kubernetes: backends not updated when i scale replication controller?

[jonaz]

Im running: containous/traefik:v1.0.0-rc2

If i scale a rc up or down: kubectl scale rc --replicas=4 app

The backends in traefik does not get updated immediately the the new IPs from the pods.
If i try accessing service it seems to fetch the new config after a few minutes. Im not really sure. But it should be instant. Or proxy to the service IP and let the service balance it?

Ingress looks like this:

kubectl get ingress app-ingress -o yaml
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  creationTimestamp: 2016-05-19T12:47:24Z
  generation: 1
  name: app-ingress
  namespace: default
  resourceVersion: "1902110"
  selfLink: /apis/extensions/v1beta1/namespaces/default/ingresses/app-ingress
  uid: d61148e3-1dbf-11e6-b22b-005056880f6d
spec:
  rules:
  - host: app.domain.com
    http:
      paths:
      - backend:
          serviceName: app
          servicePort: 8080
        path: /
status:
  loadBalancer: {}

service looks like this:

kubectl get svc app -o yaml
apiVersion: v1
kind: Service
metadata:
  creationTimestamp: 2016-05-19T12:47:08Z
  name: app
  namespace: default
  resourceVersion: "1902061"
  selfLink: /api/v1/namespaces/default/services/app
  uid: cc2c9d93-1dbf-11e6-b22b-005056880f6d
spec:
  clusterIP: 10.3.130.87
  ports:
  - name: http
    port: 8080
    protocol: TCP
    targetPort: 8080
  selector:
    app: app
  sessionAffinity: None
  type: ClusterIP
status:
  loadBalancer: {}

[jonaz]

I think this is related to #449 since i get "Last kubernetes config received less than 2s, waiting..." and it keeps repeating it self resulting in the config not getting updated.

[errm]

sounds like this is caused by #449.

[errm]

@emilevauge I am thinking that even in the case that the provider is spamming updates at the server, the correct behaviour would be to update the configuration every 2 seconds? If this is not the case it seems like there might be a bug somewhere in Server#listenProviders ?

[emilevauge]

@errm, the property ProvidersThrottleDuration as been created to avoid being spammed by providers (for example if you kill 50 containers one shot on Docker). This is not a bug, but a feature 😄.
What can be done here is to decrease this property to 100ms for example.
Finally, a deepEquals check is made in listenConfigurations https://github.com/containous/traefik/blob/master/server.go#L178, I think it would be better to move it in listenProviders https://github.com/containous/traefik/blob/master/server.go#L144, before the Providers Throttle check. Then we would not even wait if configuration doesn't change and skip the event.
WDYT?

[errm]

Yeah, thats how I think it should work . . .