I am running a locally compiled cyrus 2.0.16 on a Red Hat 7.1 system.
Twice now one of my users (only one, but this one happens to be my
boss!) has observed a wierd, transient failure mode: in this mode, imapd
hangs on any attempt to write to a folder. Here is an example, caught by
ethereal:
0000000c LIST "" inbox.Sent
* LIST () "." "inbox.Sent"
0000000c OK Completed (0.000 secs 2 calls)
0000000d APPEND inbox.Sent {353}
That's it -- no word from imapd ever again. Here, for reference, is how
it's supposed to look:
00000006 LIST "" inbox.Sent
* LIST () "." "inbox.Sent"
00000006 OK Completed (0.000 secs 2 calls)
00000007 APPEND inbox.Sent {355}
+ go ahead
Date: Mon, 26 Nov 2001 22:25:29 -0800 (PST)...
00000007 OK [APPENDUID 1001546367 273] Completed
But in the hanging mode, the client never gets a "+ go ahead" from
imapd. Once this behaviour starts, it occurs for any imapd process that
my boss creates until the cyrus master process is killed and restarted.
It does not occur for other users' imapd processes, even while it is
occuring for my boss's. Mail delivered via lmtp continues to arrive
normally in my boss's inbox, even while this is occuring. The last time
this occured (about 10 days ago), I reconstructed my boss's mailboxes
and assumed that was that, but today it happened again.
My boss uses the same clients (netscape and pine) as most of my other
users. The only thing unusual about his account is its size: 246 MB
distributed over 299 folders (we have no quotas). I am wondering if
sometimes an operation on a large file or directory might time out and
leave cyrus in an inconsistent state.
One last piece of information: when setting up cyrus, I chose to ignore
the documentation's advice to set the user, quota, and partition
directories to update synchronously. The documentation implies that the
only consequence of this is possible data loss during a hard shutdown,
and since that has never happened to us (our server has a UPS and we a
re in a hospital with its own emergency power) I choose to accept that
risk. I am now wondering if asynchronicity might have additional
consequences.
That's the story. Any ideas? Can anyone suggest a way to get more
information out of cyrus, e.g. an strace or ltrace of a running imapd
process so I can see what it is doing when it dies?
