On Thu, 2023-01-26 at 15:12 -0800, Roman Gushchin wrote: > On Thu, Jan 26, 2023 at 08:41:34AM +0100, Michal Hocko wrote: > > On Wed 25-01-23 15:14:48, Roman Gushchin wrote: > > > On Wed, Jan 25, 2023 at 03:22:00PM -0300, Marcelo Tosatti wrote: > > > > On Wed, Jan 25, 2023 at 08:06:46AM -0300, Leonardo Brás wrote: > > > > > On Wed, 2023-01-25 at 09:33 +0100, Michal Hocko wrote: > > > > > > On Wed 25-01-23 04:34:57, Leonardo Bras wrote: > > > > > > > Disclaimer: > > > > > > > a - The cover letter got bigger than expected, so I had to split it in > > > > > > > sections to better organize myself. I am not very confortable with it. > > > > > > > b - Performance numbers below did not include patch 5/5 (Remove flags > > > > > > > from memcg_stock_pcp), which could further improve performance for > > > > > > > drain_all_stock(), but I could only notice the optimization at the > > > > > > > last minute. > > > > > > > > > > > > > > > > > > > > > 0 - Motivation: > > > > > > > On current codebase, when drain_all_stock() is ran, it will schedule a > > > > > > > drain_local_stock() for each cpu that has a percpu stock associated with a > > > > > > > descendant of a given root_memcg. > > > > > > Do you know what caused those drain_all_stock() calls? I wonder if we should look > > > into why we have many of them and whether we really need them? > > > > > > It's either some user's actions (e.g. reducing memory.max), either some memcg > > > is entering pre-oom conditions. In the latter case a lot of drain calls can be > > > scheduled without a good reason (assuming the cgroup contain multiple tasks running > > > on multiple cpus). > > > > I believe I've never got a specific answer to that. We > > have discussed that in the previous version submission > > (20221102020243.522358-1-leobras@xxxxxxxxxx and specifically > > Y2TQLavnLVd4qHMT@xxxxxxxxxxxxxx). Leonardo has mentioned a mix of RT and > > isolcpus. I was wondering about using memcgs in RT workloads because > > that just sounds weird but let's say this is the case indeed. Then an RT > > task or whatever task that is running on an isolated cpu can have pcp > > charges. > > > > > Essentially each cpu will try to grab the remains of the memory quota > > > and move it locally. I wonder in such circumstances if we need to disable the pcp-caching > > > on per-cgroup basis. > > > > I think it would be more than sufficient to disable pcp charging on an > > isolated cpu. > > It might have significant performance consequences. > > I'd rather opt out of stock draining for isolated cpus: it might slightly reduce > the accuracy of memory limits and slightly increase the memory footprint (all > those dying memcgs...), but the impact will be limited. Actually it is limited > by the number of cpus. I was discussing this same idea with Marcelo yesterday morning. The questions had in the topic were: a - About how many pages the pcp cache will hold before draining them itself? b - Would it cache any kind of bigger page, or huge page in this same aspect? Please let me know if I got anything wrong, but IIUC from a previous debug (a)'s answer is 4 pages. Meaning even on bigger-page archs such as powerpc, with 64k pages, the max pcp cache 'wasted' on each processor would be 256k (very small on today's standard). Please let me know if you have any info on (b), or any correcting on (a). The thing is: having this drain_local_stock() waiver only for isolated cpus would not bring the same benefits for non-isolated cpus in high memory pressure as I understand this patchset is bringing. OTOH not running drain_local_stock() at all for every cpu may introduce performance gains (no remote CPU access) but can be a problem if I got the 'wasted pages' could on (a) wrong. I mean, drain_local_stock() was introduced for a reason. > > > This is not a per memcg property. > > Sure, my point was that in pre-oom condition several cpus might try to consolidate > the remains of the memory quota, actually working against each other. Separate issue, > which might be a reason why there are many flush attempts in the case we discuss. > > Thanks! > Thank you for reviewing! Leo
