earlyoom is what we use to address this. We can't tolerate any kind of swapping at all in our workloads, where it is better for the system to kill one process

earlyoom is what we use to address this. We can't tolerate any kind of swapping at all in our workloads, where it is better for the system to kill one process to save the others, than for the system to slow down or lock up.