Ubuntu HPC NFS client setup on compute nodes

From Notes_Wiki

Home > Ubuntu > Ubuntu HPC setup with slurm and linux containers > Ubuntu HPC NFS client setup on compute nodes

Mounting NFS on Cluster Nodes

This section explains how to mount the shared /export directory on all compute nodes, login node, and the master node in a Slurm cluster.

1. For Compute Nodes (Bare Metal or VMs)

a. Install NFS Client Packages

 apt install -y nfs-common 

b. Create Mount Point

 mkdir /export 

c. Verify NFS Export from Master

 showmount -e master # OR showmount -e master-ib 

d. Configure Persistent Mount in /etc/fstab

Add the following line to /etc/fstab:

 <master-node-name>:/export /export nfs defaults,nofail,soft,bg 0 0 

Notes:

  • Use nofail to allow booting even if NFS is not available.
  • Use soft and bg to ensure mounting happens in the background and retries without blocking the boot.
  • If using InfiniBand, replace <master-node-name> with the corresponding InfiniBand hostname (e.g., master-ib).
  • You can optionally add the async mount option to improve performance (at the cost of consistency), depending on your HPC workload requirements.

e. Mount All Entries

 mount -a 

2. For Containers (slurm-login1, slurm-master) on Infra Node

Since the login node and master node are LXD containers running on the infra (NFS) node, we do not use NFS clients to mount /export. Instead, use LXD disk device passthrough from the host (infra node):

Steps (Run from Infra Node):

a. Enable Privileged Container

 lxc config set slurm-login1 security.privileged true 

b. Restart Container

 lxc stop slurm-login1 lxc start slurm-login1 

c. Attach /export/home to Container

 lxc config device add slurm-login1 export_home disk source=/export/home path=/export/home 

Repeat similar steps for slurm-master and any other container that needs access to /export/home.

Note:

  • This ensures that containers running on the infra node share the same /export/home directory via direct host path mapping instead of NFS mount.
  • This method is faster and avoids networking overhead between containers and host.

Home > Ubuntu > Ubuntu HPC setup with slurm and linux containers > Ubuntu HPC NFS client setup on compute nodes