Upgrades¶
Upgrading Wazuh is a single field change. The operator turns it into an ordered, rollout-aware upgrade rather than swapping all images at once.
How to upgrade¶
Watch it progress:
What the operator does¶
- The new version flows into every component image
(
<registry>/wazuh-<component>:<version>). - The indexer is updated first. The reconcile waits for the StatefulSet to roll out
at the new revision — it checks
observedGeneration,updatedReplicas, andcurrentRevision == updateRevision, not justreadyReplicas. Old-revision pods that are still Ready do not let the upgrade race ahead. - Then the manager master and workers roll (one pod at a time), then the dashboard.
status.observedVersiononly advances once every component is on the new version, so theVERSIONcolumn always reflects what is actually running.
This indexer → manager → dashboard ordering matches Wazuh's recommended upgrade direction and keeps the indexer ahead of the components that read from and write to it.
Pods keep their PVCs across the rollout, so /var/ossec/etc, logs, API configuration
and indexer data are preserved — the same guarantee as Wazuh's manual Kubernetes
upgrade.
Config changes also roll pods¶
The Wazuh images read configuration only at startup. The operator stamps a
wazuh.bnerd.com/config-checksum annotation (a hash of the rendered config) onto each
pod template, so any config change — including replica-count changes that alter the
indexer's cluster.initial_master_nodes — triggers a rolling restart automatically.
The checksum is stable for unchanged config, so steady-state reconciles cause no churn.
Recommended checklist¶
- Snapshot the indexer (manual) and confirm
status.phase: Ready. - Bump
spec.version; watch the cluster return toReady. - Verify indexer cluster health and that agents reconnect via the dashboard.
- Upgrade agents separately, after the manager is on the new version.
Limitations¶
- Downgrades are not supported (Wazuh/OpenSearch do not support them).
- Cross-major migrations that require a reindex are not orchestrated; plan them manually. For new deployments the operator always issues 4.8+-style certificates, so the pre-4.8.0 certificate-recreation step does not apply.
- No automated pre-upgrade backup yet — take an indexer snapshot before major upgrades.
See Day-2 Operations for scaling and other limitations.