Using Podman with HPC Client
Using Podman with the HPC Client is simple, but takes some consideration.
Configuration
HPC Client Configuration
With the below changes in place, the only thing that needs to change in the HPC Client configuration is by setting the following line in the settings/credentials.sh file:
Software Requirements
As mentioned in the system requirements, the following software has been tested on Ubuntu 22.04, Jammy Jellyfish:
$ apt list --installed crun containernetworking-plugins podman-plugins \
podman-docker slirp4netns podman
containernetworking-plugins/jammy,now 0.9.1+ds1-1 amd64 [installed]
crun/unknown,now 101:1.14.4-0ubuntu22.04+obs70.5 amd64 [installed]
podman-docker/unknown,now 4:4.6.2-0ubuntu22.04+obs81.12 all [installed]
podman-plugins/unknown,now 4:4.6.2-0ubuntu22.04+obs81.12 amd64 [installed]
podman/unknown,now 4:4.6.2-0ubuntu22.04+obs81.12 amd64 [installed]
slirp4netns/unknown,now 100:1.1.8-4 amd64 [installed]
Ensuring these libraries are installed on the nodes of the Slurm cluster will allow the engine to execute gears. If they are not installed, consult with your system administrator to facilitate their installation.
GPU Requirements
Executing gears on GPU nodes requires the HPC Client and Slurm to be configured to schedule those nodes (see Using GPUs on a Slurm Cluster).
Furthermore, GPU execution has been successfully performed with the nvidia driver below
$ apt list --installed nvidia-driver-545
nvidia-driver-545/jammy-updates,now 545.29.06-0ubuntu0.22.04.2 amd64 [installed]
Engine Requirements
The Flywheel Engine must be updated to enable both Podman execution with or without GPUs. Please request an engine with these features from Flywheel staff and install according to the Updating the Flywheel Compute Engine document.