CA DRA: review DRA-related error policy #7784
Description
Which component are you using?:
/area cluster-autoscaler
/area core-autoscaler
/wg device-management
Is your feature request designed to solve a problem? If so describe the problem this feature should solve.:
Cluster Autoscaler tends to error out and break the whole loop in case of any unexpected errors, and the DRA MVP PR mostly follows this approach for simplicity. This is not a good direction in general, we've had a number of issues in GKE CA where a bug related to a small subset of pods/nodes would break CA completely because of it.
Describe the solution you'd like.:
We should holistically rethink if CA can proceed with the loop when it encounters DRA-related errors (and ideally non-DRA-related errors as well but that's a separate issue).
Additional context.:
This is a part of Dynamic Resource Allocation (DRA) support in Cluster Autoscaler. An MVP of the support was implemented in #7530 (with the whole implementation tracked in kubernetes/kubernetes#118612). There are a number of post-MVP follow-ups to be addressed before DRA autoscaling is ready for production use - this is one of them.