#Pause and resume

The Operator pauses an instance when spec.podAnnotations["buf.red/pause-timestamp"] is non-empty. Pausing reduces managed workload replicas to zero and makes the instance unavailable.

#Safety requirements

• Stop clients and writes before pausing.
• Use persistent storage if data must survive Pod deletion.
• A pause is not a backup. Complete external data protection first.
• Do not pause during scaling, rebalancing, upgrade, certificate rotation, or another rolling update.

#Pause

PAUSED_AT="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
kubectl -n default patch valkey valkey-cluster --type=merge \
  -p "{\"spec\":{\"podAnnotations\":{\"buf.red/pause-timestamp\":\"${PAUSED_AT}\"}}}"
kubectl -n default get valkey valkey-cluster -w

The expected phase is Paused. Confirm that the managed Pods have terminated:

kubectl -n default get pods -l buf.red/name=valkey-cluster
kubectl -n default get pvc -l buf.red/name=valkey-cluster

#Resume

Remove the map entry with a JSON merge patch:

kubectl -n default patch valkey valkey-cluster --type=merge \
  -p '{"spec":{"podAnnotations":{"buf.red/pause-timestamp":null}}}'
kubectl -n default get valkey valkey-cluster -w

Wait for Ready, verify all expected members and Cluster slots or the current primary, and then reconnect applications.