Hi dann, * dann frazier [Tue Nov 17, 2020 at 03:05:32PM -0700]:
> grml2usb autopkgtest frequently fail in Ubuntu's CI. For example: > > https://objectstorage.prodstack4-5.canonical.com/v1/AUTH_77e2ada1e7a84929a74ba3b87153c0ac/autopkgtest-hirsute/hirsute/amd64/g/grml2usb/20201116_194744_cbe0f@/log.gz > An excerpt showing the error: > 2020-11-17 14:14:49,362 Installing default MBR > 2020-11-17 14:14:49,363 executing: dd if='/dev/loop0' of='/tmp/tmpa2umy9df' > bs=5 > 12 count=1 > 2020-11-17 14:14:49,381 executing: dd if=/usr/share/grml2usb/mbr/mbrldr > of=/tmp/ > tmpa2umy9df bs=439 count=1 conv=notrunc > 2020-11-17 14:14:49,385 executing: dd if='/tmp/tmpa2umy9df' of='/dev/loop0' > bs=5 > 12 count=1 conv=notrunc > 2020-11-17 14:14:49,401 executing: sync > 2020-11-17 14:14:49,433 Probing device via 'blockdev --rereadpt /dev/loop0' > blockdev: ioctl error on BLKRRPART: Device or resource busy > 2020-11-17 14:14:49,452 Execution failed: ("Couldn't execute blockdev on '%s' > (i > nstall util-linux?)", '/dev/loop0') Wow, interesting. > I am able to reproduce this on an OpenStack instance with a failure rate of > 33% (36 failures, 110 passes). My theory is that the sync is not sufficient, > and that we really need to do a targeted fsync of the file. With the > attached patch, I've yet to see a failure in 42 iterations. Very nice catch! Thanks for your bug report and many thanks also for providing a patch with it! Quoting your other mail (Message-ID: <20201117222219.GA35687@xps13.dannf> / https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=975015#10): > Looking at git history, I see this commit that makes an assertion > counter to my test results: > http://git.grml.org/?p=grml2usb.git;a=commit;h=7a6e10df > Previous to that commit, I did not see fsync being used. Michael: is > that commit message based on experimentation? If so, perhaps neither > sync approach is really sufficient, and maybe we need to implement a > retry loop :/ I'm afraid I don't have any further details available anymore nowadays, so I'd tend to say I was wrong. :) Your patch looks fine to me and I'm all for giving this a try, if it should fail anywhere I'm more than happy to give a closer look at it. regards -mika-
signature.asc
Description: Digital signature