Release notes for version 0.51.0

Release date: 12 Oct 2026

• bump version of VM components to v1.154.0
• fix alertmanager.persistentVolume.volumeAttributesClassName being ignored because of a typo in the template.
• bump victoria-metrics-common dependency to 0.4.*. Resource names are unchanged.
• fix Alertmanager HA mode (alertmanager.mode: statefulSet with alertmanager.replicaCount > 1): replicas didn't form a cluster and vmalert couldn't reach them, because the per-pod addresses didn't resolve. Peers and vmalert notifiers now use the per-pod DNS names of the StatefulSet governing Service. The unused -headless Service is removed. The upgrade is in place: the StatefulSet serviceName is unchanged.
• rename Alertmanager gossip container ports from cluster-tcp/cluster-udp to mesh-tcp/mesh-udp, matching the VictoriaMetrics operator. They are only rendered in HA mode.
• store the generated Alertmanager config in a Secret instead of a ConfigMap, as it usually contains credentials. The name is unchanged (<alertmanager fullname>-config). Alertmanager pods are restarted once on upgrade. Added alertmanager.configSecret to use an existing Secret. alertmanager.configMap is deprecated in favour of it and still works for an existing ConfigMap.
• fix alertmanager.webConfig being ignored: it is now stored in the generated config as webconfig.yaml and passed via --web.config.file, and probes use HTTPS when it enables TLS.
• honour web.listen-address and web.route-prefix set via alertmanager.extraArgs in the Alertmanager container port and in the vmalert notifier URLs.
• rename config volumes to match the VictoriaMetrics operator: alerts-config to rules-out in vmalert and config to config-out in Alertmanager. Pods are restarted once on upgrade.
• expose Alertmanager gossip ports (tcp-mesh, udp-mesh) on its Service in HA mode.
• fix alertmanager.persistentVolume.name in statefulSet mode: the container mounted the default volume name instead of the overridden claim template name.