Client request tracing. Allow to get the request body in the trace.

[kowalski]

Long story short

I would like to get the request_body in the tracing events. Obviously it cannot be included in on_request_start because body can come from generator.

I would be happy to contribute another event, say on_body_sent which would fire after the body is written after the body is finished.

[pfreixes]

Please go ahead!

[asvetlov]

Saving response body would be interesting as well.

[kowalski]

Ok, great! Thanks for encouragement.

Before I get into coding I would like to discuss an idea around this.
The barest possible idea is to simply save the request body bytes into a variable and pass it to a Signal after everything is written. Unfortunately I can see how in certain use cases this could have catastrophic performance and memory footprint.
So instead I would like to propose, that I will move registering for grabbing a request/response body into a method on:

class TraceConfigContext(SimpleNamespace):

    def register_content_handlers(self, request_stream, response_stream):
          ....

The idea would be that this method is called from on_request_start callback and the results are gathered in on_request_end.

Also this means, that if user code chooses not to read the response body it will not be available for tracing. I mean getting 400 status when the code looks like:

async with session.get(url) as response:
    if response.status == 400:
        raise SomeError
    # here the normal processing follows, call response.json() and do something with in

What do you think ?

[pfreixes]

Umm ... once the strings are allocated in memory I would say that internally the only thing that is given as a parameter is just a String object that points to that region. So, passing the request or the response as a parameter won't affect the memory, and only if the signal handle mutates the string a new copy will be made.

am I missing something?

[kowalski]

Sorry for not being clear.
I meant a scenario when request body is not passed as string, but as for example aiohttp.payload.BufferedReaderPayload pointing to a file which will not fit in memory.

[asvetlov]

We should call signal as many times as we call transport.write. In this case everything should fit in memory.

[pfreixes]

We might thing on rename the signals as on_request_chunk_sent and on_response_chunk_received, if chunk is used this function will be called many times if not will be called once.

[asvetlov]

Fixed by #2767

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a [new issue] for related bugs.
If you feel like there's important points made in this discussion, please include those exceprts into that [new issue].
[new issue]: https://github.com/aio-libs/aiohttp/issues/new