05/26/2023, 4:42 PM
Hello. We install k8s cluster (EKS), k8s worker nodes, cilium (component for the networking) and flux (helm operator). Cilium and flux are installed through helmv2 pulumi lib, NewChart function.
If cilium is uninstalled from the platform, then flux can't be uninstalled. Also cilium chart includes one deployment that also can't be deleted when cilium daemonset is removed. When we destroy the stack, cilium daemonset uninstalled earlier than other components (hubble and flux) and then stack fails to be executed because hubble and flux pods can't be deleted. We tried to use custom timeouts but seems it doesn't work.
Also it doesn't work with NewRelease function from helmv3 pulumi lib.
Why we think that delete timeout can help? Nodes are removed in the same stack and if pulumi waits a bit more while nodes are removed, then pods are removed automatically and the whole stack succeeds. So the question. Why delete custom timeout doesn't work for NewRelease function? And how to make it work?