Out-of-memory victim selection with BPF

Posted Aug 17, 2023 18:36 UTC (Thu) by roc (subscriber, #30627)
In reply to: Out-of-memory victim selection with BPF by roc
Parent article: Out-of-memory victim selection with BPF

It *might* be good for Linux to have a way to disable overcommit on a per-process basis, so when you've got some special processes that really can handle OOM, they can get that special treatment.

Out-of-memory victim selection with BPF

Posted Aug 17, 2023 21:20 UTC (Thu) by flussence (guest, #85566) [Link] (2 responses)

I'd like that for the opposite - most of my desktop behaves fine with overcommit off, but a handful of programs insist on allocating a hundred gigabytes and crash when they don't get it.

Out-of-memory victim selection with BPF

Posted Aug 18, 2023 10:36 UTC (Fri) by khim (subscriber, #9252) [Link]

And who would write such desktop for you? Currently the worst offenders which allocate tons of memory they would never use are precisely programs that show desktop, launch other programs and so on.

Out-of-memory victim selection with BPF

Posted Aug 18, 2023 23:40 UTC (Fri) by roc (subscriber, #30627) [Link]

I don't know what you're running on your desktop but I don't believe "most" programs are resilient to OOM on most people's desktops. Certainly anything using Python, Javascript, Java or almost any other language than C or maybe C++ cannot be resilient to OOM. Any application that seriously tries to be resilient to OOM must have a rigorous testing regime using fault injection to test the OOM paths, and I haven't seen *any* Linux desktop application or toolkit with such a testing regime. Heck, do X or Wayland have such a regime? Because if they don't, there's no hope at all.