Too low resource requirements and too high heap memory causes processes to go OOM

[mvanniekerkHartwig]

Description of the bug

When testing out oncoanalyser against some more difficult samples I ran into quite some erratic behaviour of stages running out of memory.
In general this was caused by the java process allocating 95% of available memory, followed by the java process invoking an external (usually R) script. This would cause the container to go over the memory limits set in the base.config, and the container would be killed by the supervisor (in my case k8s).
Even without an external application (outside of java) being invoked this can happen. The Xmx parameter only sets the size of the heap. The GC operates outside of this. So if the heap fills up, the GC kicks in, filling up the remaining memory, and causing the whole container to go OOM.
Also, some stages, like purple and sage are set to a resource class that is too low for the amount of work they need to do, especially for more difficult samples.

Command used and terminal output

No response

Relevant files

No response

System information

No response