After resizing traefik does not extend units to more backends
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Snap |
Triaged
|
High
|
Unassigned |
Bug Description
Scenario
- bootstrap with --accept-defaults
- add node with just control and compute role
- add node with just control and compute role
sunbeam cluster resize
and wait for the end of it
monitor api with
while true; do openstack server list && echo OK; done
kill keystone-0 pod
and it takes 1min around to have the API back
Seems that Traefik loadbalancer got just one server backend
ubuntu@node08:~$ kubectl exec -n openstack -c traefik -t traefik-0 -- cat /opt/traefik/
ingress_
http:
….
domains:
- main: 10.20.21.12
sans:
- '*.10.20.21.12'
services:
juju-openstack
loadBalancer:
servers:
- url: http://
and that is true also for traefik-1 and traefik-2
description: | updated |
This is a bug in the ingress v1 interface - its not actually providing IngressPerApp as the leader unit populates the app databag with its FQDN, rather than that of the K8S service it sits behind.
The ingress v2 interface has the right behaviour but is currently only in the traefik/edge channel.