feature: add open telemetry integration

[fcarreno-collab]

ref: https://app.clickup.com/t/86b7bdn9g

[caseylocker]

You're defining randomString in api/tests/open_telemetry_instrumentation_tests.py but not using it:

class OpenTelemetryInstrumentationTest(APITestCase):
    def randomString(self, str_len):
...

Wrapt was downgraded in the requirements.txt. Was that intentional?
wrapt==2.0.1 -> wrapt==1.17.3

[fcarreno-collab]

You're defining randomString in api/tests/open_telemetry_instrumentation_tests.py but not using it:

class OpenTelemetryInstrumentationTest(APITestCase):
    def randomString(self, str_len):

@caseylocker I removed this code you are right, its not been used, this was to generate a random string for the access_token we use to authenticate the requests on the other endpoints but here that does not happen

Wrapt was downgraded in the requirements.txt. Was that intentional? wrapt==2.0.1 -> wrapt==1.17.3

@caseylocker it was on propuse because newer versions create dependency conflicts when we try to install the open telemetry instrumentation libraries

[caseylocker]

LGTM. I've confirmed that the wrapt downgrade was a requirement by the Opentelementry instrumentation packages. They explicitly require wrapt<2.0.0. No other packages require wrapt 2.x

[smarcet]

@fcarreno-collab please review comments

[dariobressanExo]

@smarcet please can you review this.

[smarcet]

@fcarreno-collab please review

[fcarreno-collab]

@smarcet changes related your comments already on PR

[smarcet]

@fcarreno-collab please review comments

[fcarreno-collab]

@smarcet branch updated with requested changes

[matiasperrone-exo]

@smarcet please review again as all the requested/agreed changes were incorporated.

[romanetar]

span.set_attribute() returns None, so cf_ray is always falsy and the second attribute is never set. The cf.ray_id attribute is written correctly (as a side-effect of the first call), so
the test test_cf_ray_header passes despite this bug, masking it.

[andrestejerina97]

Thank you

[romanetar]

Any environment without this variable explicitly set to false will try to instrument. If OTEL_EXPORTER_OTLP_ENDPOINT is not configured, otel_endpoint mode will fail silently or log errors
on every request. Defaulting to False is safer.

[andrestejerina97]

Thank you

[andrestejerina97]

It's ready to review @romanetar

[romanetar]

LGTM

[smarcet]

@andrestejerina97
The cf.ray_id baggage fallback that shipped in mailing-api (PR #5) is missing here. When a request carries cf.ray_id via OTEL baggage but has no Cf-Ray header — e.g. service-to-service calls where the Cloudflare edge header isn't forwarded — the span never gets the primary cf.ray_id attribute (only baggage.cf.ray_id), so it won't correlate with upstream spans.

Suggested fix to match mailing-api's request_hook (also collapses the two redundant if cf_ray: blocks into one):

Suggested change

	if cf_ray:
	span.set_attribute("cf.ray_id", cf_ray)
	if cf_ray:
	span.set_attribute("http.request.header.cf-ray", cf_ray)
	# Attach baggage if present
	baggage_val = baggage_api.get_baggage("cf.ray_id")
	if baggage_val:
	span.set_attribute("baggage.cf.ray_id", baggage_val)
	if cf_ray:
	span.set_attribute("cf.ray_id", cf_ray)
	span.set_attribute("http.request.header.cf-ray", cf_ray)
	# Attach baggage if present
	baggage_val = baggage_api.get_baggage("cf.ray_id")
	if baggage_val:
	if not cf_ray:
	span.set_attribute("cf.ray_id", baggage_val)
	span.set_attribute("baggage.cf.ray_id", baggage_val)

Also worth covering the header-absent path in tests: test_traceparent_and_baggage injects baggage without a Cf-Ray header but only asserts baggage.cf.ray_id, so this gap is currently untested — add an assertion that cf.ray_id == "xyz" there.

[andrestejerina97]

Solved

[smarcet]

@andrestejerina97 please review comments