On Tue 17-02-26 10:19:08, JP Kobryn (Meta) wrote: > On 2/17/26 4:37 AM, Michal Hocko wrote: > > On Mon 16-02-26 23:48:42, JP Kobryn (Meta) wrote: > > > On 2/16/26 1:07 PM, Michal Hocko wrote: > > [...] > > > > [*] btw. I believe you misaccount MPOL_LOCAL because you attribute the > > > > target node even when the allocation is from a remote node from the > > > > "local" POV. > > > > > > It's a good point. The accounting as a result of fallback cases > > > shouldn't detract from an investigation though. We're interested in the > > > node(s) under pressure so the relatively few fallback allocations would > > > land on nodes that are not under pressure and could be viewed as > > > acceptable noise. > > > > This is really confusing. You simply have no means to tell the > > difference between the requested node and the real node used so you > > cannot really say whether the memory pressure is because of fallbacks or > > your mempolicy configurations. That means that you cannot tell the > > difference between the source of the pressure and victim of that > > pressure. > > What if I excluded the fallback cases? I could get the actual node from > the allocated page and compare against the requested node or node mask.
I think it would make sense to send the per-node reclaim stats separately as there doesn't seem to be any dispute about that. For mempolicy stats try to define semantic for each mempolicy first. What exactly do you miss from existing numa_*? Do you want to count number of requests/successes. Do you want to track failures? In what kind of granularity (track fallback nodes)? -- Michal Hocko SUSE Labs
