Prometheus

Prometheus can now change its log level at runtime via the new runtime.log_level configuration field, scrape targets over Unix domain sockets, parse the OpenMetrics 2.0 format, and accept zstd-compressed scrape responses; TSDB also stabilizes XOR2 float chunk encoding and reduces head CPU use on initialization. The --log.level command-line flag is deprecated, so configurations must move the default log level setting into runtime.log_level, and anyone reading TSDB chunks directly (e.g. Thanos sidecar) must confirm XOR2 support before enabling it.

Matters for: Operators running large Prometheus fleets with sidecars like Thanos that read TSDB chunks directly, and teams using AWS service discovery for RDS or ElastiCache monitoring.

Prometheus 3.14 adds native service discovery for Oracle Cloud Infrastructure (oci_sd_configs), enables PromQL duration expressions by default, and promotes first_over_time to a stable function that no longer needs a feature flag. It also fixes multiple TSDB bugs that could cause silent data loss or corruption, including issues tied to stale_series_compaction_threshold, out-of-order ingestion during compaction, and native histogram data after restart.

Matters for: Relevant to teams running Prometheus against Oracle Cloud Infrastructure workloads and to operators relying on out-of-order ingestion or native histograms in production TSDB storage.