You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
CMPs should be able to be run in their own pods instead of relying on resources for the repo-server pod.
Motivation
Our nodes are resource constrained. We run a couple of CMPs in our repo-server pod and the resources required to make everyone happy exceed the node size. But 99% of the time none of the containers need any resources, not the argocd-repo-server itself and certainly not any of the CMPs.
Proposal
Whether the argocd-repo-server relies on a new CRD to define a new pod type and instantiates it, or whether it relies on some other magic is not something I care about. The end result should be a way to define a CMP such that argocd-repo-server can use it to resolve CMP handled content, and when CMPs aren't needed, their cost to the k8s cluster should be approximately 0.
Summary
CMPs should be able to be run in their own pods instead of relying on resources for the repo-server pod.
Motivation
Our nodes are resource constrained. We run a couple of CMPs in our repo-server pod and the resources required to make everyone happy exceed the node size. But 99% of the time none of the containers need any resources, not the argocd-repo-server itself and certainly not any of the CMPs.
Proposal
Whether the argocd-repo-server relies on a new CRD to define a new pod type and instantiates it, or whether it relies on some other magic is not something I care about. The end result should be a way to define a CMP such that argocd-repo-server can use it to resolve CMP handled content, and when CMPs aren't needed, their cost to the k8s cluster should be approximately 0.
See also #15006.
The text was updated successfully, but these errors were encountered: