[Bug]: A default of 0 GPUs should be possible

[marianne013]

Search before creating an issue

• I have searched existing issues and confirmed this is not a duplicate

Bug Description

We have set/Resources/Computing/CEDefaults/NumberOfGPUsin the Configuration, as we have a maximum of one or two test queues with GPUs (and none on the pre-prod servers).
This results in the following errors:

2026-04-13T15:02:36,534211Z WorkloadManagement/SiteDirectorLz INFO: Getting GPUs from /Resources/Computing/CEDefaults/NumberOfGPUs
2026-04-13T15:02:36,534314Z WorkloadManagement/SiteDirectorLz ERROR: Could not find NumberOfGPUs: missing siteName or gridCE or queue. Returning '0'

Steps to Reproduce

Install DIRAC v9.0.20 and set /Resources/Computing/CEDefaults/NumberOfGPUs = 0
Error appears in SiteDirector log.

Expected Behavior

A sensible default should be possible without triggering an error.

Actual Behavior

See above

Environment

Vanilla DIRAC 9.0.20

Relevant Log Output

See above

Additional Context

Simon says: The code logic looks like the default option will always be picked if the default value is not set to zero. And it it is set to zero, it triggers an error. Would it not make more sense to check the specific cases first and then only use the default if nothing specific found ?

[fstagni]

I am not sure I understand: /Resources/Computing/CEDefaults/ is for the... defaults, or for what the pilot fills in at runtime (in the pilot.cfg). I do not think you should specifically set this one.

If you have a queue with a GPU, you should set something like Resources/Sites/DIRAC/CEs/myGridCE.test.uk/Queues/gpuss/NumberOfGPUS=2

[marianne013]

Hi Federico,
When I search for "NumberOfGPUs" I get one mention:
https://dirac.diracgrid.org/en/latest/CodeDocumentation/WorkloadManagementSystem/Utilities/JobParameters.html#DIRAC.WorkloadManagementSystem.Utilities.JobParameters.getNumberOfGPUs
and the accompanying text is not clear. (If not set return 0, but can't be set to zero ?). Also this piece of documentation talks about "GPUs" when it might or might not mean NumberOfGPUs. Maybe this bug should be amended to a documentation issue ?
Related to this after removing /Resources/Computing/CEDefaults/NumberOfGPUs = 0 I am still left with
WorkloadManagement/SiteDirectorGridpp WARN: Could not find AvailableRAM: missing siteName or gridCE or queue. Returning 0 in my SiteDirector logs.
"AvailableRAM" returns no results when I search the DIRAC documentation :-(
Tagging @sfayer so he doesn't need to subscribe to this ticket.
(Fixed a typo: peace != piece, phonetically spelled languages for the win)

[fstagni]

Seems like it's just a documentation issue. Did you put somewhere something like Resources/Sites/DIRAC/CEs/myGridCE.test.uk/Queues/gpuss/NumberOfGPUS=2 ? Does it work?

The warning is also harmless. The CS parameter is called "MaxRAM". Feel free to amend the documentation.

[marianne013]

Shouldn't whoever implemented the AvailableRAM feature document it ? It would be nice to be able to find the documentation without having to do a git search on the code :-S

[fstagni]

The memory limits for the CEs are documented in https://dirac.diracgrid.org/en/latest/AdministratorGuide/Resources/computingelements.html . MemoryLimitMB is what you are looking for, IIUC.

I have added these 2 to the config example in #8519, let me know if this is not what you were looking for.