[CASSANDRA-19517] Raise priority of TCM internode messages during critical operations

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 5.1-alpha1
• Component/s: Transactional Cluster Metadata
• Change Category: Operability
• Complexity: Normal
• Platform: All

Source Control Link

https://github.com/apache/cassandra/commit/67139d5c334e58fdc8d9f09f9288155448666cd3

Test and Documentation Plan

Includes tests. Additional stress testing will be done during release qualification.

• Assignee: Alex Petrov
• Reporter: Alex Petrov

Authors

Alex Petrov

Reviewers

Marcus Eriksson, Sam Tunnicliffe

• Created: 03/Apr/24 16:14
• Updated: 06/May/24 07:13
• Resolved: 06/May/24 07:13

Description

In a busy cluster, TCM messages may not get propagated throughout the cluster, since they will be ordered together with other P1 messages (for TCM_ prefixed verbs), and with P2 with all Paxos operations.

To avoid this, and make sure we can continue cluster metadata changes, all TCM_-prefixed verbs should have P0 priority, just like Gossip messages used to. All Paxos messages that involve distributed metadata keyspace should now get an URGENT flag, which will instruct internode messaging to schedule them on the URGENT_MESSAGES connection.

Issue Links

• links to: Trunk PR