IMAP/Murder mass deleting users

Henrique de Moraes Holschuh hmh at debian.org
Wed Aug 31 12:58:09 EDT 2005


On Tue, 30 Aug 2005, Jim Herbert wrote:
> Any help anyone can provide with the following issue will be greatly
> appreciated.  I believe I have stumbled into a seriously dangerous
> scenario with Cyrus IMAP+Murder, in which IMAP mass-deletes users.

I have had similar experiences when I severely fucked up a test system and
had inconsistent databases across the Murder.

Basically, if you *KNOW* the backends are never screwed up in the sense that
maiboxes are duplicated across the backends, you should always use -ma
instead of -m.   This is relatively easy to ensure if you are not using
autocreate or something else that could cause unexpected mailbox creations
at the wrong backend.

In fact, given that cyrus murder might very well choose to keep the mailbox
with no mail in it and trash the one with all user mail in it, I use -ma
always, and watch the logs for duplicates/weirdness.

The safe way to rebuild a murder cluster if the mupdate master is borked is
to:

1. Kill all the databases in the mupdate master

2. for every backend node do:
   mupdate -ma

Then inspect the logs to see if there are any colisions.

> else.  For me, I'd like to feel comfortable knowing that this can't happen
> again.  So, any help that anyone can lend in locating and eliminating this
> issue would be greatly appreciated.

ctl_mboxlist -ma will ensure it won't happen again the way you experienced
it...  duplicate mboxes in a murder are bad in the sense that you don't know
exactly where email will end up, but at least it WILL end up in one of the
duplicates IME with 2.1 (didn't try with 2.2 or 2.3), which is MUCH better
and easier to fix than email getting deleted :)

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh


More information about the Cyrus-devel mailing list