NFS Automount Hang

[John Riedel]

I just ran into a strange problem, and I didn't know if this is to expected or a bug/limitation.

I'm running on HPUX 11.0 with the Dec. 2000 patch bundle.

I have an automount map auto_data for mount points that start /data, it is an indiirect map.

The hang happens when you cd /data/x and you do a pwd. An ls works just fine. This only occurs on the server that /data/x points to. On any clients you don't have a problem. If you cd /disk5/data/x you have no problems.

We traced things back by messages in syslog to a typo in the auto_data map. If you have another map, such as /data/y and it points to a non-existant directory on the same server as /data/x. Then the hang will occur.

This happens even though nothing is trying to mount /data/y

Anyone know what's going on? I couldn't find any patches that talk about this type of problem.

We tend to create new mount points and push out the maps, then create the directories, this will trigger this hang. We've had, by mistake, maps pointing to the wrong place for years on 10.20 and have never seen this type of problem.

I've noticed that 11.0 tries to mount things on the same machine with LOFS, while in the past, it just dropped you in the real directory.. This may be an interaction between LOFS and autofs/automount

[Berlene Herren]

Have you tried automount tracing?

Two levels of automounter tracing are available:
Detaled (level 3) Includes traces of all automounter requests and replies, mount attempts, timeouts, and unmount attempts. You can start level 3 tracing while the
automounter is running. Basic (level 1) Includes traces of all automounter requests and
replies. You must restart the automounter to start level 1 tracing.
To Start and Stop Automounter Detailed Tracing
1. Log in as root to the NFS client.
2. Issue the following commands (PID is the process ID returned by the
ps command):
#ps -ef | grep automount
#kill -SIGUSR2 PID
Level 3 tracing is appended to the file /var/adm/automount.log.
To stop level 3 tracing, issue the same commands listed above to send the
SIGUSR2 signal to the automounter. The SIGUSR2 signal is a toggle that turns tracing on or off depending on its current state.
If you have basic (level 1) tracing turned on when you send the SIGUSR2 signal to the automounter, the SIGUSR2 signal turns tracing off.
To Start and Stop Automounter Basic Tracing
1. Log in as root to the NFS client.
2. Add ?2> tracefile? to the AUTO_OPTIONS variable in the
/etc/rc.config.d/nfsconf file, as in the following example:
AUTO_OPTIONS=?-f $AUTO_MASTER 2> /var/adm/automount.log?
This change redirects standard error to the file /var/adm/automount.log. Automounter basic trace output is sent to standard error.
3. Issue the following command to get a list of all the automounted directories on the client:
/usr/bin/grep tmp_mnt /etc/mnttab
4. For every automounted directory listed by the grep command, issue the following command to determine whether the directory is
currently in use:
/usr/sbin/fuser -cu local_mount_point
This command lists the process IDs and user names of everyone using the mounted directory.
5. Warn any users to cd out of the directory, and kill any processes that are using the directory, or wait until the processes terminate. You can issue the following command to kill all the processes using the mounted directory:
/usr/sbin/fuser -ck local_mount_point
6. Issue the following commands to kill the automounter (PID is the process ID returned by the ps command):
ps -ef | grep automount
kill -SIGTERM PID
CAUTION Do not kill the automounter with -SIGKILL (-9).
7. Issue the following command to start the automounter with tracing enabled:
/usr/sbin/automount options -T
options is the list of options configured in the AUTO_OPTIONS variable in the /etc/rc.config.d/nfsconf file. You can also source the /etc/rc.config.d/nfsconf file, and then enter the automount command as follows:
/usr/sbin/automount $AUTO_OPTIONS -T
To stop automounter logging, kill the automounter and restart it, start it without the -T option.


Berlene

[John Riedel]

We haven't tried tracing. It would require that we reproduce a condition that hangs all of our users, beacuse of an application that checks that status of some directories thru a server process. We don't have that many systems on 11.0 to do testing.