Never scale up the deployment back to one replica once scale down to zero

[jmazumder]

Environment:

• Kubernetes distribution (e.g. AKS, ACS Engine, GKE, EKS, etc.):
• Kubernetes version (use kubectl version):
• Osiris version (specify by SHA or semver):
• Install method (specify exact helm install command used):
  helm install

What happened?
Never scale up again the deployment back to one replica

What you expected to happen?
Should scale up again the deployment back to one replica

How to reproduce it (as minimally and precisely as possible):
I tried to validate the example deployment in my cluster and once Osiris-enabled deployment scale to zero replica and hello-osiris pod get terminated it never scale up again the deployment back to one replica

Anything else that we need to know?

[fgrehm]

I had the same problem here, by looking at my osiris-osiris-edge-activator pod I found this:

2020-07-27T23:44:50.13420145Z I0727 23:44:50.133925       1 request_handling.go:10] Request received for for host foo:8080
2020-07-27T23:44:50.134241715Z E0727 23:44:50.133963       1 proxy.go:97] Error executing start proxy callback for host "foo:8080": No deployment found for host foo:8080

I just had to massage my service's osiris.deislabs.io/loadBalancerHostname annotation and things worked. Hope that helps!

[pavel-kurnosov]

@fgrehm I found the same behaviour, but ideally, it should work without this annotation as well.

[Whyeasy]

So I found the issue why this is happening.

The DNS entries Osiris checks for are the following:

<service_ip>:<service_port>
<service_name>.<namespace>.svc.cluster.local:<service_port>
<service_name>.<namespace>.svc.cluster.local:tls
<service_name>.<namespace>:8080
<service_name>.<namespace>:tls

I forked this project to also upgrade it to Go 1.15 and Go Modules. I also added the short DNS entry in to the code so it responds to <service_name>:<service_port>

hopefully this can help you 😄