send partial bodies

[stillinbeta]

adds http2_connection:send_body/4, which can take an arg
{send_rst_stream, false}. This lets handlers call send_body
multiple times instead of needing to buffer output

[joedevivo]

The name of this option should be 'end_stream', 'rst_stream' is for error
cases

On Thursday, March 31, 2016, Liz notifications@github.com wrote:

adds http2_connection:send_body/4, which can take an arg
{send_rst_stream, false}. This lets handlers call send_body

multiple times instead of needing to buffer output

You can view, comment on, or merge this pull request online at:

#55
Commit Summary

• Tests and skeleton for new send body
• Support sending body in multiple segments

File Changes

• M src/http2_connection.erl
  https://github.com/joedevivo/chatterbox/pull/55/files#diff-0 (29)
• M test/client_server_SUITE.erl
  https://github.com/joedevivo/chatterbox/pull/55/files#diff-1 (47)
• A test/double_body_handler.erl
  https://github.com/joedevivo/chatterbox/pull/55/files#diff-2 (51)

Patch Links:

• https://github.com/joedevivo/chatterbox/pull/55.patch
• https://github.com/joedevivo/chatterbox/pull/55.diff

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#55

[stillinbeta]

no idea why that test's failing, and I can't seem to repro it locally. maybe a flaky test?

[joedevivo]

It's failing consistently in travis, which makes me think it's not a flaky test. It does work for me locally, so I'm not really sure what's going on, but I'll investigate.

[joedevivo]

Looks like I finally got one to pass, but I need to see some consistency.

[joedevivo]

Four in a row passed, so I'll look at the code now :D

[joedevivo]

There's a better way to test this using http2c. It will allow you to inspect the response as a series of frames. You can then assert that the response is made of two data frames, and that they're broken up correctly.

[stillinbeta]

I think that's a level of abstraction lower than we need to test. as a user I don't care how you send my data, just that you do as fast as possible.

[joedevivo]

It's not really doing that though. For better or worse, this test would pass if the server held back all the response body until it was ready to send it all.

It might mean that we have to add some kind of streaming functionality to the http2_client, so that we can see the response coming in pieces. We might have to add some kind of timer delay to the double_body_handler to take advantage of that.

[stillinbeta]

Just travis things ¯_(ツ)_/¯

[joedevivo]

Need to change this line to something along the lines of

queued_data=<<Stream#stream.queued_data,Body>>,

but really it needs to be safer than that, something like

NewBody = case is_binary(Stream#stream.queued_data) of
    true ->
        <<Stream#stream.queued_data,Body>>;
    false ->
        Body
end,
{NewSWS, NewS} =
   s_send_what_we_can(Conn#connection.send_window_size,
                      Conn#connection.send_settings#settings.max_frame_size,
                      Stream#stream{
                          queued_data=NewBody,

That and a supporting test and this PR should be ready to merge

[joedevivo]

A+! Would merge again