 3.13.5 are out with updated CA certificates and Go security fixes. Other fixes this week: discovery refresh metrics surviving reloads, remote write config reloads that keep queues consistent, and clearer query cost timeout errors.

Prometheus 3.15.1 and 3.13.5 are released. Both update container CA certificates to distrust the emSign ECC Root CA - G3 and Comodo AAA Services root after the reported compromise of eMudhra intermediate CA private keys, and ship builds that fix vulnerabilities in net/http, net/textproto, crypto/tls and html/template. Also this week: discovery refresh metrics are kept for providers still running after a reload, remote write defers config changes until every config builds so a failed reload no longer leaves queues unmanaged, the tsdb agent clears pending data when Commit or Rollback fails, and query cost timeouts are reported correctly instead of as bad data.

Prometheus updates this week: the remote write receiver moves to AppenderV2, query cost limits are now enforced more strictly, histogram comparisons no longer skip buckets, duplicate exemplar label names are rejected, OTLP exemplars are sorted by timestamp, and promtool can creat

Changes in Prometheus this week: The remote write receiver now writes through AppenderV2 and ExemplarAppenderV2. Query cost enforcement and estimation are bounded so range selectors can't buffer storage input before sample-read limits apply, and duration limits cover construction, queueing and completion. PromQL fixes stop ==,!=, changes() and count_values() from skipping histogram buckets, and group_right now keeps fill sides correct. On exemplars: OTLP exemplars are sorted by timestamp before appending, the out-of-order counter again counts only out-of-order exemplars, and exemplars with duplicate label names are rejected. promtool can create blocks with --allow-incompatible-block-duration, and type and unit are dropped consistently with name.

This week: exemplars preserved across compaction and restart, future-dated exemplars no longer collapse remote-write shards, alert restoration fixed for short for durations, and q

This week in Prometheus: exemplars are preserved across compaction and restart, so a series is no longer dropped while it still has a recent exemplar. Future-dated exemplars no longer advance sent timestamps in a way that pushed remote-write shards down to one. Rules whose for duration is shorter than the alert for grace period now restore previously firing alerts instead of skipping them. Remote read gains a gate wait duration histogram. BucketQuantile and BucketFraction return NaN for empty input instead of panicking, printing a partial AST after a failed parse no longer panics, CopyToSchema keeps counter reset hints, metadata enrichment handles double _total and _info suffixes, and stripDomain no longer truncates IPv6 zone identifiers.

V3.13.4 released: HTTP/2 security fixes, agent WAL improvements, head chunk isolation fixes, and OpenMetrics 2.0 performance optimizations.

Prometheus v3.13.4 is out with important security updates for HTTP/2 handling and stability improvements. This release fixes WAL corruption tracking, preserves series records during checkpoints, enforces query isolation on head chunks, and optimizes OpenMetrics 2.0 parsing. Security patches address gRPC vulnerabilities and HTTP/2 ALPN negotiation issues under recent Go versions.

Prometheus 3.15.0 released. New integral() function in PromQL, fixes for range query evaluation, WAL handling, and remote storage data races. Custom PromQL functions now supported via parser options.

Prometheus 3.15.0 is now available. This release adds the integral() function to PromQL for calculating range vector integrals, fixes critical issues in range query evaluation that were wasting storage I/O, improves WAL persistence and replay, and resolves a data race in remote storage during config reloads. The parser now supports custom PromQL functions through parser options, enabling better extensibility.

Prometheus 3.15.0-rc.0 is available. Extended range selectors enabled by default. Range queries now evaluate correctly without inflating sample counts. Multiple performance and reliability improvements included.

Prometheus 3.15.0-rc.0 release candidate is now available. Key improvements include extended range selectors enabled by default in PromQL, corrected range query evaluation to prevent unnecessary sample reading, and performance optimizations in scraping and staleness tracking. Several bug fixes address histogram conversion, WAL handling, and API result ordering.