On Sat, Nov 21, 2020 at 09:44:29PM +0000, Daire Byrne wrote: > ----- On 21 Nov, 2020, at 01:03, Jeff Layton jlayton@xxxxxxxxxx wrote: > > On Fri, 2020-11-20 at 17:44 -0500, J. Bruce Fields wrote: > >> On Fri, Nov 20, 2020 at 05:38:31PM -0500, J. Bruce Fields wrote: > >> > I think the first one's all that's needed to fix the problem Daire > >> > identified. I'm a little less sure of the rest. > > I can confirm that patch 1/8 alone does indeed address the reported revalidation issue for us (as did the previous patch). The re-export server's client cache seems to remain intact and can serve the same cached results to multiple clients. Thanks again for the testing. > I should also mention that I still see a lot of unexpected repeat > lookups even with the iversion optimisation patches with certain > workloads. For example, looking at a network capture on the re-export > server I might see 100s of getattr calls to the originating server for > the same filehandle within 30 seconds which I would have expected the > client cache to serve. But it could also be that the client cache is > under memory pressure and not holding that data for very long. That sounds weird. Is the filehandle for a file or a directory? Is the file or directory actually changing at the time, and if so, is it the client that's changing it? Remind me what the setup is--a v3 re-export of a v4 mount? --b. > But now I do wonder if these NFSv4 directory modifications and > pre/post change attributes could be one potential contributor? I might > run some production loads with a v3 re-export of a v3 server to see if > that changes anything. > > Many thanks again for the patches, I will take the entire set and run > them through our production re-export workloads to see if anything > shakes out. > > Daire
