On 22/08/2015 12:30 PM, Heiko Schlittermann wrote:
Hi Daryl,
Daryl Richards <[email protected]> (Do 20 Aug 2015 18:00:05 CEST):
1ZSOZc-0002m2-U8 1ZSOZd-0002mG-02 1ZSOZd-0002lz-0d 1ZSOZl-0002oF-IJ
1ZSOZl-0002oI-FN 1ZSOZl-0002oX-Hw 1ZSOaY-0002rn-Dz 1ZSOaY-0002rq-DP
1ZSOaY-0002rz-D3
The dumpdb line shows the destination host, and various message ids, but not
the one that is frozen.. ??
Yes, the message ids shown are normally IDs of messages waiting for that
host. Do you run lots of queue runners in parallel?
Well, all those message ids where no longer in the queue itself. mailq only
showed the one message, as delivered but frozen. That list was also
generated about 45 minutes after the message was actually sent through. I do
not run queue runners in parallel.
If you're familiar with git, you may cherry-pick the commits
dadff1d47e54962b0fdf98e8ce5cef42b6cb7fb5
6b51df8340eacc95e3def9a4376506610e91996c
to exim-4_86.
Or, you may try the patch I appended here.
It makes me worry that you don't use split_spool_directory.
Can you reproduce the crash?
(I did about what Wolfgang Breyha did too: blocked the outgoing
connections for a short while to get the wait-<transport> db filled.
Then cp wait-<transport> away and re-enable the connections again.)
During flushing the waiting message Exim crashed. And, additionally I
could reproduce the behaviour if I copied back the saved
wait-<transport> file, because now it clearly contained an "expired"
message ID.
So, can you - with a plain exim-4_86 reproduce the crash?
Does it still crash with the above commits applied?
If it crashes, can you compile Exim with -ggdb and let it core dump?
Hello Heiko,
I'm not that up on git, etc, but I can and do build from FreeBSD ports.
It was easy enough to add your patch into that tree and build a
version... And, it's been running 24 hours now and the problem *seems*
to be gone. The offending emails are generated many times a day, so I
didn't have to try to inject them manually.
I will keep watching it and will let you know if any more end up frozen.
--
## List details at https://lists.exim.org/mailman/listinfo/exim-dev Exim
details at http://www.exim.org/ ##
