On Fri, Dec 02, 2022 at 04:15:15PM +0300, Andrei Borzenkov wrote:
On Fri, Dec 2, 2022 at 4:07 PM Daniel Morris <danielm@ecoscentric.com> wrote:
Not sure why OOM killer targetted plasmashell with a score of 0 this week, instead of one/many of the 34 chrome(ium) instances with a score of 300, or another 8 chromium instances with a score of 200, eg:
Nov 30 14:57:18 chunk kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Nov 30 14:57:18 chunk kernel: [ 17992] 507 17992 4366216 2998934 26849280 35029 0 plasmashell Nov 30 14:57:18 chunk kernel: [ 2515] 507 2515 296308145 15562 1798144 10297 300 chrome Nov 30 14:57:18 chunk kernel: [ 18647] 507 18647 8566333 6256 1388544 3418 200 chrome
Well, plasmashell consumes much more real memory so from this listing it is logical. oom_score_adj is not "OOM score", it is a hint.
Aha. Thanks for the explanation. I'd over-guestimated a sum of the real memory for all the chromes, which turns out to be less than a quarter of plasmashell. But in any case I suppose the oom reaper deals with individual processes and not the sum of many suspect ones named the same. Daniel