|
|
Log in / Subscribe / Register

Out-of-memory victim selection with BPF

Out-of-memory victim selection with BPF

Posted Aug 18, 2023 13:28 UTC (Fri) by mathstuf (subscriber, #69389)
In reply to: Out-of-memory victim selection with BPF by rrolls
Parent article: Out-of-memory victim selection with BPF

> a program should only ever get OOM-killed if that program is the one trying to allocate memory in the first place.

A program allocating all memory it can and just sitting on it can then cause tiny utilities to fail to allocate a reasonably sized request.

There's actually a leak in macOS of Mach ports we're running into on our CI machines that causes process launch to fail, so any kind of remediation without an already-open System Monitor is DOA (e.g., SSH can't launch `zsh` for any kind of administration or even investigation).

I think that some more structured decision making behind which process needs to go is better than "whoever happens to get unlucky". Not that Linux's OOM killer is a *good* implementation, but I'd much rather something other than the processes I'm using to try and wrangle things be eligible for being next in line.


to post comments


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds