Prometheus v3.11.0 released on 02-04-2026
Prometheus v3.11.0 is out now. It delivers a major set of TSDB improvements and experimental storage controls to cap disk usage and speed restart recovery for operators running Prometheus at scale.
See the GitHub release notes for full details, changelogs and download links.
What’s in this release
- TSDB: new storage.tsdb.retention.percentage setting to cap TSDB disk usage by percentage.
- TSDB: experimental fast-startup mode that writes series_state.json to the WAL to accelerate restart recovery.
- TSDB: experimental st-storage to persist ingested start timestamps and expose them via Remote Write v2.
- TSDB: experimental xor2-encoding, a new float chunk encoding optimised for scraped data and start-timestamp encoding; old temporary checkpoints are removed when creating checkpoints.
- Service discovery and cloud-provider changes: Hetzner label deprecations and replacements (hcloud labels stop working after 1 July 2026), new AWS ElastiCache and RDS roles, Azure Workload Identity support, Vultr client library upgrade, and Kubernetes SD improvements including node role selectors and new pod-based controller labels.
- PromQL and UI: new native-histogram trimming operators (/), experimental variadic histogram_quantiles, PromQL performance work, a new discovery metric prometheus_sd_last_update_timestamp_seconds, and UI autocomplete and tooltip fixes.
- Reliability and bug fixes across remote write, OTLP, the Agent, alerting/rules, scraping, tracing and TSDB (including missing mmap of histogram chunks during WAL replay, retention time unit fix, buffer-pool memory-leak fixes and various robustness improvements).
Upgrade notes
- Hetzner SD deprecations: replace
__meta_hetzner_datacenterwith__meta_hetzner_robot_datacenterfor the robot role, and replace__meta_hetzner_hcloud_datacenter_location/__meta_hetzner_hcloud_datacenter_location_network_zonewith__meta_hetzner_hcloud_location/__meta_hetzner_hcloud_location_network_zone. Note the hcloud labels will stop working after 1 July 2026 — update relabel rules before then. - If you rely on the
use-uncached-iofeature flag, check platform support first: Prometheus now fails early on unsupported environments. - Rollback: if a revert is needed, obtain the previous release binary from the Prometheus GitHub releases page and restore your prior configuration and data as appropriate.
Share any notes on behaviour, performance or upgrade issues so others can benefit — report problems or experiences via the Prometheus GitHub issue tracker or community channels.