Francesco Romani

Results 485 comments of Francesco Romani

> TODO: make sure we handle burst of deletions No code change seems needed. Add a code comment to explain why.

existing tests should now pass and cover the non-regression promise. I'll add tests to demonstrate and verify the new behaviour and then lift the WIP.

> @fromanirh I rethink this problem a bit: the NRT plugin's filter() logic relies on the NRT object's "Available" field: > > https://github.com/kubernetes-sigs/scheduler-plugins/blob/7b3182b2e1750f673d00ae0edcb5a3609a528e4a/pkg/noderesourcetopology/pluginhelpers.go#L143 > > While this field gets frequently...

Apologies for joining the discussion a bit late, I had a couple hectic days. First of all thanks @swatisehgal for the brilliant summarization; you described the problem pretty much like...

> > > We ONLY need: Node A has two NUMA nodes and one has 3 CPUs and another one has 4 and based on _just_ that information we can...

> Good conversations and now we have two problems upfront: > > 1. strategy to reserve pod resources - either pessimistically sub from all NUMA nodes, or intelligently sub from...

update: one of the biggest issues emerged during the review was the cache invalidation algorithm. I made some progress here and found possible way forwards I'm playing with, but I'm...

Over the past months I (with help from my team) redesigned the logic. The PR now contains the new logic and is reviewable. WIP is kept because I want to...