[Dovecot] 1.0rc7 getting near
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
Also everyone who are still experiencing hangs could you try the suggestions in this mail and tell me what happened: http://dovecot.org/list/dovecot/2006-August/015377.html
Timo Sirainen said the following On 11/8/2006 1:37:
Especially people who had compiling problems with 1.0rc6.
I was one of them, the original rc6 gives this error when it starts up: Aug 7 16:49:02 mail dovecot: Dovecot v1.0.rc6 starting up Aug 7 16:49:03 mail dovecot: Login process died too early - shutting down Aug 7 16:49:03 mail dovecot: child 7709 (login) killed with signal 11 Aug 7 16:50:00 mail dovecot: Login process died too early - shutting down Aug 7 16:50:00 mail dovecot: child 7737 (login) killed with signal 11 Aug 7 16:50:00 mail dovecot: child 7738 (login) killed with signal 11 Aug 7 16:50:00 mail dovecot: child 7739 (login) killed with signal 11
The night build does not have this error and everything is ok.
Ciao, luigi
-- / +--[Luigi Rosa]-- \
I've been in jail longer than Nelson Mandela. Maybe you want me to run for president. --John Mason, The Rock
On 11 août 06, at 01:37, Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
I used not to be able to compile it (OS X, problem with quota.c), now
it compiles fine and seems to be running fine from what I can tell.
Alan
--
The hacker: someone who figured things out and made something cool
happen.
.O.
..O
OOO
On 11 août 06, at 09:29, Alan Schmitt wrote:
On 11 août 06, at 01:37, Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
I used not to be able to compile it (OS X, problem with quota.c),
now it compiles fine and seems to be running fine from what I can
tell.
Well, I spoke too fast. After noticing that I had run out of PIDs, I
had to killed dovecot because I had a lot of:
root 6030 0.0 -0.0 0 0 ?? Z 1Jan70 0:00.00
(dovecot-auth)
I downgraded to rc2 which does not exhibit this issue.
Alan
--
The hacker: someone who figured things out and made something cool
happen.
.O.
..O
OOO
On Fri, 2006-08-11 at 13:48 +0200, Alan Schmitt wrote:
On 11 août 06, at 09:29, Alan Schmitt wrote:
On 11 août 06, at 01:37, Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
I used not to be able to compile it (OS X, problem with quota.c),
now it compiles fine and seems to be running fine from what I can
tell.Well, I spoke too fast. After noticing that I had run out of PIDs, I
had to killed dovecot because I had a lot of:root 6030 0.0 -0.0 0 0 ?? Z 1Jan70 0:00.00
(dovecot-auth)
Oh. Fixed now: http://dovecot.org/list/dovecot-cvs/2006-August/006270.html
On 11 août 06, at 17:10, Timo Sirainen wrote:
On Fri, 2006-08-11 at 13:48 +0200, Alan Schmitt wrote:
Well, I spoke too fast. After noticing that I had run out of PIDs, I had to killed dovecot because I had a lot of:
root 6030 0.0 -0.0 0 0 ?? Z 1Jan70 0:00.00 (dovecot-auth)
Oh. Fixed now: http://dovecot.org/list/dovecot-cvs/2006-August/006270.html
Thanks a lot. Unfortunately I won't be able to test this until early
december as I rely on this server and I won't have physical access to
it until then (and running out of pids is bad when I'm not by the
server).
Alan
--
The hacker: someone who figured things out and made something cool
happen.
.O.
..O
OOO
Timo Sirainen wrote:
On 11 août 06, at 01:37, Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
I tried 20060812 nightly on NetBSD-current, and it fails (rc2 works "almost" perfectly) during auth:
Aug 12 21:42:07 intrepid dovecot: Dovecot v1.0.rc6 starting up Aug 12 21:42:15 intrepid dovecot: child 3534 (auth) killed with signal 11 Aug 12 21:42:16 intrepid dovecot: child 9520 (auth) killed with signal 11 Aug 12 21:42:16 intrepid dovecot: auth(default): Error: write_full() failed: Broken pipe Aug 12 21:43:16 intrepid dovecot: imap-login: Disconnected: method=PLAIN, rip=192.168.1.8, lip=75.80.183.211, TLS
This is build with:
$ ./configure --prefix=/software/dovecot-20060812 --disable-ipv6 --with-ioloop =poll --with-notify=none --without-passwd-file --without-shadow --without-pam -- with-checkpassword --without-bsdauth --without-gssapi --without-vpopmail --witho ut-pop3d --with-storages=maildir --without-sql-drivers --without-deliver --witho ut-static-userdb --disable-asserts
Using kqueue/kqueue rather than poll/none has the same results (I tried to use kqueue first!).
Recommendations on how to debug?
ScottEOn Sat, 2006-08-12 at 22:03 -0700, Scott Ellis wrote:
I tried 20060812 nightly on NetBSD-current, and it fails (rc2 works "almost" perfectly) during auth:
Aug 12 21:42:07 intrepid dovecot: Dovecot v1.0.rc6 starting up Aug 12 21:42:15 intrepid dovecot: child 3534 (auth) killed with signal 11
Could you get gdb backtrace from it? The core should be written to /var/run/dovecot directory as long as you had high enough ulimit -c and hadn't changed auth_user from root.
And what passdb and userdb are you using? checkpassword?
On Wed, Aug 16, 2006 at 05:00:33PM +0300, Timo Sirainen wrote:
On Sat, 2006-08-12 at 22:03 -0700, Scott Ellis wrote:
I tried 20060812 nightly on NetBSD-current, and it fails (rc2 works "almost" perfectly) during auth:
Aug 12 21:42:07 intrepid dovecot: Dovecot v1.0.rc6 starting up Aug 12 21:42:15 intrepid dovecot: child 3534 (auth) killed with signal 11
Could you get gdb backtrace from it? The core should be written to /var/run/dovecot directory as long as you had high enough ulimit -c and hadn't changed auth_user from root.
And what passdb and userdb are you using? checkpassword?
I'm using checkpassword as the userdb and passdb. Still getting this problem with the "2nd" -latest of today (per the kqueue on BSD thread). Backtrace is as follows:
Core was generated by `dovecot-auth'. Program terminated with signal 11, Segmentation fault. #0 0x00000000004160d6 in sigchld_handler (signo=20, context=0x0) at passdb-checkpassword.c:147 147 request = hash_lookup(module->clients, POINTER_CAST(pid)); (gdb) bt #0 0x00000000004160d6 in sigchld_handler (signo=20, context=0x0) at passdb-checkpassword.c:147 #1 0x0000000000422599 in signal_read (context=0x0) at lib-signals.c:90 #2 0x0000000000422260 in io_loop_handler_run (ioloop=0x546000) at ioloop-kqueue.c:143 #3 0x000000000042161b in io_loop_run (ioloop=0x546000) at ioloop.c:280 #4 0x0000000000410dc6 in main (argc=1, argv=0x7f7fffffec00) at main.c:322 (gdb)
This is on NetBSD/amd64 -current (as of August 8th).
Needless to say, I can't try out the "fixed" kqueue functionality yet due to this issue. :-)
ScottEQuoting Timo Sirainen:
Ah, thanks, fixed now in CVS.
But now epoll doesn't work any more:
Aug 17 22:31:12 ymmv dovecot: IMAP(jh): file ioloop.c: line 71 (io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:12 ymmv dovecot: child 29248 (imap) killed with signal 6 Aug 17 22:31:12 ymmv dovecot: IMAP(jh): file ioloop.c: line 71 (io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:12 ymmv dovecot: child 29249 (imap) killed with signal 6 Aug 17 22:31:13 ymmv dovecot: imap-login: Login: user=<jh>, method=CRAM-MD5, rip=84.56.226.250, lip=85.214.25.134, TLS Aug 17 22:31:19 ymmv dovecot: IMAP(jh): file ioloop.c: line 71 (io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:19 ymmv dovecot: child 29250 (imap) killed with signal 6 Aug 17 22:31:20 ymmv dovecot: imap-login: Login: user=<jh>, method=CRAM-MD5, rip=84.56.226.250, lip=85.214.25.134, TLS Aug 17 22:31:30 ymmv dovecot: IMAP(jh): file ioloop.c: line 71 (io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:30 ymmv dovecot: child 29251 (imap) killed with signal 6
On 17.8.2006, at 23.42, Jakob Hirsch wrote:
Quoting Timo Sirainen:
Ah, thanks, fixed now in CVS.
But now epoll doesn't work any more:
Aug 17 22:31:12 ymmv dovecot: IMAP(jh): file ioloop.c: line 71
(io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:12 ymmv dovecot: child 29248 (imap) killed with signal 6
Hmm. Looks like io_remove() is called twice, which really shouldn't
be happening. I tried various ways to reproduce this, but couldn't.
Could you get a gdb backtrace of this? http://dovecot.org/bugreport.html
On Thu, 2006-08-17 at 23:56 +0300, Timo Sirainen wrote:
Aug 17 22:31:12 ymmv dovecot: IMAP(jh): file ioloop.c: line 71
(io_remove): assertion failed: (io->refcount > 0) Aug 17 22:31:12 ymmv dovecot: child 29248 (imap) killed with signal 6Hmm. Looks like io_remove() is called twice, which really shouldn't
be happening. I tried various ways to reproduce this, but couldn't.
Could you get a gdb backtrace of this? http://dovecot.org/bugreport.html
Well, everything except Maildir+IDLE. Fixed in CVS. I guess I'll build one more snapshot and release rc7 in a couple of hours.
Quoting Timo Sirainen:
Aug 17 22:31:12 ymmv dovecot: child 29248 (imap) killed with signal 6 Hmm. Looks like io_remove() is called twice, which really shouldn't
be happening. I tried various ways to reproduce this, but couldn't.
Could you get a gdb backtrace of this? http://dovecot.org/bugreport.html Well, everything except Maildir+IDLE. Fixed in CVS. I guess I'll build one more snapshot and release rc7 in a couple of hours.
ah, was just about to send the backtrace. works now, tnx!
On Fri, Aug 11, 2006 at 02:37:25AM +0300, Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
Hello Timo,
I haven't tested it yet, but at least it compiles (no autoconf problem anymore).
-- Thomas Hummel | Institut Pasteur hummel@pasteur.fr | Pôle informatique - systèmes et réseau
On Fri, Aug 11, 2006 at 02:49:03PM +0200, Thomas Hummel wrote:
Hello Timo,
I haven't tested it yet, but at least it compiles (no autoconf problem anymore).
As far as I tested it, it works for me on FreeBSD 6.1.
-- Thomas Hummel | Institut Pasteur hummel@pasteur.fr | Pôle informatique - systèmes et réseau
Timo Sirainen wrote:
To avoid having some stupid new problems with 1.0rc7, could you test this night's snapshot once it arrives in a couple of hours:
http://dovecot.org/nightly/dovecot-20060811.tar.gz
Especially people who had compiling problems with 1.0rc6.
So far compiles and works quite well on OBSD 3.9
rc6 didn't compile, rc5 didn't work with --with-ioloop=kqueue (TLS, bsdauth, with & without certificate authentication)
participants (8)
-
Alan Schmitt
-
Jakob Hirsch
-
Luigi Rosa
-
Michal Soltys
-
Scott Ellis
-
Scott Ellis
-
Thomas Hummel
-
Timo Sirainen