Deploy Bundle Edge Version

[moula]

Bonjour,
I tried to deploy the edge version of the bundle on my data-center. Everything installs except slurmrestd. Thank's.

[jaimesouza]

Hello @moula! Thanks for contacting us! We are working to fix that.

[jaimesouza]

@moula could you try again, please? You can also try the latest on the candidate channel (--channel candidate).

[jaimesouza]

We have fixed the issue and released a new revision of the slurm charms to edge. Thank you for trying the slurm charms! Let us know if you need any help.

[moula]

@jaimesouza I'm trying again, I'll get back to you as soon as it's done.

[moula]

@jaimesouza it works, but during the deployment I had to reboot the machines manually which is not the same thing with version 8.5. I will add Nvidia GPUs and monitoring tomorrow in order to test it in use. Merci.

[jamesbeedy]

Hello @moula! Thanks for stopping by!

We have another charm called, nvidia-gpu. We will be deprecating the built-in charm infiniband and gpu actions soon and replacing them with individual charms.

To add cuda drivers to your deployment, you could deploy and relate the nvidia-gpu charm to the nodes where your gpus are.

juju deploy nvidia-gpu --channel edge
juju relate nvidia-gpu slurmd

[moula]

@jamesbeedy Thank you very mauch.
I am testing it. If you have other howtos or videos for example for integration with COS or testing code with GPU, I'm interested in testing all that. A note: is the migration of percona-cluster planned?
I know this has been done before : canonical/slurmdbd-operator#9

[jamesbeedy]

Hey @moula,

We will be collaborating with @NucciTheBoss and @dvdgomez from the HPC team at Canonical for the next amount of time to revise/refactor the Slurm charms. Any changes you see in the Canonical/slurm*-operator forks will eventually end up getting PR’d into the omnivector-solutions/slurm*-operator repos.

[NucciTheBoss]

Hi there @moula! Yes, I have the migration completed and plan on submitting for upstream to the Omnivector folks soon. I am also tackling the versioning issue mentioned in charmed-hpc/slurmdbd-operator#5, which is where I think you received the original install error that Jamie and James fixed.

[moula]

Bonjour @NucciTheBoss Yes You are all doing a good job. Thank you so much.

[moula]

@jamesbeedy Thank you.
For the installation of GPU drivers, I got the message "Machine needs reboot". I rebooted 5 times each node



for it to work!!!.

[jamesbeedy]

Hey @moula , It takes a few minutes following a reboot for that message to disappear. We can try and clean up the messaging around rebooting by making a shorter period in between polling to see if the machine needs a reboot. Most likely your first reboot worked 🙂

[moula]

@jamesbeedy keep it up . Thank's.