Doug Fawley

> One more: > 4. Maintain the same order provided by the name resolver. > How important is this One related thing that does matter is that if an address...

> Moreover, the size is sent from the server in the settings frame, but the client never reads and applies it. Is that correct? This seems fine. Our default is...

We looked into this migration a year or two ago, but it's a non-trivial amount of work (due to the API changes) for a relatively small benefit (it's been rewritten...

...which was a dupe of #3554 I think we should have an issue for this open, but we have no upcoming plans to do anything here.

FWIW I created https://github.com/golang/protobuf/issues/1442 so we can try to get a non-deprecated way of maintaining our existing API.

> The error must either be checked on the first Read Does this mean we need to propagate the error returned when we're trying to read the server's preface, then?

Note that this would necessitate moving the reading of the preface from `reader` and to `newHTTP2Client`. It would be done in blocking fashion, which we've discussed doing for awhile now,...

> I think this is not suitable for a production use case, should be disabled by default. We agree; tracing was disabled by default in #1509. We may delete the...

@elvizlai can you be a little more specific? Why does istio/envoy retry strategy not work? gRPC retry does not work if the server sends headers. This is by design. Please...

> May be I should check if next middleware returns error, then using Trailer instead of Header? I believe the gRPC library should be (but isn't) automatically doing what you...