Hi Marco, Here is the revert on top of 6.1 (currently only build-tested): https://github.com/ceph/ceph-client/commit/b45280b209388082a0880eb7d348436cc0a6ee6d Also, can you share "ceph report | jq .osdmap" output captured before, during (in the middle of it going at <10M/s) and after the test? Thanks, Ilya On Mon, Dec 19, 2022 at 12:59 PM Roose, Marco <marco.roose@xxxxxxxxxxxxx> wrote: > > [...] > > Marco, if you have time, it’d be great if you tested Liunx 5.6-rc7 and 5.6, where the commit entered Linus’ master branch. > > already did it: as expected RC7 is fine 5.6 shows the exact problem > > Kind regards, > Marco Roose > > -----Original Message----- > From: Paul Menzel <pmenzel@xxxxxxxxxxxxx> > Sent: 19 December 2022 12:46 > To: Xiubo Li <xiubli@xxxxxxxxxx>; Roose, Marco <marco.roose@xxxxxxxxxxxxx> > Cc: Ilya Dryomov <idryomov@xxxxxxxxx>; ceph-devel@xxxxxxxxxxxxxxx > Subject: Re: PROBLEM: CephFS write performance drops by 90% > > Dear Xiubo, > > > Am 19.12.22 um 11:48 schrieb Roose, Marco: > > > my colleague Paul (in CC) tried to revert the commit, but it was'nt > > possible. > > > -----Original Message----- > > From: Xiubo Li <xiubli@xxxxxxxxxx> > > Sent: 19 December 2022 01:16 > > To: Roose, Marco <marco.roose@xxxxxxxxxxxxx>; Ilya Dryomov > > <idryomov@xxxxxxxxx> > > Cc: Ceph Development <ceph-devel@xxxxxxxxxxxxxxx> > > Subject: Re: PROBLEM: CephFS write performance drops by 90% > > […] > > > Since you are here, could you try to revert this commit and have a try ? > > > > Let's see whether is this commit causing it. I will take a look later > > this week. > > Unfortunately, reverting the commit is not easily possible, as the code was changed afterward too. It’d be great if you provided a git branch with the commit reverted. > > Marco, if you have time, it’d be great if you tested Liunx 5.6-rc7 and 5.6, where the commit entered Linus’ master branch. > > > Kind regards, > > Paul
