Don't set externalTrafficPolicy when externalService is ClusterIP

[baby-gnu]

Describe the bug

I can't deploy the helm with the following values.yaml:

front:
  hostPort:
    enabled: false

  externalService:
    enabled: true
    type: ClusterIP

I get the following error:

Error: Service "mailu-front-ext" is invalid: spec.externalTrafficPolicy: Invalid value: "Local": may only be set when `type` is 'NodePort' or 'LoadBalancer'

Environment

• Kubernetes Platform Scaleway

Additional context

I have an existing nginx ingress as LoadBalancer with correct TCP service forwarding.

[emouawad]

+1 same here

[jlxq0]

Same. Did anyone figure this out?

[WebSpider]

I'll see if i can provide a PR for this

[jlxq0]

What's the workaround, @WebSpider?

[WebSpider]

Sorry, real life happend. Thanks for tagging me, ill PR tonight.

Its a matter of putting an extra 'if' in the template that creates the service.

[jlxq0]

How do we get this merged now? 😇

[mreho]

Hello, same issue here, I have a Octavia LoadBalancer in front of my K8S cluster (with Proxy Protocol, X-Forwarded-For headers enabled, and Real IP correctly set up), so the externalTrafficPolicy should be disabled for me too.

Maybe it can be a good thing to disable it by default, since most of the people running "heavy" apps like Mailu have load balancers in front of their cluster..

[larsskj]

I'm hit as well - though my situation seems to be a little different.

My local on-prem cluster is running MetalLB in order to create K8s services of type LoadBalancer complete with external IPs on my local network. I use it a lot - not least for services not running HTTP(S) like my Git server using SSH.

And Mailu with all the different mail related ports.

This has worked fine for a long time, but stopped working last night when I upgraded MetalLB to the latest incarnation. Everything worked fine - except for Mailu: It took some time to figure out why, but externalTrafficPolicy was the problem for me as well. Changing it from Local to Cluster like the 15+ other services using a LoadBalancer solved the problem.

Why do you set it to Local in the first place? It seems like a weird decission to me? Why not just leave the setting alone like all my other Helm charts?

[mreho]

Hello @larsskj,
My loadbalancer is not managed by Kubernetes so I set all my services to ClusterIP (LoadBalancer would make my service in Pending state, waiting for an external IP), that's why externalTrafficPolicy is not applicable in my case.