kueue
kueue copied to clipboard
Introduce ResourceFlavor fallback mechanism
What would you like to be added: A mechanism that would allow fallback to a different ResourceFlavor, if Kueue cannot start a job on assigned flavor.
Why is this needed: Currently, there is no fallback mechanism to a different flavor in Kueue. It means that if there is a free capacity in Kueue, but there are stockouts on the cloud provider side, Kueue will assign the same flavor over and over to a given Workload. This results in wasteful assignments VMs to a Workload that will not start (e.g. Workload will get repeatedly 5 VMs, when it needs 10 of them to start)
Users would like to be able to configure Kueue in a way, so that in case there are stockouts, Kueue will try a different flavor.
Completion requirements:
This enhancement requires the following artifacts:
- [ ] Design doc
- [ ] API change
- [ ] Docs update
The artifacts should be linked in subsequent comments.