Rolling Restart

A rolling restart is the technique of restarting cluster nodes one at a time, keeping the remaining nodes running so the service stays available to clients throughout the operation. The cluster must maintain quorum at every step to continue handling writes and electing a primary.

How it works #

On a 3-node InnoDB Cluster, the typical sequence is:

1. Identify the node to restart (a secondary is the safest starting point).
2. Confirm the cluster is ONLINE and all members are in sync.
3. Restart the node (systemctl restart mysqld or equivalent).
4. Wait for the node to rejoin the group before moving to the next one.

# Check cluster status before each step
mysqlsh -- cluster status

The restarted node reconnects to the group via Group Replication and catches up on missed transactions through distributed recovery. Only after its status returns to ONLINE should the next node be restarted.

When to use it #

A rolling restart is the standard procedure whenever MySQL configuration parameters in my.cnf require a process restart — for example changes to innodb_buffer_pool_size, innodb_log_file_size, or Group Replication settings. It is also the preferred approach for applying OS patches or minor version upgrades without scheduling a formal maintenance window.

The main constraint is time: if a restarted node has a significant transaction backlog to recover, the distributed recovery phase can take several minutes, stretching the total operational window.