# 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:

```bash
kubectl -n scylla patch scyllacluster/scylla --type=merge \
  -p='{"spec": {"forceRedeploymentReason": "restart-2024-01-15"}}'
```

Wait for the rolling restart to complete:

```bash
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 `forceRedeploymentReason` must be different from the current value to trigger a restart. Repeating the same string has no effect.                   |
| 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 (`maxUnavailable: 1`) ensures that at most one node per datacenter is unavailable at any given time, even during a rolling restart.                      |

## Related pages

- [StatefulSets and racks](https://operator.docs.scylladb.com/v1.22/understand/statefulsets-and-racks.md) — how StatefulSets manage rolling updates with partition-based rollouts
- [Upgrade ScyllaDB](https://operator.docs.scylladb.com/v1.22/upgrade/upgrade-scylladb.md) — version upgrades that include an automatic rolling restart