TKFE Deployed Cluster fails to initialize slurm about hpc-toolkit HOT 6 CLOSED

Hi Matthew,

I have contacted NAG about this issue. May I ask you what version of the toolkit you are using? are you using the latest in main or develop?

Thanks!

from hpc-toolkit.

Hi Carlos,

I am using the latest in main.

Thanks for the help in advance.

from hpc-toolkit.

Hi Matthew,

I deployed both the the version in main and the version in develop and was unable to reproduce the issue. In both cases the controller exposed the mount points correctly.

So I am wondering if this is not a network configuration issue.

• Could you please let me know if you default network allows connections between the nodes?
• If you create two VMs in the default network, can they ping each other?
• Alternatively, could you try creating a new vpc and new subnetwork and test if the cluster provisions normally in that case?

Thanks!

from hpc-toolkit.

Hi Carlos,

I tried again on a new VPC and it worked. The default VPC did allow nodes on the same subnet to ping each other though... At any rate, thanks for the help! Just a quick question, are cloud filestore instances supposed to be created in a default cluster configuration? Or does it just use buckets?

Thanks,
Matthew

from hpc-toolkit.

Hi Matthew,

Glad it worked!
I don't think either filestore or buckets are used by default. The nfs shares are exported from the slurm controller and used in the cluster, unless you manually create a filestore and use that mount point during the cluster creation.

cheers,

Carlos

from hpc-toolkit.

I will now mark this issue as closed. Please feel free to reopen it if you feel this was not sufficiently addressed.

from hpc-toolkit.