Description of problem: The Replication Worker spawns a subprocess (evm:dbsync:replicate) that alone can exceed the memory threshold that is permitted by default for the Replication Worker (200MiB by Default). This process is currently not counted against the Replication Worker Memory threshold. This can lead administrators to believe a Replication Workload should only consume an amount of memory below the Replication Worker Threshold. For 5.5.2.0 during stress tests that have lasted around 4 hours. I have witnessed the following Memory usages: MiqReplicationWorker - 175-176MiB RSS, 168-170MiB PSS evm:dbsync:replicate - 206-211MiB RSS, 199-204MiB PSS For a combined Replication Workload using 381-388MiB RSS When viewing the processes PSS (Proportional Set Size): MiqReplicationWorker - 168-170MiB PSS evm:dbsync:replicate - 199-204MiB PSS Version-Release number of selected component (if applicable): 5.5.2.0 How reproducible: Exposed in all tests against 5.5.2.0 Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
I'm converting this to a documentation issue to ensure the need to account for approximately 200Mb of additional memory (beyond the amount configured for the MiqReplicationWorker) is accounted for in during configuration planning. Another possible action would be to include a one line note with the Replication Worker configuration text indicating an additional 200Mb of system memory will be consumed by replication - beyond the configurable worker amount.
Assigning to Suyog for review.