If two Services select the same pods then pods won't start

[lkysow]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request. Searching for pre-existing feature requests helps us consolidate datapoints for identical requirements into a single place, thank you!
• Please do not leave "+1" or other comments that do not add relevant new information or questions, they generate extra noise for issue followers and do not help prioritize the request.
• If you are interested in working on this issue or have submitted a pull request, please leave a comment.

---

Overview of the Issue

If two kube services have the same selector of mesh pods then Consul will register 4 service instances (two for each pod, sidecar and app) and then connect-init will loop forever because it's waiting for exactly 2 service instances.

Possible solutions

We can't register two service instances for one pod because the pod's sidecar can only receive traffic for one service name. So I think we need to provide a way to exclude one of the services from syncing. I checked and if you label the service, the labels are available at the endpoints level so we could do something basic like:

labels:
  consul.hashicorp.com/connect-ignore: true

[chauhanshubham]

Hello, any further updates on this? I'd like to implement it :)

[david-yu]

@chauhanshubham Could you tell me more about why you would need this particular feature? In this case we are ignoring one of the services via a label. Are you also looking to deploy something like NGINX Ingress controller with Consul K8s sidecar proxies?

[chauhanshubham]

hi @david-yu i was just looking at it as a good first issue. Since the connect-init loop seems to get stuck in this scenario, I thought it should probably get fixed, irrespective of any specific usecases.

[mike-code]

@david-yu Actually I exactly have a case with nginx. My understanding is that having nginx with sidecar proxy is a valid scenario because otherwise I won't be able the to allow the nginx's loadbalancer service to communicate with applications that have sidecar proxies enabled.

Nginx deployment deploys two services nginx-controller and nginx-controller-admission (one being of type LoadBalancer and the other of ClusterIP) and both are bound to the same deployment. What would be the right way of dealing with such scenarios?

[david-yu]

@mike-code We will be releasing the functionality that addresses this issue in our next release most likely around mid-December. Stay tuned!

[markharding]

@mike-code We will be releasing the functionality that addresses this issue in our next release most likely around mid-December. Stay tuned!

@david-yu any update on this?

[ishustava]

Hey @markharding the fix for this was released, and you can now add the consul.hashicorp.com/service-ignore label to your service as mentioned in #858. If you are still running into issues using the latest release of the helm chart, please create a new issue and describe the steps for us to reproduce. Thanks!