Fix empty nodelist in ondemand apps

[sjpb]

When topology feature is enabled, slurm does not tolerate the default empty --nodelist= parameter for the desktop and rstudio apps. This fixes it by making it conditional.

Also turns on topology in CI to try to catch any future similar issues.

Failure was e.g in login node's /var/log/ondemand-nginx/demo_user/error.log:

App 17541 output: [2025-12-09 11:54:29 +0000 ]  INFO "execve = [{}, \"sbatch\", \"-D\", \"/home/demo_user/ondemand/data/sys/dashboard/batch_connect/sys/bc_desktop/slurm/output/8dc04023-7393-4a92-918f-06e876446f38\", \"-J\", \"ood-desktop\", \"-o\", \"/home/demo_user/ondemand/data/sys/dashboard/batch_connect/sys/bc_desktop/slurm/output/8dc04023-7393-4a92-918f-06e876446f38/output.log\", \"-p\", \"cclake_himem\", \"-t\", \"01:00:00\", \"--export\", \"NONE\", \"--nodes=1\", \"--ntasks=1\", \"--nodelist=\", \"--parsable\", \"-M\", \"stg\"]"
App 17541 output: [2025-12-09 11:54:29 +0000 ] ERROR "ERROR: OodCore::JobAdapterError - sbatch: error: Batch job submission failed: Requested node configuration is not available"

[elelaysh]

Why don't rstudio, matlab and codeserver have the --node arg in the launch script?

[sjpb]

Why don't rstudio, matlab and codeserver have the --node arg in the launch script?

Because the example apps with those didn't add those I think. Its arguable whether all or none should have it TBH! I can sort-of see that for some cases you might want to land on a specific node, but I'm not sure any of these apps are one such case.

[elelaysh]

Fix confirmed on an appliance instance.
Indeed the issue is not present if the topology plugin is not active.

[elelaysh]

Why don't rstudio, matlab and codeserver have the --node arg in the launch script?

Because the example apps with those didn't add those I think. Its arguable whether all or none should have it TBH! I can sort-of see that for some cases you might want to land on a specific node, but I'm not sure any of these apps are one such case.

I would like to see unified general parameters (cores, mem, gpus, nodes, ...) for all apps, because of DRY principle + you never know: a user might want to have codeserver to gdb attach to a running program, ...