Open viceice opened 2 years ago
Couldn't track this on baremetal. Could you please confirm on other platforms like minikube or kind?
seeing this on k3s
What k3s version do you use? I see some possibly related issues in k3s project: https://github.com/k3s-io/k3s/issues/2722
I use v1.23.6+k3s1
, so i don't think it's the containerd issue. I also run k3s on plain ubuntu 20.04 virtual maschines.
Found the same bug, but with an inhouse chart. Removing exec
probes in favor of anything else fixed the problem. It seems related to bottom notices at https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/#configure-probes although it doesn't make sense as k3s uses containerd instead of docker-shim.
Root cause seems to be that probe invocation processes are children of the main pod process which needs to reap them once they shut down.
Edit: also k3s v1.23.6+k3s1
@ Ubuntu 20.04.
For my own images i use dumb-init
as entrypoint, with will do this job very well
I've now build a custom image which starts dumb-init before keydb, so hopefully no more zombies 🤞
I've got a lot zombi processes.
values.yml