Ticket 22034

Summary:	X11 Forwarding Not Working
Product:	Slurm	Reporter:	Aaron Jezghani <ajezghani3>
Component:	Other	Assignee:	Connor <connor>
Status:	OPEN ---	QA Contact:
Severity:	3 - Medium Impact
Priority:	---	CC:	mcmullan
Version:	24.11.1
Hardware:	Linux
OS:	Linux
Site:	GA Tech	Alineos Sites:	---
Atos/Eviden Sites:	---	Confidential Site:	---
Coreweave sites:	---	Cray Sites:	---
DS9 clusters:	---	Google sites:	---
HPCnow Sites:	---	HPE Sites:	---
IBM Sites:	---	NOAA SIte:	---
NoveTech Sites:	---	Nvidia HWinf-CS Sites:	---
OCF Sites:	---	Recursion Pharma Sites:	---
SFW Sites:	---	SNIC sites:	---
Tzag Elita Sites:	---	Linux Distro:	---
Machine Name:		CLE Version:
Version Fixed:		Target Release:	---
DevPrio:	---	Emory-Cloud Sites:	---

Description Aaron Jezghani 2025-02-10 06:02:49 MST

Following our recent adoption of job_container/tmpfs plugin and upgrade to Slurm 24.11.x (we're on 24.11.1 now, but problem observed in 24.11.0 as well), we're seeing issues with X11 forwarding.

In the configuration, we have both `PrologFlags=Alloc,Contain,X11` and `X11Parameters=home_xauthority` set. Direct SSH to the node allows X11 forwarding as expected, but running under an interactive job fails.

Checking things like $DISPLAY and `xauth list` present identical information between the manual SSH and interactive job launch, so as far as I can tell, things look to be landing properly.

I have noticed that occasionally, an error will pop up when trying to launch an X11-based application: 'salloc: error: _half_duplex: wrote -1 of 1748' - unsure if this is telling or a red herring, though.

Please let me know if additional information is required.

Comment 1 Connor 2025-02-11 10:40:05 MST

This looks similar to an ongoing issue we have. We are actively working on a patch to handle x11 connections better.

Aside from reverting the change in ecfc7f6ff7 we don't have a work around at this time unfortunately.

-Connor

Comment 2 Aaron Jezghani 2025-02-12 17:26:17 MST

To test this change, which nodes need the update - e.g., just compute or head node, or is the library part of the entire pipeline?

Comment 3 Connor 2025-02-13 07:03:01 MST

It would be the compute nodes and where the the slurm commands are run. "salloc" and "srun" will pull in this library and as well as the slurmd/slurmstepd services.


-Connor

Comment 4 Connor 2025-04-08 13:20:06 MDT

Hey Aaron,


Just providing you with an update: we have a patch that is currently in review. Hopefully we'll have it queued for a release soon.

-Connor

Comment 5 Aaron Jezghani 2025-04-08 13:38:08 MDT

Connor - 

Thanks for the update! Looking forward to it.

Aaron