Re: [exim] write to main log fails after log cycling: Bad fi…

I wanted to follow up to my previous email with a few more useful
details I neglected to mention previously:
I'm running on Debian: Linux hostname 3.16.0-4-amd64 #1 SMP Debian
3.16.7-ckt25-2 (2016-04-08) x86_64 GNU/Linux
With this version of exim: Exim version 4.84_2 #2 built 13-Mar-2016 17:47:19
logrotate is version 3.8.7, but as I mention the problem persists if I
use exicyclog.

> I have anacron set up to do daily tasks, including updating peerguardian
> lists -- which generates a mail almost immediately, and generates a second
> mail a few minutes later -- and cycling logs, including exim's logs.
> Everything works great, except that about half the time, I get this in my
> syslog:
> Jul 8 07:38:54 hostname exim[6319]: 2016-07-08 07:38:54 <mailid> failed to
> write to main log: length=92 result=-1 errno=9 (Bad file descriptor)
> Jul 8 07:38:54 hostname exim[6319]: write failed on panic log: length=116
> result=-1 errno=9 (Bad file descriptor)
>
> I figured it had exim's logfile open, the logs got cycled, and the file
> descriptor was bad. But...
> 1) I thought it should handle that gracefully
> 2) It only prints the error to syslog about half the time
> 3) Oddest thing: if I look in exim's main log, the first mail always works
> fine (and goes to the new log file); the second email, which gets sent
> about three minutes later, is the only one that has a problem.
>
> When it has a problem, the mainlog looks like this:
> 2016-07-08 07:35:15 <mailid1> <= smarthostlogin@??? U=root P=local
> S=55147
> 2016-07-08 07:35:15 <mailid1> => loginname <root@hostname> R=local_user
> T=mail_spool
> 2016-07-08 07:35:15 <mailid1> Completed
> 2016-07-08 07:38:54 <mailid2> <= smarthostlogin@??? U=root P=local
> S=1074
> 2016-07-08 07:46:46 Start queue run: pid=6394
> 2016-07-08 07:46:46 <mailid2> => loginname <root@hostname> R=local_user
> T=mail_spool
> 2016-07-08 07:46:46 <mailid2> Completed
>
> As you can see, the first mail after log cycling works. The second mail,
> sent about three minutes later, doesn't complete immediately; but the next
> time the queue runs, at 7:46, it does get sent. In fact, the second
> message shows up in my mailbox twice.
>
> It's not the worst problem in the world, but it would be nice if I could
> figure out what's going on. I was previously using /usr/sbin/logrotate,
> and then switched to using exicyclog to cycle exim's logs, but the problem
> is exactly the same -- about half the time the above problem shows up (and
> the other half, everything works just like it should, with no error
> messages to syslog). Any ideas? Anything I can try? Thanks!
>
>