Hi Timo, Thanks for your quick response, apologies for my slow one.
Timo Sirainen <t...@iki.fi> writes: >> I'm using 2.2.13 on Debian stable, and I perform regular dsync >> backups. Typically if there is an error, dsync can recover from it the >> next time it is run, but I've been encountering a new error that doesn't >> seem to self-heal: >> >> # /usr/bin/dsync -u micah backup ssh -i /root/.ssh/id_rsa >> backmaildir@internal.example /usr/bin/dsync -u micah >> dsync-remote(micah): Error: >> /srv/maildirbackups/m/micah/daily.1/mailboxes/INBOX/dbox-Mails/dovecot.index: >> Couldn't set transaction log view (seq 9..13) >> dsync-remote(micah): Error: Failed to read transaction log for mailbox INBOX >> >> Re-running the dsync never fixes the issue. >> >> I'm happy to provide any additional information if so required to help >> track down this problem! > > Can you create a tarball of such mailbox directories that reproduces the > problem? Probably this would work: > > - copy source and destination dboxes to /tmp/test/1 and /tmp/test/2 > - delete all the mailboxes except INBOX I noticed an interesting bit of information doing this: the users that are having a problem have no dbox-Mails directory in their source mailboxes/INBOX directory - but on the destination where I am sending the backups, there is one. If I remove the backup's INBOX/ and then re-do the dsync, things work again. > - delete all the u.* files from the INBOXes. This probably doesn't > affect the dsync run - if it does run the mails instead through > http://dovecot.org/tools/dbox-anonymize.pl I have no u.* files on either side. There are m.* files though? > - delete dovecot.index.cache files > - make sure you can still reproduce this error by running: doveadm -o > mail=sdbox:/tmp/test/1 sync sdbox:/tmp/test/2 did you mean s/sdbox/mdbox/ here? also, I was using 'backup' and not sync, I dont know if that matters? I tried this with some that had issues, but I couldn't reproduce the error this way :p > - tar czf test.tar.gz /tmp/test > > Then send the tarball to me and I can fix the bug. Oh, also it would > be a good idea to test with a newer Dovecot version since there are > some indexing fixes since v2.2.13.. I'm working on getting a newer version of dovecot installed to test this. > Also I added some code that should give a better-debuggable error message if > this happens: http://hg.dovecot.org/dovecot-2.2/rev/8146fdc0de34 great, this will certainly be helpful! micah
