[YUNIKORN-3128] Rollback allocation on bind failure

[jeffgao01]

What is this PR for?

This PR is regarding https://issues.apache.org/jira/browse/YUNIKORN-3128

• Define BindFailed taskEventType
• Rollback allocation on bind-volumes-to-pod or bind-pod-to-node failure
• Modify the task state to new, for follow-up retry

What type of PR is it?

• - Bug Fix
• - Improvement
• - Feature
• - Documentation
• - Hot Fix
• - Refactoring

Todos

• - Task

What is the Jira issue?

• https://issues.apache.org/jira/browse/YUNIKORN-3128

How should this be tested?

• unit test

Screenshots (if appropriate)

Questions:

• - The licenses files need update.
• - There is breaking changes for older versions.
• - It needs documentation.

[wilfred-s]

@jeffgao01 please run make lint and make test over your PR and fix any issues you find

[wilfred-s]

Looking at the change I think we're missing the part that needs to communicate the failure back to the core. The core already has assigned a node to the allocation. That also needs to be undone. Without that the core still thinks that the allocation is assigned to a node and will never retry it.

[jeffgao01]

Looking at the change I think we're missing the part that needs to communicate the failure back to the core. The core already has assigned a node to the allocation. That also needs to be undone. Without that the core still thinks that the allocation is assigned to a node and will never retry it.

Actually, the current change is rolling back the allocation as a whole.
But I synced with Wilfred, a more fine-grained approach is to partially rollback the allocation, to the extent that it can be retried (but the existing metadata/history is retained).