Currently each container sees the whole instance memory instead of just the memory of the plan.
For example, I spawned an instance with 512MB of RAM and VSZ showed 5Gig.
This is problematic for JVM based software as it will try to use as much RAM as possible even if you limit the heap memory with with
-Xmx256m -Xss512k
. Where it would work fine if the JVM saw that the host only had the allocated memory by the plan (with potentially a buffer that would act like kubernetes limit vs request).