Gateway updates is causing downtime? about contour HOT 11 CLOSED

Yep, can definitely repro on GKE, a few seconds after making the Gateway change to add/remove a Listener/LB service port, connections start failing for ~30 seconds or so

from contour.

Thanks for your help @skriss - I'm going to close this out and figure out how to raise an issue with GKE team.

from contour.

Sorry logs are here
contour.logs.gz

from contour.

Following up - Contour updates a Service type=LoadBalancer when listeners are added on the Gateway (eg. 443).

For some reason on GKE when this update to the Service happens drops traffic on other listener ports (eg. 80).

from contour.

Follow up work @skriss mentioned is to see if this pattern is exhibited on other LBs. If so contour may want to add a workaround where they always have a port 433 on the K8s Service definition.

from contour.

Initial tests on Azure/AKS are not showing this problem, the LB remains "up" while additional ports are added/removed

from contour.

AWS (EKS + NLB) also do not show this problem, LB remains up while adding/removing additional ports

from contour.

Thanks for testing - can you run your tests against GKE. It would expect it to fail but worth doing the validation.

from contour.

Yep, will do, need to track down an account

from contour.

@skriss what cluster version are you using?

from contour.

@skriss what cluster version are you using?

1.27.8-gke.1067004

from contour.