On Tue, Aug 22, 2023 at 09:00:06AM -0700, Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote: > We can probably also kick FLUSH_TIME forward as well. True, they guard same work. > Perhaps I can move both into do_stats_flush() then. If I understand > correctly this is what you mean? Yes. > What do you think? The latter is certainly better looking code. I wasn't sure at first about moving stats_flush_threshold reset before actual flush but on second thought it should not be a significant change - readers: may skip flushing, the values that they read should still be below the error threshold, - writers: may be slowed down a bit (because of conditional atomic write optimization in memcg_rstat_updates), presumably not on average though. So the latter should work too. Michal
Attachment:
signature.asc
Description: PGP signature
