Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
On Sat, Nov 13, 2004 at 09:03:43PM -0500, Joe Maimon said: Stephen Gran wrote: On Thursday 11 November 2004 05:56 pm, Stephen Gran wrote: So, when start-stop-daemon (or daemon) sends a kill signal, it ends up signalling the wrong thread, and it takes a long time for the signal to work. Unfortunately in my role as packaager, I can't make too many assumptions about what methods people are using to run things on their system. Thanks, though. Perhaps a Pidfile option? Well, of course we're signalling the PID in Pidfile, the problem is that on 2.4 linux kernels, that's not the right thread. To show this, do this: [EMAIL PROTECTED]:~$ ps -el | egrep '(PID|clamav-milter)' F S UID PID PPID C PRI NI ADDR SZ WCHAN TTY TIME CMD 040 S 109 9222 1 0 69 0 - 1322 select ?00:00:10 clamav-milter 040 S 109 9223 9222 0 68 0 - 1322 poll ?00:00:13 clamav-milter 040 S 109 9224 9223 0 69 0 - 1322 rt_sig ?00:00:00 clamav-milter [EMAIL PROTECTED]:~$ ps -eLf | egrep '(PID|clamav-milter)' UIDPID PPID LWP C NLWP STIME TTY TIME CMD clamav9222 1 9222 0- Nov09 ?00:00:10 /usr/sbin/clamav-milter -dq --pidfile /var/run/clamav/clamav-milter.pid local:/var/run/clamav/clamav-milter.ctl clamav9223 9222 9223 0- Nov09 ?00:00:13 /usr/sbin/clamav-milter -dq --pidfile /var/run/clamav/clamav-milter.pid local:/var/run/clamav/clamav-milter.ctl clamav9224 9223 9224 0- Nov09 ?00:00:00 /usr/sbin/clamav-milter -dq --pidfile /var/run/clamav/clamav-milter.pid local:/var/run/clamav/clamav-milter.ctl [EMAIL PROTECTED]:~$ cat /var/run/clamav/clamav-milter.pid 9222 [EMAIL PROTECTED]:~$ uname -mr Linux 2.4.26-2-686-smp By contrast: [EMAIL PROTECTED]:~$ ps -el | egrep '(PID|clamav-milter)' F S UID PID PPID C PRI NI ADDR SZ WCHAN TTY TIME CMD 1 S 103 1010 1 0 76 0 - 12899 - ?00:00:00 clamav-milter [EMAIL PROTECTED]:~$ ps -eLf | egrep '(PID|clamav-milter)' UIDPID PPID LWP C NLWP STIME TTY TIME CMD clamav1010 1 1010 02 Nov11 ?00:00:00 /usr/sbin/clamav-milter -dq --pidfile /var/run/clamav/clamav-milter.pid local:/var/run/clamav/clamav-milter.ctl clamav1010 1 1011 02 Nov11 ?00:00:00 /usr/sbin/clamav-milter -dq --pidfile /var/run/clamav/clamav-milter.pid local:/var/run/clamav/clamav-milter.ctl [EMAIL PROTECTED]:~$ cat /var/run/clamav/clamav-milter.pid 1010 [EMAIL PROTECTED]:~$ uname -mr Linux 2.6.7-1-686 As you can see, the PID in pidfile is (correctly) the first thread started by init. The problem is, 9224 is the thread waiting for signals. If all the threads shared a pid, this would be fine - we would signal all the threads by signalling the PID, and things would work great. Alternately, if inter-thread communication worked OK, signalling 9222 would work, because it would pass the signal to 9224 in a relatively short time. Neither of these is true, so it takes forever without some hackery. On Linux 2.6 kernels, the various threads correctly share a PID, but have seperate LWP IDs, making this work as it should. Sigh. -- -- | Stephen Gran | But you'll notice Perl has a goto. | | [EMAIL PROTECTED] | -- Larry Wall in| | http://www.lobefin.net/~steve | [EMAIL PROTECTED]| -- pgpTxWdb1aO2o.pgp Description: PGP signature ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
Stephen Gran wrote: On Fri, Nov 12, 2004 at 10:29:20AM -0600, Jeremy Kitchen said: On Thursday 11 November 2004 05:56 pm, Stephen Gran wrote: [snip] So, when start-stop-daemon (or daemon) sends a kill signal, it ends up signalling the wrong thread, and it takes a long time for the signal to work. http://cr.yp.to/daemontools.html http://smarden.org/runit makes it very easy to send signals to applications. Unfortunately in my role as packaager, I can't make too many assumptions about what methods people are using to run things on their system. Thanks, though. Perhaps a Pidfile option? ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
On Thursday 11 November 2004 05:56 pm, Stephen Gran wrote: [snip] So, when start-stop-daemon (or daemon) sends a kill signal, it ends up signalling the wrong thread, and it takes a long time for the signal to work. http://cr.yp.to/daemontools.html http://smarden.org/runit makes it very easy to send signals to applications. -Jeremy -- Jeremy Kitchen ++ Systems Administrator ++ Inter7 Internet Technologies, Inc. [EMAIL PROTECTED] ++ www.inter7.com ++ 866.528.3530 ++ 815.776.9465 int'l kitchen @ #qmail #gentoo on EFnet IRC ++ scriptkitchen.com/qmail GnuPG Key ID: 481BF7E2 ++ jabber:[EMAIL PROTECTED] pgpJRGAW4kO7C.pgp Description: PGP signature ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
On Fri, Nov 12, 2004 at 10:29:20AM -0600, Jeremy Kitchen said: On Thursday 11 November 2004 05:56 pm, Stephen Gran wrote: [snip] So, when start-stop-daemon (or daemon) sends a kill signal, it ends up signalling the wrong thread, and it takes a long time for the signal to work. http://cr.yp.to/daemontools.html http://smarden.org/runit makes it very easy to send signals to applications. Unfortunately in my role as packaager, I can't make too many assumptions about what methods people are using to run things on their system. Thanks, though. -- -- | Stephen Gran | Old age is the harbor of all ills. -- | | [EMAIL PROTECTED] | Bion| | http://www.lobefin.net/~steve | | -- pgp4rtHD4Kd7w.pgp Description: PGP signature ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
[Clamav-users] clamav on woody: clamav-milter hangs when stopped
I have just upgraded to clamav-milter_0.80-5 on my Debian Woody system. When I run /etc/init.d/clamav-milter stop it hangs. When I do a ps ax it says sleep 0.1. Looks as if the offending line something like this: if [ -n $PID ]; then start-stop-daemon -q -K -o -p $PIDFILE $DAEMON while [ -d /proc/$PID ]; do sleep 0.1; done I think that /proc/$PID isn't going away and its staying in the loop indefinitely. I had a similar problem with the last debian release: start-stop-daemon wasn't killing the process. I've had to go back to the old hack of killing daemon and all instances of clamav-milter. Has anybody had similar problems? Any fixes? ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
Robert S schrieb: I have just upgraded to clamav-milter_0.80-5 on my Debian Woody system. When I run /etc/init.d/clamav-milter stop it hangs. When I do a ps ax it says sleep 0.1. Looks as if the offending line something like this: if [ -n $PID ]; then start-stop-daemon -q -K -o -p $PIDFILE $DAEMON while [ -d /proc/$PID ]; do sleep 0.1; done I think that /proc/$PID isn't going away and its staying in the loop indefinitely. I had a similar problem with the last debian release: start-stop-daemon wasn't killing the process. I've had to go back to the old hack of killing daemon and all instances of clamav-milter. Has anybody had similar problems? Any fixes? Please have a look at Debian bug #278198 (http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=278198) for an in-deep discussion of what's going on/wrong there. I'm sure Stephen Gran (the pkg maintainer) will come up with a new revision (and a woody backport) RSN. Thomas ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
Re: [Clamav-users] clamav on woody: clamav-milter hangs when stopped
On Thu, Nov 11, 2004 at 11:05:43AM +0100, Thomas Lamy said: Robert S schrieb: I have just upgraded to clamav-milter_0.80-5 on my Debian Woody system. When I run /etc/init.d/clamav-milter stop it hangs. When I do a ps ax it says sleep 0.1. Looks as if the offending line something like this: if [ -n $PID ]; then start-stop-daemon -q -K -o -p $PIDFILE $DAEMON while [ -d /proc/$PID ]; do sleep 0.1; done I think that /proc/$PID isn't going away and its staying in the loop indefinitely. I had a similar problem with the last debian release: start-stop-daemon wasn't killing the process. I've had to go back to the old hack of killing daemon and all instances of clamav-milter. Has anybody had similar problems? Any fixes? Please have a look at Debian bug #278198 (http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=278198) for an in-deep discussion of what's going on/wrong there. I'm sure Stephen Gran (the pkg maintainer) will come up with a new revision (and a woody backport) RSN. What kernel are you running? (I'm asking Robert, obviously). The short version is that I have found through trial and error that the 2.4 series of linux kernels don't implement threads that well - each thread has a different PID (instead of more properly appearing as the same PID, but with a different LWP ID), and inter-thread signal communication is not very quick. 2.6 handles this much better. So, when start-stop-daemon (or daemon) sends a kill signal, it ends up signalling the wrong thread, and it takes a long time for the signal to work. Hence the parent stays around for a long time, and the shutdown takes a long time to happen. The alternative is just to send a kill signal and move on, but then you have problems on restart if the old process hasn't died yet. The woody backport out there has my first hackish and totally non-portable attempt to work around this - a better and portable one is sitting on my development machine while I think about this some more. The new one borrows liberally from squid's shutdown script, and will tell you that it's still waiting for the process to go away, so it at least won't appear to 'hang', although it may still take some time. So, for Robert, can you do a couple of things for me? Let me know the output of: uname -a ps -el | egrep '(PID|clamav-milter)' And also try running the milter (and clamd - they're tied together unfortunately, in this) without daemon and let me know how a shutdown goes? To do this, stop clamd and the milter, comment out Foreground, and start them both again. Then try: time /etc/init.d/clamav-milter stop. Reply either on or off list - this is more of packaging issue than a clamav one, strictly, so I don't mind taking it to private email. Thanks, -- -- | Stephen Gran | Just because you're paranoid doesn't| | [EMAIL PROTECTED] | mean they AREN'T after you. | | http://www.lobefin.net/~steve | | -- pgpl6CiepROsG.pgp Description: PGP signature ___ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
