[CASSANDRA-8274] Node fails to rejoin cluster on EC2 if private IP is changed

• Type: Bug
• Status: Open
• Priority: Low
• Resolution: Unresolved
• Fix Version/s: 3.11.x
• Component/s: Legacy/Distributed Metadata

Environment

Amazon EC2

• Severity: Low
• Since Version: 1.2.16
• Assignee: Unassigned
• Reporter: Joseph Clark
• Created: 06/Nov/14 23:42
• Updated: 16/Apr/19 09:31

Description

Nodes in Amazon AWS EC2 Classic (not a VPC) may be assigned a new private IP if the node is stopped and then started again. In this case we have puppet update the configured listen_address to the new private IP. However, once the cassandra service starts, it is unable to communicate with the existing nodes(single region) and vice versa.

'nodetool status' shows that each node believes that it is 'UN' and the other node is 'DN'.

'nodetool gossipinfo' on the node that remained running shows the old private IP listed as the 'INTERNAL_IP' of the node that was stopped and restarted.

The situation is resolved by restarting the cassandra service on the node that remained running. Once it has restarted, the INTERNAL_IP is correctly updated to the new private IP. 'nodetool status' shows that both nodes are up and the cluster appears to function normally.

This appears to me to be the root cause of https://issues.apache.org/jira/browse/CASSANDRA-7292. Possibly https://issues.apache.org/jira/browse/CASSANDRA-8072 as well, but I am not convinced they are actually duplicates.