torque-server init script fails during installation and removal

The attached patch fixes this problem, plus another couple of packaging issues that was addressed after the initial upload to the NEW queue.

motu-sru, please allow an SRU for hardy for the torque package.

Patch incorporating modifications according to discussion with Luca Falavigna on IRC

ACK from motu-sru.
Please, do not ship linda override in the upload, it is just a cosmetic change.
Also, provide a good TEST CASE to speed up verification process. Thanks!

Why did you remove the pid file check? There is no rationale for this in the changelog, nor a separate bug report about this. PID files are a good thing to not break processes in chroots, or processes started as different users, etc.

> Why did you remove the pid file check? There is no rationale for
> this in
> the changelog, nor a separate bug report about this. PID files are a
> good thing to not break processes in chroots, or processes started as
> different users, etc.

Yes, that's how I thought the --pidfile option worked too. However,
the purpose of this option is to specify a PID file created by the
server process itself, and because pbs_server does not do that, it is
not relevant in this case.

Cheers,
Morten

Accepted into -proposed, please test and give feedback here

jd, please test the updated package in hardy-proposed and report back here!

Cheers,
Morten

Unfortunately, the --pidfile options are still present in this release, and they cause the initscript to break.

Also, /var/lib/torque/mom_priv/jobs, /var/lib/torque/spool and /var/lib/torque/undelivered need to be added to the dirs for torque-mom. (/var/lib/torque/spool and /var/lib/torque/undelivered need to have 1777 permissions)

Looking in more detail, they are gone from the server init script, but not the mom or scheduler one.

Also, the mom won't run properly unless it has a config file in /var/lib/torque/mom_priv/config. It can be empty, but it has to be present.

Finally, the postinst script should create the database, if it doesn't exist yet. Here's a sample script to do it:
/usr/sbin/pbs_server -t create && sleep 2 && /usr/bin/qterm

Martin, please release the version in hardy-proposed into hardy-updates.

Its taken me a while to get back to this, but i'm having similar issues with the package from proposed. I've taken action on each of the comments by Ted, removing the --pidfile arguments, creating the database, running `touch /var/lib/torque/mom_priv/config` and changing permissions to 1777 on both /var/lib/torque/spool and /var/lib/torque/undelivered.

First i tried simply to install the -proposed version and it failed, so then I made the changes above and tried a reinstallation and it continues to fail. It would seem though that I can start each of the services, mom, scheduler and server and see them running is ps -ef. And since i'm brand new to this world i wanted to install the torque-gui, which seems to install, but i can't find where to launch it from (which is a side question).

In conclusion, I'd say the -proposed does not fix the bug at hand, even though i *may* have a working torque server

If there is anything else I can provide, please ask, I'd really like to use this.

All crash files added

After changing /var/lib/torque/server_name from torqueserver to the name of my machine, I am able to run qmgr and configure the server, as well as successfully configure nodes and have them display as 'free' via pbsnodes -a.

So it looks like i do have a working server and clients.

Thanks for the very useful comments. I will fix these issues and make another upload shortly.

I have looked here to find a solution to the problem related to the bug, synthetically about just removing the torque (especially torque-mom, torque-scheduler and torque-server). The working solution is that stated in first post "This can only be fixed by editing the torque-server init script, so it always returns 0." that translates in putting a line with "exit 0" before line "This can only be fixed by editing the torque-server init script, so it always returns 0." in /etc/init.d/torque-mom, /etc/init.d/torque-scheduler and /etc/init.d/torque-scheduler.

I have looked here to find a solution to the problem related to the bug, synthetically about just removing the torque (especially torque-mom, torque-scheduler and torque-server). The working solution is that stated in first post "This can only be fixed by editing the torque-server init script, so it always returns 0." that translates in putting a line with "exit 0" before line "This can only be fixed by editing the torque-server init script, so it always returns 0." in /etc/init.d/torque-mom, /etc/init.d/torque-scheduler and /etc/init.d/torque-scheduler.

I can't get any of this to work and I would like to just remove it by hand as it interferes with just about everything I try to install now. Apt-get remove doesn't work so can anyone offer some hints to get rid of it. Ta

alidaf: To remove the packages, add the line "exit 0" to the top (like 3 or so is fine) of each of the torque scripts in /etc/init.d, as Giovanni describes above. Then you can remove/purge the packages with dpkg.

Ping? This package is a catastrophe.

It is pretty bad, but once you have it installed, it does work properly. What problems are you having?

I still get this error with the version in hardy-proposed:

Setting up torque-server (2.1.8+dfsg-0ubuntu1.1) ...
 * Starting Torque batch queue server: PBS_Server: No such file or directory (2) in chk_file_sec, Security violation with "/var/lib/torque/spool/"
PBS_Server: No such file or directory (2) in chk_file_sec, Security violation with "/var/lib/torque/pbs_environment"
PBS_Server: PBS_Server, pbsd_init failed
invoke-rc.d: initscript torque-server, action "start" failed.
dpkg: error processing torque-server (--configure):
 subprocess post-installation script returned error exit status 3
Errors were encountered while processing:
 torque-server

A fix would really be appreciated. This makes the torque packages unusable.

If you want to fix it yourself, just create the following directories before you install the package:
/var/lib/torque
/var/lib/torque/server_priv
/var/lib/torque/server_priv/jobs
/var/lib/torque/server_priv/queues
/var/lib/torque/server_priv/accounting.

On clients, you need the following:
/var/lib/torque/mom_priv
/var/lib/torque/mom_priv/jobs
/var/lib/torque/spool (with permissions 1777)
/var/lib/torque/undelivered (with permissions 1777)

Also, you'll need a /var/lib/torque/server_name file on your servers and clients listing the torque server DNS name.

That's everything I have listed in my puppet config. Ideally all of these should be put in the package.

This bug was fixed in the package torque - 2.1.8+dfsg-0ubuntu3

---------------
torque (2.1.8+dfsg-0ubuntu3) intrepid; urgency=low

  * Build for intrepid.

torque (2.1.8+dfsg-0ubuntu1.2) hardy-proposed; urgency=low

  * Fix remaining issues with init scripts (LP: #223649)
  * Add missing empty directories to package torque-client

 -- Morten Kjeldgaard <email address hidden> Wed, 27 Aug 2008 21:48:57 +0200

Accepted into -proposed, please test and give feedback here. Please see https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

The latest torque packages in hardy-proposed work for me.

Thanks!

There's one small remaining problem with the torque-server init script: it's not idempotent. I found this out when running a "dpkg --configure torque-server" (to finish configuring the package) while the server was running.

I wasn't sure what was best, so I filed a separate bug report about this: https://bugs.launchpad.net/ubuntu/+source/torque/+bug/270574

There's one issue with the packages in hardy-proposed. The following directories:

/var/lib/torque/mom_priv
/var/lib/torque/mom_priv/jobs

should be in the torque-mom package, not in the torque-client one. As is it, it makes torque-mom broken unless torque-client is also installed on the same machine (which is not always the case).

Another issue. The /var/lib/torque/spool and /var/lib/torque/undelivered directories are not created with mode 1777 (see comment 21), which make submitted jobs fail when torque tries to run them on the compute note.

Yet another reason why the hardy-proposed package is not ready for prime time: the directory /var/lib/torque/server_priv/acl_svr is missing from the torque-server package. Without this directory, the operators and managers settings of the torque-server are not preserved across executions.

Three other directories missing (all in /var/lib/torque/server_priv):

acl_hosts
acl_users
acl_groups

Marking verification-failed for now, this will probably require a new fix for Intrepid too.

Reopening for all releases, since this still isn't fixed. I removed the SRU from hardy-proposed, since it failed verification.

This bug was fixed in the package torque - 2.3.6+dfsg-0ubuntu1

---------------
torque (2.3.6+dfsg-0ubuntu1) jaunty; urgency=low

  * New upstream release (LP: #235385).
  * Torque-server init script fails during installation and
    removal (LP: #223649) fixed, "set -e" removed from init script.
  * Package torque-scheduler 2.1.8+dfsg-0ubuntu1.1 failed to install
    (LP: #244440 LP: #270574) fixed, "set -e" removed from init script.
  * /etc/init.d/torque-mom not idempotent, and stop doesn't work
    (LP: #256998) Fixed, "set -e" removed from init script.
  * Torque-scheduler prints errors during package configuration
    (LP: #270653). Reason: missing dir sched_config. Fixed,
    package installs FIFO scheduler config file in
    /var/lib/torque/sched_priv/.
  * Package torque-mom 2.1.8+dfsg-0ubuntu1 failed to
    install/upgrade (LP: #276575 LP: #291674). Reason: missing directories.
    Fixed, /var/lib/torque/server_priv/jobs/,
    /var/lib/torque/server_priv/queues/,
    /var/lib/torque/server_priv/accounting/ and
    /var/lib/torque/mom_priv/jobs/ are now installed in their respective
    packages.
  * Package torque-gui missing most of the files it needs to run!
    (LP: #281360). Reason: missing *.tk etc. files from src/gui
    Fixed: /usr/lib/xpbs now shipped in package
  * changed patch system to quilt.

 -- Morten Kjeldgaard <email address hidden> Mon, 16 Feb 2009 17:32:28 +0100

Hello

since I have a lot of problems with Kubuntu releases > 8.04, I want to stick with 8.04 for the moment. Where can I find a torque-server deb which works?

I tried to read all the posts and it seems there have been a release in hardy-proposed but
I can't find it in updates.

thanks

Uwe Brauer

Intrepid Ibex reached end-of-life on 30 April 2010 so I am closing the report. The bug has been fixed in newer releases of Ubuntu.

Hardy has seen the end of its life and is no longer receiving any updates. Marking the Hardy task for this ticket as "Won't Fix".