test: adding permit to allow for sibling pod scheduling #74
Conversation
Problem: the submit of the first index works for more controlled lengths (e.g., lammps takes a while) but was having issues with really quick jobs. Solution: try restoring the queue that allows for enabling siblings pods so any group can be scheduled. Signed-off-by: vsoch <[email protected]>
|
Failure due to controller entry point change from 5 days ago. Hopefully won’t require a Kubernetes component version update. kubernetes-sigs/scheduler-plugins@4d3d41c |
|
I think when we merge flux-framework/fluxion-go#8 that should update fluence go bindings to 1.21, and then we can attempt updating here. sig-scheduler plugins is at go 1.21 https://github.com/kubernetes-sigs/scheduler-plugins/blob/51d27b6e06b339bfa413d0415d80c86d01097b44/go.mod#L3. |
|
I'm going to try building with that branch. If it works, I'll merge there and update here, and if this passes we can merge into the other PR branch. That's a lot of "ifs" 😆 |
|
Looks like the upstream is still a moving target! They bumped kubernetes now up to 1.29x. kubernetes-sigs/scheduler-plugins@2d20310. It changed a few function signature, trying to figure that out now. |
Problem: we need to update to a newer go to keep up with the sig-scheduler upstream, and also the rainbow scheduler integration. Solution: upgrade to 1.21. This also required some refactor of the main.go and fluence due to changes in function signatures. This is a test to see if tests are passing - the fluxion-go bindings used here are from a branch (not merged yet) that can be used for the PR this one is going into, and before merging that final one we should merge and release the bindings more properly. Signed-off-by: vsoch <[email protected]>
Problem: the submit of the first index works for more controlled lengths (e.g., lammps takes a while) but was having issues with really quick jobs.
Solution: try restoring the queue that allows for enabling siblings pods so any group can be scheduled.