-
Type: Improvement
-
Resolution: Won't Fix
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
Atlas Streams
When running a jstest which repeatedly starts and stops a stream processor (https://github.com/10gen/mongo-enterprise-modules/blob/03031668b9efbfc90fa54a2d031d7a629c2a2fe2/jstests/streams/sp_repeated_starts.js), a slow increase is observed in the memory usage of the mongod process (the RES col in top)
When trying an asan build, the problem is no longer observed. So this looks like a tcmalloc issue, though, the rate of growth is very small and so this may not be a real problem.
Using the following tcmalloc settings does not help either:
- Call MallocExtension::instance()->ReleaseToSystem() periodically to force tcmalloc to release memory sooner than it normally would have
- Run with the kernel settings suggested by tcmalloc: https://google.github.io/tcmalloc/tuning.html#system-level-optimizations
See attached graphs for the memory usage for asan build vs a build with ^ tcmalloc settings.
Additionally, I also tried a build where the checkpoint-at-stop feature was turned off but it showed the same behavior