Readiness Probe does not get reflected for NEG enabled ingress

[freehan]

When the following conditions apply:

• NEG is enabled on the ingress referenced service
• Service is ClusterIP type

Then LB health check is created with default setting.

[manigandham]

We see this happen to on the latest version 1.11.2-gke.18 GKE clusters. NEG health checks are not reflecting any of the settings in the deployment YAML and remain unhealthy.

[hrdymchl]

+1

Posted about my issue on SO here, and this sounds like the same problem. Am also on version 1.11.2-gke.18.

[uschen]

same issue here 1.11.3 and 1.11.4

the NEG will retain the old pod IPs for at least 5min, in the meantime, requests will be responded with 502.

[freehan]

same issue here 1.11.3 and 1.11.4

the NEG will retain the old pod IPs for at least 5min, in the meantime, requests will be responded with 502.

@uschen Based on you description, it might be another issue. I created an umbrella issue for container native load balancing #583. Can you take a look and see if it matches any known issues?

If not, would you mind creating an issue with more detail? It would be great if it could include repro steps.

[freehan]

PR to fix the health check bug is in flight: #582