When restarting an entire cluster all at once, verify that nodes that were already part of the cluster are restarted first. Once they are fully online, it is safe to boot any new nodes being added into the cluster. Booting these new nodes too early can lead to IP address conflicts, and the new nodes will fail to come online.

Verify the updated USB flash drives, configuration server, or CSN is prepared and begin the reboot process after the cluster is shut down.

This is the recommended power-on sequence:

1. Start a single node.

2. Verify the node boots properly. This process verifies that the software loads correctly and the node can communicate with the network.

3. Repeat on each remaining node in the cluster.

When initially booting a node, the hpStartDelay parameter provides a 15-minute delay window by default for all nodes to boot up and join the cluster before Swarm begins to check for missing cluster nodes. As long as all nodes in the cluster are running within this window, failed volume recovery (FVR) and erasure-coded recovery (ECR) operations can be avoided.

This window exists immediately following a node reboot. After 15 minutes, the recovery operations begin immediately after Swarm detects a missing node.

Restoring a Previous Version

The software can usually be reverted to a previous version using the same method performed during an upgrade if a legacy software version on any node needs to be installed.

Limitations

• Reverting to a previous version is not supported when upgrading to version 6.0 or later from a version prior to 6.x due to internal Swarm changes.

• The 6.5.1 release can be downgraded to any 6.x release except 6.5.0.