fix: HTTP/2 sends now wait up to 5 seconds for flow-control credit instead of returning send_buffer_full

[speeddragon]

Disclaimer: This PR code was generated by AI.

I'm trying to address #895. I'm having the same issue with ex_aws, and this PR modifies the request to wait until a specific time rather than returning {error, send_buffer_full}.

This change allows waiting for the buffer to clear instead of returning an error.

How to use it with ExAws:

  config :ex_aws, :hackney_opts,
    h2_send_timeout: 5_000

Or per request:

  ExAws.request(operation, http_opts: [h2_send_timeout: 10_000]) 

TODO

• Run rebar3 eunit
• Update documentation about the new property

[benoitc]

It seems that what you want is blocking send i but is there any reason you need a configurable send wait per request ? I don't see a real use case for h2_send_timeoutand the complexity it adds.

This will change the behabiour though, instead of returnong once the 1MB buffer cap is crossed {error, send_buffer_full} , the application will have to wait for a {error, timeout}

[speeddragon]

Thanks for the feedback.

It seems that what you want is blocking send i but is there any reason you need a configurable send wait per request ?

It doesn't need to be per request. I'm not sure we can define a more low-level place and have it picked up by other high-level (connection vs request).

I've also noticed this doesn't preserve the default behaviour. I will make this optional.

This will change the behabiour though, instead of returnong once the 1MB buffer cap is crossed {error, send_buffer_full} , the application will have to wait for a {error, timeout}

Is there a way to handle send_buffer_full outside of hackney? Retry doesn't look like an option because will lead to the same error. Modifying the buffer size doesn't look great as well.

Maybe I'm missing the right way to handle send_buffer_full outside of hackney.