Since NXP-19721 there is a performance regression because the lock is held during the drain method, which add latency to post commit threads that submit log entries.
In case of slow audit backend, the result is that postcommit threads are waiting and can not be reused, so it creates lots of new threads consuming resources that slow down the system.
The lock should be release before draining and acquire when sending signal.
- is related to
-
NXP-19721 Fix Elasticsearch random bug on TestAuditWithElasticSearch.shouldSupportMultiCriteriaQueries
- Resolved
-
NXP-19472 process core events in audit inlined
- Resolved
-
NXP-25136 Fix performance regression on audit bulker on 8.10
- Resolved
-
NXP-22762 Investigate about performance regression seen on 17w18
- Resolved