Closed asnaylor closed 10 months ago
Thanks @asnaylor. I think the easiest fix is to add this to default_pull_args
.
The only question is whether using fuse-overlays rather than native adds a lot of slowdown on logins. If that's the case, we could add some logic for podman-hpc to determine if it's on a compute or login, although that would make things a bit messy. I'll do some testing.
Should be addressed by https://github.com/NERSC/podman-hpc/pull/82
Was trying to pull
nvcr.io/nvidia/tritonserver:22.02-py3
on a Perlmutter compute node but the squash failed. I can pull this image fine on a login node.As a temporary fix @lastephey suggested doing: