Michael Fraenkel

On further review, there seems to be an expectation of how the server should behave with these requests. If the default behavior for a GET / is to return no...

A similar issue occurs with http2/5.1/6. It all comes down to when the second request appears on the Server, the stream may be in half-closed state or it may actually...

Here is the verbose for 5.1/5 and 5.1/6 which have similar issues. 5.1/5 ``` [send] SETTINGS Frame (length:6, flags:0x00, stream_id:0) [recv] SETTINGS Frame (length:24, flags:0x00, stream_id:0) [send] SETTINGS Frame (length:0,...

Define immediately.he test case sends a HEADER frame with EndStream=true. This can result in a HEADER w/EndStream=true or HEADER + DATA (EndStream=true). Like I said, I haven't seen what the...

Well, now that I look at it closer its a bit more tricky. As I showed above, there is a RST_STREAM queued on stream 1, but because an END_STREAM was...

I am now convinced this test is racy. The race is ultimately between the responses of the first and second requests. If the first request sends a response, the stream...

Had to go track this down since its been a while. It started from https://github.com/containerd/containerd/pull/2475 and I have a branch, https://github.com/fraenkel/protobuild/tree/relative_outputs TBH, I don't know how far I got since...

In our environment, we too are hitting rate limits on AWS. I have already increased our aws retries to 10 although now I am considering 13 with a much longer...

Your key difference is that you are using haproxy and not the GCE Ingress which places requirements on how healthcheck works. Combining Attempt 2 and 4 is probably the best...

Yes, those three provide the ability to use the generated certificate or add your own on a separate listener and expose it via Ingress or a LoadBalancer.