Post #1736087
2022-11-22 20:57 UTC
@itamarst you read my mind: yes we want to use that. Specifically we have pre-allocated child processes which are the first to die. They're empty processes we keep around to speed up loading a new page; the idea is to detect when they die and take that as a sign of memory running low. The problem is that we currently don't have a way to tell if a process was reaped by the OOM killer but @xlerb is working on it so... soon, probably!
Replies (1)
-
@itamarst@hachyderm.io 2022-11-22 21:57
@gabrielesvelto @xlerb Nice! My use case is somewhat different—a Python memory (and performance) profiler running in production, and if we're about to hit OOM I want to dump a report ASAP so users can debug what happened. The other thing I've considered is using PSI plus free memory/swap as a heuristic, perhaps inspired by whatever Ubuntu's systemd-oomd config is doing.