This repository was archived by the owner on Jun 29, 2022. It is now read-only.
This repository was archived by the owner on Jun 29, 2022. It is now read-only.
Document the process of ingress nodes upgrade without downtime #485
Open
Description
If a worker node pool is running Contour and MetalLB on them. What is the {node, bootstrap kubelet} upgrade process look like? How can we achieve it without downtime?
Do we drain the node one node at a time? Let it reboot and then move to another one? How does this work? There are problems with draining nodes, they don't drain Daemonsets by default.
What are the implications of draining Daemonsets? If you plan on removing MetalLB and Contour which run as Daemonset but it will also drain Daemonsets like calico, kube-proxy?
What is the convergence time for BGP how do we make sure that the BGP peer of the "node to be upgraded" will deregister the node and then we can safely remove the node?