I'm running Ambari 2.1.2.1, which includes the fix for AMBARI-13307 / Agent instance memory footprint likely gradually growing.
However, I noticed today (after an agent crashed) that the agent had been consuming 37GB of memory, and I've got an agent on another machine consuming 15GB of memory after it was restarted 11/13. (And a couple others running in the ~8-10GB range). The rest of the agents are only consuming ~1.6GB of memory - which seems to be machines that only have datanodes/nodemanagers. So whatever the issue is, it seems to only be affecting a small number of agents. The symptoms here are very similar to the leak that appeared in/around 2.1.x that I (thought) would've been fixed by AMBARI-13307. Checking logs, it appears the rate of memory increase is approximately the same as before the 2.1.2.1 upgrade, so 13307 doesn't seem to have fixed this particular issue. If you need more information about this just let me know. Thank you!
