etcd Patch Releases: v3.7.4, v3.6.17, v3.5.36

SIG-etcd has published patch releases for all three supported versions of etcd v3.7.4, v3.6.17 and v3.5.36. These updates improve recovery behavior and concurrency safety, and include additional version-specific fixes. Users should update at the next scheduled maintenance window after the releases become available.

Updates may be found here:

Official container images are available from gcr.io.

Recovery and concurrency fixes

All three releases fix an issue where using --force-new-cluster could permanently prevent a member from starting when the backend’s consistent index was ahead of the last index in its write-ahead log (WAL). The fix prevents the adjusted commit index from exceeding the WAL’s last index.

The releases also define alarmStore as an atomic.Pointer, fixing a race condition when accessing the active alarm store.

Version-specific fixes

v3.7.4 fixes a startup panic that could occur while replaying legacy v2 member-attribute entries from the WAL.

v3.5.36 fixes an issue where a leader that removed itself from the cluster could answer a ReadIndex request with a stale index.

Dependency updates

All three releases are built with Go 1.26.11. Official container images now use Debian 13.7, which distrusts eMudhra certificates.

v3.7.4 and v3.6.17 update golang.org/x/net to v0.62.0, addressing CVE-2026-78661 and CVE-2026-97045.

Additionally, v3.6.17 also updates github.com/klauspost/compress to v1.20.2 to address CVE-2026-116591.

For the full list of changes, see the changelogs: