Can't access/use schedmd public images at all

[Mgcini Keith Phuthi]

Hi,

I am a newbie and I've been struggling with this for a while. I can't seem to be able to use any of the images that were provided by schedmd but I could a few months ago. For example if I run:

gcloud compute instances create slurm-hpc-instance \
        --zone=us-central1-a \
        --image-family=schedmd-slurm-21-08-5-hpc-centos-7 \
        --image-project=schedmd-slurm-public \
        --maintenance-policy=TERMINATE \
        --machine-type=c2d-standard-8 --boot-disk-size=200GB

I get the following error for any image from the schedmd project:

ERROR: (gcloud.compute.instances.create) Could not fetch resource:
 - Required 'compute.images.useReadOnly' permission for 'projects/schedmd-slurm-public'

This happens when I try and deploy a cluster with Terraform or use packer as well it only works when I deploy with the default example.tfvars but I want to use a different image based on hpc-centos-7. Is there some permission/role I'm missing? Are the images no longer available publicly?

I checked my policies and I should be allowed to use any public images.

Thanks,

Keith

[Wyatt Gorman]

Hi Keith,

Can you please make sure you're using the latest version of the scripts from Github? There is no "schedmd-slurm-21-08-5-hpc-centos-7" image family, so we would suggest using the latest image. The latest image in the V4 scripts was specified as "projects/schedmd-slurm-public/global/images/family/schedmd-slurm-21-08-8-hpc-centos-7", can you try that image instead? You mention that you want to use a different image, can you clarify what other image you want to use, and why?

Thanks,

Wyatt Gorman HPC Solutions Manager https://cloud.google.com/hpc

--
You received this message because you are subscribed to the Google Groups "google-cloud-slurm-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to google-cloud-slurm-...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/google-cloud-slurm-discuss/806a6ccc-a10d-4e0e-adef-079a468dccb1n%40googlegroups.com.

[Mgcini Keith Phuthi]

Hi Wyatt,

Thanks! I was going off of past advice that said I should use whatever slurm version is on the schedmd website and I hadn't noticed the version in the github repo. Changing the version fixed the issue. By different I just meant a modified where I install some software on the image before deploying the cluster, which is why I wanted to spin up a VM first. I'll try and play around with Packer as well.

Best!

Keith

[Tom Downes]

Keith-

You might consider following the image building example in the Cloud HPC Toolkit. It is built around the latest v4 image and Packer:

https://github.com/GoogleCloudPlatform/hpc-toolkit/tree/main/examples#image-builderyaml-

The Toolkit is a new product from Google intending to simplify HPC deployments, including Slurm, with open source tools such as Terraform, Packer, and Ansible. The startup script in the example is purposely simple but can be extended using bash or Ansible playbooks. It may be helpful to look through the examples/ and community/examples directory of the repo, in addition to the modules/README.md.

Tom Downes Software Engineer, High Performance Computing +1-331-625-1145 210 N Carpenter St, Chicago, IL 60607

To view this discussion on the web visit https://groups.google.com/d/msgid/google-cloud-slurm-discuss/2191eb87-e1e2-4234-ad51-8af9e7306807n%40googlegroups.com.