Unable to read msg box when job attempts to run on node

Users/Site Administrators
1

Hello,

I am getting the following log message when attempting to submit new jobs. The job attempts to run but then gets put into a Held state. All nodes have AD authentication, so we do not use the /etc/passwd file for users.

pbs_mom(Thread 0);tpp_mbox_read;Unable to read from msg box
05/20/2024 14:54:54;0028;pbs_mom;Job;8.wwadm01;No Password Entry for User ********

Any suggestions?

2

Can you please try to SSH as that “user” from PBS server host to compute node host and then back ? Whether this works ?

3

We do not allow ssh to compute nodes as regular users. Only when jobs owned by that user are running on a compute node are they supposed to be allowed to ssh in.

Even as root I am unable to run an ID lookup of a user in our AD. I have the gateway set on the compute nodes to go through the head to go out to the domain controllers, but still nothing.

Any ideas in terms of networking I should check?

4

To make sure user can ssh into the compute node and back , it is better to disable the $restrict_user setting for one experiment to make sure user can ssh successfully.

Reference: Linux Error “No passwd entry for user” | Baeldung on Linux

As root user ssh to the compute node and then do su - whether it works

5

I am able to freely ssh between nodes as a regular user. Jobs submit now and finish but during the exit process I still get the “Unable to read msg box” error

I am also getting the following sys_copy failed, return value=1 with a connection refused when the compute nodes attempt to copy back to the login node/submit host. There is currently no firewall on the login node, and I am able to ping and ssh freely between nodes. Any ideas?

6

copy back will preserve permissions (group, user , other), so that might be one of the caveats
you can increase the mom log debug level and , retest with a sample job and check the mom logs. There might be more clues.

8

So the problem ended up being we needed to declare which paths should be allowed to copy in the mom config file on the compute nodes.

Editing the /var/spool/pbs/mom_priv/config file did the trick.

Thanks for the support