...
Since Workbench users each run in their own namespace, no users would have access to this data because they cannot access the PVC.
No Need for NFS: Copy the Data Manually to the Same Location on Each Node
No changes, Workbench will assume the data is already located on the host at some pre-specified path.
Cluster Admin will be responsible for ensuring that the data is properly copied to all nodes and accessible to containers (with correct permissions).
No need for NFS in this case, because the data is small (~500MB) and read-only
Pros:
- Scales better than NFS - no concurrency problems
Cons:
- Manual steps still needed to scale-up cluster (copying data, changing permissions, etc)
Suck it Up and Just Create Mount the NFS Manually On Each Host
...
Cluster Admin will be responsible for ensuring that the data is properly mounted to all nodes (with correct permissions) and resilient to node restarts.
In practice, this will likely be an external NFS export or globus endpoint mounted manually to each node.
...