Details
-
Bug
-
Resolution: Fixed
-
Minor
-
None
-
Security Level: Public
-
None
-
2.0-1495
Description
Steve Yen: On "the initial cache miss looks weird", that weird time is correlated when compaction kicks in. Not sure why compacting means a higher cache miss rate, though.
Perhaps @Aaron Miller, @Chiyoung Seo, @Liang Guo might have a theory? Does this need a jira issue to find an answer?
Aaron Miller: Compaction shouldn't affect cache misses, unless the compactor was eating up RAM and causing ejections. I think we'd see that happening in one of these graphs though. Plus the compactor is designed to operate with a pretty small memory footprint. It's probably a red herring.
Dustin Sallings in reply to Aaron Miller: System RAM consumption shouldn't cause an increase in cache misses unless it's happening inside ep-engine now and the memory used is considering in accounting for evictions (that is, unless someone decided it should).
Steve Yen: @Aaron Miller @Dustin Sallings - hmmm, graphs were showing 0 vb_active_eject and 0 vb_replica_eject throughout. So... strange mystery.