Envoy graceful shutdown

[stevesloka]

Feature Request

When restarting Envoy (i.e. upgrading, draining nodes, etc), the pod receives a SIGTERM from Kubernetes and all connections are dropped. We should signal to Envoy that it's going down and attempt to drain before rolling the pod: envoyproxy/envoy#1990 (comment)

We should be able to use the lifecycle hooks and add something like this, however, curl isn't in the envoy pod. So either we will need to add that which has our users utilize a custom image (not favorable), or we add another side-car which listens for the SIGTERM and sends the POST request to Envoy to shut down.

lifecycle:
  preStop:
  exec:
    command: ["curl","-XPOST","http://localhost:9001/healthcheck/fail"] 

// #135

[rosskukulinski]

-1 to custom envoy image. But +1 on this feature.

[phylake]

No curl but you do have wget

[stevesloka]

We have this fixed in Contour (projectcontour/contour#781). I'll leave this open so that we can update the examples in Gimbal to reflect.

[azuretek]

There's still an issue projectcontour/contour#781 (review)

[davecheney]

@steveslola can you please take this one. Thank you.