Yes, old PIDs left from previous jobs don't utilize any resources, because they are stuck in non-interuptable sleep, but "cpu.stat" in existing cgroupv2 is not cleared for new job => systime and usertime for jobs running in existing subgroup gets time from all previous jobs... so CPU account is completely wrong.
ah, shoot - I had not thought about that!I have to check at us, if I find some current examples at us and check their cpu.stats! :-/
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature