About VNC session

Hello,

I have a problem where VNC session launches only after I do ssh to slurm compute node manually or if I do pre-login in slurm compute node (GUI) already phyisally in the server room. I am using mate desktop session in slurm compute node with ubuntu 24.04. I tried with a lot of options with batch job script to automate the launch of the vnc session but nothing works. As I am struggling more than a month with this, would be greatful I am if someone help me with this.

Thanks in advance!

Hello,
Is there anyone to help me with this?

Hi sorry for the delay. IIRC this has something to do with systemd’s loginctl enable-linger functionality. Searching for linger on this site seems to confirm this as well as a google search of “slurm vnc launches only after ssh login

Something like this in prologue

loginctl enable-linger $ SLURM_JOB_USER

With something like this in the epilogue

loginctl disable-linger $SLURM_JOB_USER

Hi @jeff.ohrstrom,

Many thanks for your kind repsonse! I jus had a look at linger and got to understand how it works with vnc. But there’s another problem which I didn’t added in above details. Even thogh I enbale loginctl enable-linger for a specific user like below (an example);

$ loginctl enable-linger harjad

But then inside the vnc session, the web browser (Firefox, Chromium & Steam) is not working. Is this something related to desktop session or with remote enable loginctl? Cause thes ework well when I already login into the mate desktop (GUI) physically on a server. Do you have any suggestion or recommendaiton for this? This would be very helpful to go forward with my work as I was hanged up with this as well.

Thanks in advance!

I’m not 100% sure, but I’d suggest you start them from the command line to see the stdout output to see what the error could be.

Thanks again @jeff.ohrstrom

I already tried running firefox from Mate terminal where it throws the same error as below;


failed to load module "appmenu-gtk-module" No protocol specified Error: cannot open display: :4


But gtk moule already exists and the display is also open at :4 I don understand why this throws same error again and again. But this error doesn’t appear and also it works fine when I already login pysically in GUI node (slurm compute node). I tried my best to resolve this error but nothing works out. If you have any suggestion then please let me know!

Thanks again!

I’m not sure what could be the issue. Can you share the output.log and/or vnc.log from the job? That may give us some insight.

Thanks again @jeff.ohrstrom
I have enclosed the vnc and output logs here as txt files here. I don know whether is this a correct way to share the files or not. if not, please let me know the exact format which you feel comfortable. Your inputs are indedd;

Output_log.txt (23.1 KB)

VNC_log.txt (9.0 KB).

Thanks again!

Hi @jeff.ohrstrom Hope you are doing well.

Any luck from the log file which I have shared?

Thanks in advance!

Sorry I’m just looking at them now. There are many errors in your output_log.txt that seem concerning to me. Unfortunately though, I’m not entirely sure what any of them mean. My very rough understanding seems to suggests that there’s some issue with your DBUS session and things being able to connect to it, though what that could be exactly I’m sorry I just don’t know.

@jeff.ohrstrom , we are looking at this live at ISC, and it appears to me to be an X issue of some sort. The OOD stuff all looks very similar to our production setup, and I think the enable-linger/disable-linger stuff is addressing a symptom but not the cause. Unfortunately, I’m not very versed in the X11 startup/config stuff, but given that a user has to be logged in on the physical node this points at something about the non-interactive X configuration. We did get a failure to open a display error when trying to launch firefox in the terminal. (“Error: cannot open display: :8”). If I try to run Chromium, it says “Missing X server or $DISPLAY.” despite $DISPLAY being set. So I think there is some kind of X or environment issue.

Yea it seems to me that DBUS isn’t starting right or things can’t communicate to X through DBUS.

The output logs seems to confirm this - lots of programs can’t figure out the session from DBUS. Though what could be causing it, I’m not sure.