[CASSANDRA-12344] Forward writes to replacement node with same address during replace

• Type: Improvement
• Status: Open
• Priority: Normal
• Resolution: Unresolved
• Fix Version/s: 5.x
• Component/s: Cluster/Membership, Consistency/Bootstrap and Decommission, Legacy/Distributed Metadata
• Assignee: Paulo Motta
• Reporter: Paulo Motta

Authors

Paulo Motta

• Created: 29/Jul/16 19:47
• Updated: 07/Mar/23 10:54

Description

On CASSANDRA-8523 it was added support to forwarding writes to a replacement node via a new gossip state BOOTSTRAPPING_REPLACE.

Currently this is limited to replacement nodes with a different address of the original node, because if a replacement node with the same address of a normal endpoint joins gossip with a non-dead state, it will become alive in the Failure Detector and reads will be forwarded to it before the node is ready to serve reads.

This ticket is to add support to forwarding writes to replacement nodes with the same IP address as the original node.

The initial idea is to allow marking a node as unavailable for reads on TokenMetadata, what will allow the replacement node with the same IP join gossip without having reads forwarded to it. This will be enabled by CASSANDRA-11559.

Issue Links

• is related to: CASSANDRA-8523 Writes should be sent to a replacement node which has a new IP while it is streaming in data
• is related to: CASSANDRA-14084 Disks can be imbalanced during replace of same address when using JBOD
• requires: CASSANDRA-11559 Enhance node representation