[SCALE k3s] LoadBalancer pods remain active after changing service types

Description

After creating an app using a LoadBalancer service type and then subsequently changing the service type to ClusterIP or NodePort, the old loadbalancer pod is not destroyed, and remains active.

The svclb pod will be re-created even after manually destroying the pod from the command-line.

The old svclb pods are re-created even after a system restart.

To reproduce
1. Create an app from TrueCharts catalog with "Simple" service type (which is a loadbalancer).
2. Once the app is active, edit the app configuration to change the service type from simple to ClusterIP or NodePort.
3. Run the command `k3s kubectl get pods -A` to see the svclb pod persists.
4. To confirm this, you can still access the app from the old loadbalancer-configured port (http://hostIP:oldLBport)
5. Try to delete the loadbalancer pod with `k3s kubectl n ix[APPname] delete pods [svclb POD name] and after successfully deleting it, the svclb pod will be recreated

Problem/Justification

None

Impact

None

Activity

Show:

Waqar Ahmed November 29, 2021 at 2:07 PM

can you please create upstream issues at k3s-io as the load balancer service is being brought in by upstream and hence they should be making sure we don't have redundant pods leftover when they are not being consumed. Thanks!

Kjeld Schouten-lebbing November 26, 2021 at 9:11 PM

This one is yours I guess...

Also might be related to:
https://github.com/k3s-io/klipper-lb/issues/3
https://github.com/k3s-io/k3s/issues/1870

It looks to be a fuckup with klipperLB

Kjeld Schouten-lebbing November 26, 2021 at 9:07 PM

Addition:

Service Type "Simple" is just a GUI Alias for kubernetes servicetype "LoadBalancer"
Service does get correctly registered as "ClusterIP" when running `k3s kubectl get svc`