autofs Bug ?
A message that is not in the archive (sent off-list or not yet received): [email protected]
Hi, John
Quoted 11 lines
Quoted 10 lines · --- John Saalwaechter
Date: Fri, 26 Mar 2004 14:44:02 -0800 (PST)
From: John Saalwaechter <[email protected]>
Subject: [autofs] autofs Bug ?
To: [email protected]Roland, I'm very interested in your comments. I'm in an environment with a relatively sophisticated automounter setup, with Sun, SGI, HP-UX, and Linux machines. The Linux machines are mostly stock Red Hat 9, with its default autofs.
Quoted 10 lines
Quoted 10 lines
Occasionally the automounter on multiple Linux machines will completely stop functioning, which correlates to them being stuck in the "D" state in "ps ax" output. Attempts to simply stop and restart the automounters with "
/etc/init.d/autofsstop;/etc/init.d/autofsstart" always make the problem worse. After the restart the load on the Linux box will rise until it's high, and there will be multiple automount daemons for the same mount point. A full system reboot is the only solution we have right now for this situation. I also suspect that the affected systems are all multiprocessors.
My problem with Fedora Core 1 was a little bit different. The automount did not only stop functioning, but the whole machine was dead. Reset button was the only possible action.
But one of my tests showed a similar result as you wrote:
The following loop
while true
do ls ~user1 ~user2 /net/host1/usr >/dev/null ; sleep 1
pkill -USR2 automount ; sleep 1 ; /etc/init.d/autofs start ; sleep 1
done
accumulated more and more automount processes.
Quoted 5 lines
Quoted 5 lines
You mention sending the USR1 signal in the stop method, then pausing, then sending USR2. Why did you think to do that? Can you please post your modified
/etc/init.d/autofsscript, or at least the stop method? That would be greatly appreciated.
The original stop function in /etc/init.d/autofs of the autofs-4.1.0-1 RPM looks as follows:
--------------------------------
echo -n $"Stopping $prog:"
if [ -z "`pidofproc $prog`" -a -z "`getmounts`" ]; then
success $"$prog shutdown"
RETVAL=0
else
killproc $DAEMON -USR2 2> /dev/null
RETVAL=$?
fi
count=0
while [ -n "`/sbin/pidof $DAEMON`" -a $count -lt 10 ] ; do
killproc $DAEMON -USR2 2> /dev/null
RETVAL=$?
[ $RETVAL = 0 -a -z "`/sbin/pidof $DAEMON`" ] || sleep 3
count=`expr $count + 1`
done
umount -a -f -t autofs
rm -f /var/lock/subsys/autofs
echo--------------------------------
where the Variable DAEMON is set to /usr/sbin/automount ,
the Variable prog is set to automount , and the
function "killproc", defined in /etc/init.d/functions , sends the given signal to the named processes.
In the above stop function, immediately after sending the USR2 signal in the else-Branch, the automount process is still alive, because it takes a while to go down. And therefore, the while-loop begins working and sends a second USR2. This was the moment when the automount process crashed, especially when there were umountable (not busy) (Network-) filesystems mounted.
The workaround is simple and seems to help:
In the else-Branch above I replaced the USR2 by USR1 , and between "fi" and "count=0" I inserted "sleep 2". The effect should be, that all umountable filesystems are umounted before the automount process receives a USR2.
As said before, it seems to work.
Sincerely,
Roland Pfretzschner
---------------------------------------------------------------------------
Dr. Roland Pfretzschner [email protected]
Sysadmin
University of Technology Dresden
Departement of Mathematics
Mommsenstr. 13
01062 Dresden, Germany
Ian Kent
maintainerQuoted 6 lines · Roland Pfretzschner
In the above stop function, immediately after sending the USR2 signal in the else-Branch, the automount process is still alive, because it takes a while to go down. And therefore, the while-loop begins working and sends a second USR2. This was the moment when the automount process crashed, especially when there were umountable (not busy) (Network-) filesystems mounted.
Yes this was from the original autofs-3 RedHat package I took the patches from. It has been modified somewhat in 4.1.1 to reduce the signalling as it causes problems with umount. I put back some code I previously had. This may not be the problem you describe but please try 4.1.1 or later and let me know how you go.
Ian
Ian Kent
maintainerQuoted 3 lines · Roland Pfretzschner
Quoted 3 lines
Quoted 3 lines
make the problem worse. After the restart the load on the Linux box will rise until it's high, and there will be multiple automount daemons for the same mount point. A full system
Oh as well.
Jim Carter identified a possible infinite expire loop. I'm not sure that this is what is happening here but I applied his work around in 4.1.1 also.
Please try this out.
Ian