Pires

What I propose is to clearly identify the VK incompatibility w/ CCM in docs and even print a warning during VK initialization. @cpuguy83 wdyt?

I tend to agree w/ the idempotent approach to the reconciler. This is actually how I'm used to writing controllers, using controller-runtime, which I would suggest to adopt here if...

Oh, as mentioned over Slack, I don't quite follow this: >Since the VK knows all pods that are scheduled on it, we can scan over these. We shouldn't scan and...

That's what `full resync period` is all about, the opportunity for controllers to relist the objects they care about and have client-go re-generate events that may have been lost, if...

You can relist w/ field selectors, namely `nodeName==VKnodename` and reduce the load on both the API server and the client-go internals. I need you to share your concerns in more...

(1) API server keeps its own cache to avoid overloading etcd when it can. (2) Unless you can show the problem happening and at what scale it happened, I can't...

@cpuguy83 I think recently it was agreed we'd remove the Helm stuff from the repo, since VK is just a library now? But the monitored namespaces is a different subject...

My proposal to close this one is to remove Helm charts, deprecate the namespaced VK in the next minor release and drop it entirely in the next release.

I still share concerns about using the same Kubernetes client the VK uses to do its thing (track Pods, Secrets, ConfigMaps, the Node, send events). This means an option is...

My idea really is to provide an event recorder while making sure it's a client instance that is not used by VK, to avoid bad providers spamming events to mess...