I have seen interesting behavior on this topic recently:
Should this trendline ever go above the redline?
The max memory was set with -m 10000 which is ~2gb below the pod resource request of 12000. The JVM restarted 42 times, about every 10 minutes, so we took some corrective actions:
- Increased pod limits to be more than the requests, from 12000 memory to 16000 memory, and 2 CPU to 8 CPU.
- Updated ignition.conf to have init-memory and max-memory limits to be at 10000 to match
-m 10000configuration.
As a background, I have restored a .gwbk for initial configuration from another gateway, which had a setting of 56000 max memory in the file. Since the change where ignition.conf setting matched -m, it only went above the redline once, (I think thanks to max pod limit increase to 16000 the wrapper didn't send a SIGKILL (9)?) and has been stable since then.
