[CASSANDRA-6455] Improve concurrency of repair process

• Type: Improvement
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 2.2.0 beta 1
• Component/s: Legacy/Streaming and Messaging
• Labels: repair
• Assignee: Yuki Morishita
• Reporter: Yuki Morishita

Authors

Yuki Morishita

Reviewers

Marcus Eriksson

• Created: 05/Dec/13 18:55
• Updated: 16/Apr/19 09:31
• Resolved: 09/Oct/14 23:13

Description

Currently, most of the repair tasks (taking snapshots, send/receiving merkle tree, compute MT difference, etc) are done on single threaded AntiEntropyStage.

This causes a problem like CASSANDRA-6415 and likely to cause unnecessary wait.

Also, repair is done one CF at the time. I think we can parallelize this(concurrency is configurable by a user based on # of CF and load of the nodes) for faster processing.

Issue Links

• is duplicated by: CASSANDRA-8177 sequential repair is much more expensive than parallel repair
• is related to: CASSANDRA-5220 Repair improvements when using vnodes
• is related to: CASSANDRA-6566 Differencer should not run in AntiEntropy Stage
• is required by: CASSANDRA-6616 Nodetool repair is taking a long time.
• is required by: CASSANDRA-6516 Force shutdown of all repair sessions sometimes fails