OpenTelemetry integration

[alessio-locatelli]

MRE

For a pure aiohttp we can track requests / responses:

import asyncio
from typing import Union
import aiohttp
from aiohttp_client_cache import CachedSession, DictCache
from opentelemetry.instrumentation.aiohttp_client import Span, create_trace_config
import yarl

def strip_query_params(url: yarl.URL) -> str:
    return str(url.with_query(None))


def request_hook(span: Span, params: aiohttp.TraceRequestStartParams):
    print("request_hook")
    print(f'{params=}, {span=}')
    if span and span.is_recording():
        span.set_attribute("custom_user_attribute_from_request_hook", "some-value")


def response_hook(span: Span, params: Union[
             aiohttp.TraceRequestEndParams,
             aiohttp.TraceRequestExceptionParams,
         ]):
    print("response_hook")
    print(f'{params=}, {span=}')
    if span and span.is_recording():
        span.set_attribute("custom_user_attribute_from_response_hook", "some-value")


async def main():
    async with CachedSession(
        cache=DictCache(),
        trace_configs=[create_trace_config(
            request_hook=request_hook,
            response_hook=response_hook,
            # Remove all query params from the URL attribute on the span.
            url_filter=strip_query_params,
    )]) as session:
        async with session.get("https://www.example.com") as response:
            await response.text()
        async with session.get("https://www.example.com") as response:
            await response.text()


asyncio.run(main())

Output:

request_hook
params=TraceRequestStartParams(method='GET', url=URL('https://www.example.com'), headers=<CIMultiDict()>), span=NonRecordingSpan(SpanContext(trace_id=0x00000000000000000000000000000000, span_id=0x0000000000000000, trace_flags=0x00, trace_state=[], is_remote=False))

response_hook
params=TraceRequestEndParams(method='GET', url=URL('https://www.example.com'), headers=<CIMultiDict()>, response=<ClientResponse(https://www.example.com) [200 OK]>
<CIMultiDictProxy('Accept-Ranges': 'bytes', 'Content-Type': 'text/html', 'Etag': '"84238dfc8092e5d9c0dac8ef93371a07:1736799080.121134"', 'Last-Modified': 'Mon, 13 Jan 2025 20:11:20 GMT', 'Vary': 'Accept-Encoding', 'Content-Encoding': 'gzip', 'Content-Length': '648', 'Cache-Control': 'max-age=2461', 'Date': 'Thu, 23 Jan 2025 14:29:15 GMT', 'Alt-Svc': 'h3=":443"; ma=93600,h3-29=":443"; ma=93600,h3-Q050=":443"; ma=93600,quic=":443"; ma=93600; v="46,43"', 'Connection': 'keep-alive')>
), span=NonRecordingSpan(SpanContext(trace_id=0x00000000000000000000000000000000, span_id=0x0000000000000000, trace_flags=0x00, trace_state=[], is_remote=False))

However, request_hook and response_hook are not triggered if a request is cached.

Maybe I would like to add span.set_attribute("custom_user_attribute_from_response_hook", "this_response_was_from_cache") or something like that, but def _request() is not called if we have a cached response, so hooks are not called as well.

Question

This may be a bit of a silly question as I am fairly new to telemetry, but what is the correct and simple way to track the number of cached responses with OpenTelemetry?
Perhaps there is a way to access aiohttp self._trace_configs, but I wonder about the way that does not require editing aiohttp-client-cache source code.

Related

• https://docs.aiohttp.org/en/stable/client_advanced.html#client-tracing
• https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/aiohttp_client/aiohttp_client.html