Contour should configure a stream_idle_timeout

[xstevens]

Describe the solution you'd like
Allow the configuration of Envoy's stream_idle_timeout. According to the Envoy documentation this can be set globally as well as per-route. Envoy defaults this to 5 minutes, but it is overridden by per-route idle timeout.

We use gRPC streaming for some APIs and we would like to keep the connection open as long as possible in many cases. We have currently enabled gRPC keep-alive, but recently realized that Envoy doesn't proxy these through to keep upstream connections alive (see envoy/2086 and envoy/5142).

What happens currently is if we don't send data in under 60 seconds (Contour's default idle timeout) then we get a gRPC error back of "stream timeout".

Anything else you would like to add:

We also need to consider how to do this in a way where we can avoid security issues such as trusting timeouts specified by clients.

Related Issues:

• Configure the per-route value for idle_timeout #944
• Contour should configure a client request timeout #1073 (comment)

Environment:

• Contour version: v0.15
• Kubernetes version: v1.15.3

[davecheney]

@xstevens in 1.0.0-beta.1 we added support for a route.timeoutPolicy.idle per route idle timeout. Is this useful for you?

[xstevens]

Yes, per route idle timeout is useful for us. Thanks!

[ewoutp]

We have same problem. Even after setting idle timeout to infinity, we still get stream timeout errors.

[skriss]

Stream idle timeout is now globally configurable for Contour -- see https://projectcontour.io/docs/v1.9.0/configuration/#timeout-configuration. Closing this out, but feel free to reach out again if needed.