Closed garymansellricardo closed 5 months ago
I have confirmed this same behaviour on Azure CycleCloud 8.5 / Slurm template 3.0.5
I can also confirm that this is existing on Slurm Cluster v.23.02.4-1, no matter which HPC Image I choose
we have fixed in cyclecloud 8.6 and azslurm 3.0.6 versions.
Hi,
I am using the default Slurm cluster config template from version 3.0.1 - and it seems to create a 2GB Azure managed disk for the shared filesystem on the scheduler node even though I am mounting /shared from an NFS file server (the cyclecloud server). I think this should not be needed?
Here is a screenshot showing the disk in Azure:
Here, you can see the disk at the Linux OS (as sdb and no mount point):
[root@nemospot-scheduler ~]# lsblk -o NAME,HCTL,SIZE,MOUNTPOINT | grep -i "sd" sda 0:0:0:0 30G sdb 0:0:0:1 2G sdc 1:0:0:0 60G ├─sdc1 1000M /boot ├─sdc2 58.5G / ├─sdc14 4M └─sdc15 495M /boot/efi sdd 1:0:0:1 75G └─sdd1 75G /mnt
Here, you can see /shared mounted from the CycleCloud server via NFS:
[root@nemospot-scheduler ~]# mount | grep shared azu-nemo-srv:/shared on /shared type nfs4 (rw,relatime,vers=4.1,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,clientaddr=192.168.140.132,local_lock=none,addr=192.168.140.20)
Would be good not to incur unnecessary Azure costs...