> > Hi Darrick, > > > > FYI, I started testing the log fixes backports from v5.14 along with > > the deferred > > patches 6-8 [1] with extra focus on recoveryloop tests. > > > > I know that Leah is also testing another batch of 5.15-only patches, so she > > may yet post another 5.15-only series before my 5.10-only series. > > > > In the meanwhile, if you have some spare time due to rc8, please try to > > look at the already posted patches 6-8 [2] that were deferred from the original > > stable submission per your request. > > This is pretty difficult request -- while I /think/ the LSN->CSN > conversion for the upper layers in patch 7 is correct, I'm not as > familiar with where 5.10 is right now as I was when that series was > being proposed for upstream. > > It /looks/ ok, but were I maintaining the 5.10 tree I'd be a lot more > comfortable if I had in hand all the results from running the long-soak > log recovery tests for a week. Sure, I can do that. > > (Which itself might be fairly difficult for 5.10...) That depends on the exact definition of running the long-soak and log recovery tests for a week. I did run the 'recoveryloop' group 100 times on baseline vs. backports and saw no regressions observed. Also ran 'auto' 10 times and 'soak' 10 times (and counting) with no regressions so far. However, some recoveryloop tests are failing in baseline, so running them "for a week" is meaningless for those tests: generic/019, generic/475, generic/648: failing often in all config generic/388: failing often with reflink_1024 generic/388: failing at ~1/50 rate for any config generic/482: failing often on V4 configs generic/482: failing at ~1/100 rate for V5 configs xfs/057: failing at ~1/200 rate for any config Note that some of those recoverloop tests may be failing due to older xfs_repair [*]. The following recoveryloop/soak tests have NOT failed so far: generic,455, generic/457, generic/646 generic/521, generic/522, generic/642 Should I just keep running those for a week? Another option is to declare those fixes too risky for LTS 5.10. That is always an option we need to consider. Anyway, I will post the full v5.14 series for review and continue running the long soak tests and the passing tests. Thanks, Amir. [*] I should note that the 5.10.y tests are run with xfsprogs 5.10 When I started testing 5.10.y, I used debian/testing with newer xfsprogs but that led to many issues compared to debain/bullseye with xfsprogs 5.10. I figured that it is not very likely for a distro to use kernel 5.10 and much newer xfsprogs (?) so I preferred to test more aligned versions for kernel/xfsprogs to match the real life use cases.
