David Capwell created CASSANDRA-18601:
-----------------------------------------
Summary: Repair vtable uses number of records rather than memory
limits, which can cause more memory allocated than expected
Key: CASSANDRA-18601
URL: https://issues.apache.org/jira/browse/CASSANDRA-18601
Project: Cassandra
Issue Type: Improvement
Components: Consistency/Repair
Reporter: David Capwell
The config “repair_state_size” is used by 2 caches: the coordinator side of
repair, and the participate side of repair. At this point in time we use
number of records rather than memory side, and the default is 100k… In one case
it was found that there were 95k records of participate history causing 550mb
of allocation!
To resolve this we should:
1) migrate to memory size
2) have the 2 caches have 2 different configs (maybe you want more history for
one than the other?); this also makes things explicit and doesn’t have any
confusion in terms of cost (atm you can have 200k records for repair with a
100k limit)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
