Impact of mismatched improved_mboxlist_sort in murder environment?

Matt Elson mailing_lists.cyrus_info at melson.fastmail.net
Mon May 16 07:12:45 EDT 2011


Hey all,

I'm in the middle of moving from a 2.2.13p1 Murder environment to a
2.3.16 Murder, moving the backends first. I've been moving people over
slowly (thousand or so now, I think) and while running a few checks, I
got the following error when I ran ctl_mboxlist -m -w

fatal error: Internal error: assertion failed: ctl_mboxlist.c: 297:
strcmp(realpart, unused_mbdata->server)

Poking around I would get this on certain mailboxes that ctl_mboxlist
would hit, so I shuffled a few of the problematic mailboxes to a test
server, where they would show the same error.

After hitting my head against the wall, I noticed that the new (2.3.16)
backends have improved_mboxlist_sort: 1 set, whereas the
not-yet-upgraded frontends (2.2.13p1)+mupdate servers, well, don't even
have the option ;).  I guessed that this mismatch might be the problem,
and while I've not done any extensive testing since the system is live,
if I shutdown the test server with my problematic mailboxes, dump the
mailboxes.db there, turn off improved_mboxlist_sort, and reload the
dump, I can issue ctl_mboxlist -m -w till my heart is content.

Outside of ctl_mboxlist functions, everything's fine and I haven't
noticed anything amiss, but (obviously) that doesn't mean there isn't
some problem waiting to bite me.

I guess the question is 1) if I'm right that my problems are likely
related to improve_mboxlist_sort, and 2) if anyone has any idea what
sort of other problems I may encounter due to this mismatch.

Due to circumstances beyond my control, getting downtime in my
environment (even for what should be a straight forward dump/reload)
anytime soon is going to be problematic, but I can push for it if
necessary.  On the other hand, if the problem is relatively harmless and
going to sort itself out once I get the mupdate server to 2.3.16, I'm
fine with waiting until then or upgrading the mupdate server sooner.  

Thanks in advance for any thoughts!

Matt,
Wishing he could go straight to 2.4.x.  Sigh.


More information about the Info-cyrus mailing list