losing mail

losing mail

[Michael R Cook]

I just observed what I had always suspected: If gnus is fetching
mail from my `nnmail-spool-file' while mailing is arriving, mail can
be lost.

I was receiving a large file, uuencode and broken up into multiple
messages.  I hit `g' a couple times as the parts were coming in.
One part never showed up in my nnml folder.  But, according to the
log kept by my procmail script, that message *was* stored into my
`nnmail-spool-file'.

Should there be some kind of locking going on between my procmail
script and the gnus nnml backend?  Hmm...

Re: losing mail

[SL Baur]

Michael R Cook <mcook@cognex.com> writes:

> I just observed what I had always suspected: If gnus is fetching
> mail from my `nnmail-spool-file' while mailing is arriving, mail can
> be lost.

Without coordinated locking between procmail and movemail, yes, mail
can be lost.

Here's a simple definitive test to see if locking is working between
procmail and movemail.

Add something like this recipe to your .procmailrc:

:0 w:
* ^X-yzzy: Gnus Locking Test
| /home/steve/bin/locktest.sh >> foo.spool


locktest.sh should look something like:
#!/bin/sh

while read line; do
        echo $line
done
sleep 120
echo You Lose
## locktest.sh ends here

Now send a message to yourself with a header `X-yzzy: Gnus Locking Test'
and verify with ps that you have a `hung' procmail process.

Now in another Xterm execute something like:

$ /usr/src/xemacs-20.0/lib-src/movemail ~/spool/foo.spool /tmp/foo-bar

With a `sleep 120' in the test script you have two minutes to execute
the movemail command.  You can shorten the wait if you like.

The movemail command should also hang, because foo.spool is locked by
procmail.  If you _do_ have locking you will have your test message
complete with an appended `You Lose' line at the bottom in /tmp/foo-bar
and ~/spool/foo.spool is zero length.

If you _don't_ have locking, you will have two non-zero length files
when you're done.  /tmp/foo-bar with a copy of the test message, and
~/spool/foo.spool with the single line `You Lose'.

Adjust paths to fit the local environment, tested against procmail-3.11pre7
on Linux-2.0.30, your mileage may (but shouldn't) vary, batteries not
included, offer void where prohibited etc. etc.

See
	http://www.miranova.com/gnus-list/Nov-1995/0379.html
for more information from the mailing list archives.

Re: losing mail

[Michael R Cook]

>>>>> "sb" == SL Baur <steve@xemacs.org> writes:

 sb> Here's a simple definitive test to see if locking is working between
 sb> procmail and movemail.

Yes, that was my problem--locking wasn't working properly (my own
fault, really).

Thanks for the thorough response.

Re: losing mail

[Edward J. Sabol]

Excerpts from mail: (27-Jul-97) losing mail by Michael R Cook
> I just observed what I had always suspected: If gnus is fetching
> mail from my `nnmail-spool-file' while mailing is arriving, mail can
> be lost.
[...]
> Should there be some kind of locking going on between my procmail
> script and the gnus nnml backend?  Hmm...

You need to recompile your movemail binary to use procmail-compatible .lock
semaphore files. Read this article <http://www.rosat.mpe-garching.mpg.de/
mailing-lists/procmail/1997-03/msg00173.html> from the procmail mailing list
Web archive for details on how to do this.