[Dovecot] dsync timeout?

[micah anderson]

Sean Kamath <kamath at moltingpenguin.com> writes:

> On Jan 30, 2013, at 3:46 PM, micah anderson <micah at riseup.net> wrote:
>> Seems that only the above process was still around and no other dsync
>> processes. I have three machines that all have this happening it seems.
>> 
>> I wonder if there is a ssh configuration option I could set to make
>> these die off.
>
> If the ssh process isn't sending anything, and just waiting for read()s, and keepalives are turned off, the SSH session might never know the remote side is long gone. . .

This time I managed to capture a process that was stuck and look at it
from the server side, and the client side:

on the server:

2000     19470  0.0  0.0   7512  3816 ?        Ss   Feb05   0:01 /usr/bin/dsync dsync-server -E -u foo
# strace -s 1024 -F -p 19470
Process 19470 attached - interrupt to quit
write(2, "dsync-remote(foo): Error: mdbox /srv/maildirbackups/foo/daily.1/storage: Duplicate GUID 96860517f68aa94f8b51000097f19f0b in m.41:682501 and m.37:653225\n", 167

on the client:

root     19001  0.0  0.0  41308  1600 ?        S    Feb05   0:00 ssh -i /root/.ssh/backmaildir_id_rsa backmaildir at hoopoe-pn /usr/bin/dsync -u foo server

# strace -s 1024 -F -p 19001
Process 19001 attached - interrupt to quit select(8, [4], [], NULL, NULL

interestingly, now that I've been watching this more, the same users
keep getting wedged. 

When I attempt to do a dsync of that user by hand, I get this:

dsync-local(foo): Error: Unexpected reply from server: 13	d2a100118c45d24f760f000097f19f0b	3561	128	\Recent	1353980259

I tried one of the other users that was stuck, and it gave me:

dsync-remote(bar): Error: Corrupted dbox file /srv/maildirbackups/bar/daily.1/storage/m.130 (around offset=22532): msg header has bad magic value

This looks like there is something corrupted with the dbox for the user
on the client side, is there something I can do to repair those?

> If any data were transmitted, it would discover the remote side is turned off.

One thing I am doing is using a ssh controlmaster socket, and if I kill
the process on the client's side, the server side process also dies.

micah