Was this page helpful?
Perform a rolling restart¶
Force a rolling restart of all ScyllaDB nodes in a cluster by changing the forceRedeploymentReason field.
How it works¶
The Operator places forceRedeploymentReason as an annotation on the StatefulSet pod template.
When you change the value, the pod template changes, which triggers the StatefulSet controller to perform a rolling update.
The Operator orchestrates this update one pod at a time, draining each node before terminating it and waiting for the replacement to become ready before proceeding to the next.
The value itself is arbitrary — it only needs to be different from the previous value.
Using a descriptive string (e.g. "config change 2024-01-15") makes it easier to identify why a restart was triggered.
Restart a ScyllaCluster¶
Patch the spec.forceRedeploymentReason field:
kubectl -n scylla patch scyllacluster/scylla --type=merge \
-p='{"spec": {"forceRedeploymentReason": "restart-2024-01-15"}}'
Wait for the rolling restart to complete:
kubectl -n scylla wait --timeout=30m --for='condition=Progressing=False' scyllaclusters.scylla.scylladb.com/scylla
kubectl -n scylla wait --timeout=30m --for='condition=Degraded=False' scyllaclusters.scylla.scylladb.com/scylla
kubectl -n scylla wait --timeout=30m --for='condition=Available=True' scyllaclusters.scylla.scylladb.com/scylla
Note
In multi-DC clusters using multiple ScyllaCluster resources, restart each datacenter’s ScyllaCluster independently.
Key considerations¶
Consideration |
Detail |
|---|---|
One at a time |
The Operator restarts one node at a time per rack. Each node is drained before termination, and the replacement must be ready before the next node is restarted. |
Unique value |
The value of |
Impact on availability |
A rolling restart temporarily reduces the number of available replicas by one. Ensure your replication factor allows for one node to be unavailable. |
PodDisruptionBudget |
The PDB ( |