Hi Everyone, When a quota is lowered we currently just stop scheduling workloads in that queue or set of queues. We then rely on the natural pod exits to lower the usage below the new quota set and then continue scheduling. This works with short lived applications. It does not work if the workload is long lived.
The proposal here is to enforce quota change immediately or within configurable graceful delay duration so that newer quota comes into effect from then onwards. I've written a design doc to enforce Quota change (lowering) through preemption. Design doc - https://docs.google.com/document/d/1MYu7msNLNCyy_B-oPsrvoC_FHXgmGCczgGwxEiwfxlc/edit?usp=sharing Please go through the doc and share your feedback. Thanks, Mani
