This is a really hard to hit issue, that requires a lot of things to be in place for it to happen. But it is an unexpected data loss issue. I'll wait tonight for the change to be merged, though I really don't like it. You could have informed me on this thread earlier. Please, in the future, keep release-managers/maintainers updated about any critical changes. The only reason this is getting merged now, is because of the Jenkins migration which got completed surprisingly quickly. On Thu, Mar 31, 2016 at 7:08 PM, Kotresh Hiremath Ravishankar <khiremat@xxxxxxxxxx> wrote: > Kaushal, > > I just replied to Aravinda's mail. Anyway pasting the snippet if someone misses that. > > "In the scenario mentioned by aravinda below, when an unlink comes on a entry, in changelog xlator, it's 'loc->pargfid' > was getting modified to "/". So consequence is that , when it hits posix, the 'loc->pargfid' would be pointing > to "/" instead of actual parent. This is not so terrible yet, as we are saved by posix. Posix checks > for "loc->path" first, only if it's not filled, it will use "pargfid/bname" combination. So only for > clients like self-heal who does not populate 'loc->path' and the same basename exists on root, the > unlink happens on root instead of actual path." > > Thanks and Regards, > Kotresh H R > > ----- Original Message ----- >> From: "Kaushal M" <kshlmster@xxxxxxxxx> >> To: "Aravinda" <avishwan@xxxxxxxxxx> >> Cc: "Gluster Devel" <gluster-devel@xxxxxxxxxxx>, maintainers@xxxxxxxxxxx, "Kotresh Hiremath Ravishankar" >> <khiremat@xxxxxxxxxx> >> Sent: Thursday, March 31, 2016 6:56:18 PM >> Subject: Re: [Gluster-Maintainers] Update on 3.7.10 - on schedule to be tagged at 2200PDT 30th March. >> >> Kotresh, Could you please provide the details? >> >> On Thu, Mar 31, 2016 at 6:43 PM, Aravinda <avishwan@xxxxxxxxxx> wrote: >> > Hi Kaushal, >> > >> > We have a Changelog bug which can lead to data loss if Glusterfind is >> > enabled(To be specific, when changelog.capture-del-path and >> > changelog.changelog options enabled on a replica volume). >> > >> > http://review.gluster.org/#/c/13861/ >> > >> > This is very corner case. but good to go with the release. We tried to >> > merge >> > this before the merge window for 3.7.10, but regressions not yet complete >> > :( >> > >> > Do you think we should wait for this patch? >> > >> > @Kotresh can provide more details about this issue. >> > >> > regards >> > Aravinda >> > >> > >> > On 03/31/2016 01:29 PM, Kaushal M wrote: >> >> >> >> The last change for 3.7.10 has been merged now. Commit 2cd5b75 will be >> >> used for the release. I'll be preparing release-notes, and tagging the >> >> release soon. >> >> >> >> After running verification tests and checking for any perf >> >> improvements, I'll make be making the release tarball. >> >> >> >> Regards, >> >> Kaushal >> >> >> >> On Wed, Mar 30, 2016 at 7:00 PM, Kaushal M <kshlmster@xxxxxxxxx> wrote: >> >>> >> >>> Hi all, >> >>> >> >>> I'll be taking over the release duties for 3.7.10. Vijay is busy and >> >>> could not get the time to do a scheduled release. >> >>> >> >>> The .10 release has been scheduled for tagging on 30th (ie. today). >> >>> In the interests of providing some heads up to developers wishing to >> >>> get changes merged, >> >>> I'll be waiting till 10PM PDT, 30th March. (0500UTC/1030IST 31st >> >>> March), to tag the release. >> >>> >> >>> So you have ~15 hours to get any changes required merged. >> >>> >> >>> Thanks, >> >>> Kaushal >> >> >> >> _______________________________________________ >> >> maintainers mailing list >> >> maintainers@xxxxxxxxxxx >> >> http://www.gluster.org/mailman/listinfo/maintainers >> > >> > >> _______________________________________________ Gluster-devel mailing list Gluster-devel@xxxxxxxxxxx http://www.gluster.org/mailman/listinfo/gluster-devel
