I have executed "./nodetool repair -pr -local -full cursor1" n times. So I believe there is no discrepancy when I executed the same repair command (n+1)th time.
But the log still shows that sstables compaction is triggered. May I know why nodetool repair triggers sstables compaction if there is no discrepancy?
@376752150_179413 Repairs don't trigger the type of compaction that you are probably thinking of.
When running a repair, the first phase is to build a Merkle tree of the data to be repaired (leaves of a Merkle tree are hashes of key values). Cassandra compares the hash values between replicas to determine which parts of the tree (replica data) to repair. To perform the Merkle tree calculation, Cassandra does a validation compaction -- a special kind of compaction.
To answer your question, parallel repairs do validation compaction in order to determine whether replicas are out of sync, and which parts of the data it needs to synchronise. For more information, see Repairing nodes particularly the section on Manual anti-entropy repair. Cheers!
5 People are following this question.
