search

Azure / cyclecloud-slurm

Azure CycleCloud project to enable users to create, configure, and use Slurm HPC clusters.
MIT License
56 stars 42 forks source link

Install script assumes cluster name should be used for /sched/<clustername> config #277

Open anhoward opened 2 weeks ago

anhoward commented 2 weeks ago

When trying to use the install script on "external" nodes, it would be good to be able to specify the cluster name to use rather than imply it from the CC settings. This would help a lot with manual installation of Slurm using this script for things like login visualization nodes that might already have the mounts and everything else present.

I'd propose something like --clustername as an argument to the script.

xpillons commented 2 weeks ago

One scenario would be if I want to create a cluster of VDI login nodes which need slurm client to be installed. Here is what is happening today without this option. The VDI cluster doesn't run the scheduler hence the configuration path for slurm is wrong. image

aditigaur4 commented 2 weeks ago

May I ask why login visualization nodes cannot be created as regular login nodes through CC?

anhoward commented 2 weeks ago

There are times we want to install slurm on nodes that aren't part of the cluster directly (they may be used for other purposes or be on-prem). In this case it's to support nodes running Open OnDemand that still need slurm installed.

-Andy

From: Aditi Gaur @.> Sent: Tuesday, August 27, 2024 12:27 PM To: Azure/cyclecloud-slurm @.> Cc: Andy Howard @.>; Author @.> Subject: Re: [Azure/cyclecloud-slurm] Install script assumes cluster name should be used for /sched/ config (Issue #277)

May I ask why login visualization nodes cannot be created as regular login nodes through CC?

— Reply to this email directly, view it on GitHubhttps://github.com/Azure/cyclecloud-slurm/issues/277#issuecomment-2313018658, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AISV6UTAH6KGXZF26GFECBDZTSSFXAVCNFSM6AAAAABNEQ3D6SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMJTGAYTQNRVHA. You are receiving this because you authored the thread.Message ID: @.***>

aditigaur4 commented 2 weeks ago

And one more question-- the install scripts do call jetpack config command. Its how it infers the cluster names etc. So is jetpack not working on these nodes? Or have they created a minimally viable jetpack config? Coz if no and if i add this change, then it will likely break somewhere else.

anhoward commented 2 weeks ago

I think for now we can assume jetpack config works, but it's just part of a different cluster. I'll follow up with you offline about how this might work for non-CC nodes.

-Andy

From: Aditi Gaur @.> Sent: Tuesday, August 27, 2024 12:35 PM To: Azure/cyclecloud-slurm @.> Cc: Andy Howard @.>; Author @.> Subject: Re: [Azure/cyclecloud-slurm] Install script assumes cluster name should be used for /sched/ config (Issue #277)

And one more question-- the install scripts do call jetpack config command. Its how it infers the cluster names etc. So is jetpack not working on these nodes? Or have they created a minimally viable jetpack config? Coz if no and if i add this change, then it will likely break somewhere else.

— Reply to this email directly, view it on GitHubhttps://github.com/Azure/cyclecloud-slurm/issues/277#issuecomment-2313036194, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AISV6UWDC7DG25DAHFYCX5LZTSTGZAVCNFSM6AAAAABNEQ3D6SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMJTGAZTMMJZGQ. You are receiving this because you authored the thread.Message ID: @.***>