feat: httpx instrumentation

[jy-tan]

Summary

Adds instrumentation for the httpx HTTP client library, supporting both synchronous (httpx.Client) and asynchronous (httpx.AsyncClient) clients.

Changes

New files:

• drift/instrumentation/httpx/__init__.py - Package exports
• drift/instrumentation/httpx/instrumentation.py - Main instrumentation

Modified files:

• drift/core/drift_sdk.py - Auto-initialize httpx instrumentation when the library is available
• drift/instrumentation/requests/instrumentation.py - Set calling_library_context during network calls
• drift/instrumentation/socket/instrumentation.py - Fix unpatched dependency detection logic

Features

• RECORD mode: Captures httpx request/response data as CLIENT spans with proper schema merges
• REPLAY mode: Returns mocked responses via find_mock_response_async/find_mock_response_sync
• Transform support: Integrates with HttpTransformEngine for drop/redact/mask transforms
• Async support: Full async/await support for httpx.AsyncClient

Socket instrumentation fix

The socket instrumentation detects unpatched dependencies by monitoring raw TCP calls during REPLAY mode. Previously, it only allowed ProtobufCommunicator (SDK's internal communication) and would incorrectly flag instrumented HTTP libraries as "unpatched".

The fix changes the check from:

if calling_library != "ProtobufCommunicator":  # ❌ Flags httpx/requests

to:

if calling_library is None:  # ✅ Only flags truly unpatched libraries

Both httpx and requests instrumentations now set calling_library_context when making network calls, preventing false positives.

Testing

• FastAPI e2e test passes (10/10 tests)
• httpx spans verified in recorded traces (HttpxInstrumentation CLIENT spans)
• No socket instrumentation warnings in REPLAY mode

[cubic-dev-ai]

4 issues found across 5 files

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="drift/instrumentation/httpx/instrumentation.py">

<violation number="1" location="drift/instrumentation/httpx/instrumentation.py:175">
P1: `span.end()` is called twice when request is dropped by transform - once here and again in the outer finally block. Remove this call since the finally block already handles span termination.</violation>

<violation number="2" location="drift/instrumentation/httpx/instrumentation.py:199">
P2: Dead code: duration calculation result is discarded. Either remove this line or assign the result to a variable for use (e.g., passing to `_finalize_span`).</violation>

<violation number="3" location="drift/instrumentation/httpx/instrumentation.py:412">
P2: Potential TypeError when headers=None in async mock lookup; guard with `kwargs.get("headers") or {}` before dict().</violation>

<violation number="4" location="drift/instrumentation/httpx/instrumentation.py:656">
P2: Potential TypeError during span finalization when headers=None; guard with `request_kwargs.get("headers") or {}` before dict().</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

P1: span.end() is called twice when request is dropped by transform - once here and again in the outer finally block. Remove this call since the finally block already handles span termination.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At drift/instrumentation/httpx/instrumentation.py, line 175:

<comment>`span.end()` is called twice when request is dropped by transform - once here and again in the outer finally block. Remove this call since the finally block already handles span termination.</comment>

<file context>
@@ -0,0 +1,992 @@
+                        json.dumps({"bodyProcessingError": "dropped"}),
+                    )
+                    span.set_status(Status(OTelStatusCode.ERROR, "Dropped by transform"))
+                    span.end()
+
+                    raise RequestDroppedByTransform(
</file context>

[cubic-dev-ai]

P2: Dead code: duration calculation result is discarded. Either remove this line or assign the result to a variable for use (e.g., passing to _finalize_span).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At drift/instrumentation/httpx/instrumentation.py, line 199:

<comment>Dead code: duration calculation result is discarded. Either remove this line or assign the result to a variable for use (e.g., passing to `_finalize_span`).</comment>

<file context>
@@ -0,0 +1,992 @@
+                finally:
+                    calling_library_context.reset(calling_lib_token)
+                    # Finalize span with request/response data
+                    (time.time_ns() - start_time_ns) / 1_000_000
+                    instrumentation._finalize_span(
+                        span,
</file context>

Suggested change

	(time.time_ns() - start_time_ns) / 1_000_000
	duration_ms = (time.time_ns() - start_time_ns) / 1_000_000

[cubic-dev-ai]

P2: Potential TypeError during span finalization when headers=None; guard with request_kwargs.get("headers") or {} before dict().

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At drift/instrumentation/httpx/instrumentation.py, line 656:

<comment>Potential TypeError during span finalization when headers=None; guard with `request_kwargs.get("headers") or {}` before dict().</comment>

<file context>
@@ -0,0 +1,992 @@
+            parsed_url = urlparse(url)
+
+            # ===== BUILD INPUT VALUE =====
+            headers = dict(request_kwargs.get("headers", {}))
+            params = dict(request_kwargs.get("params", {})) if request_kwargs.get("params") else {}
+
</file context>

Suggested change

	headers = dict(request_kwargs.get("headers", {}))
	headers = dict(request_kwargs.get("headers") or {})

[cubic-dev-ai]

P2: Potential TypeError when headers=None in async mock lookup; guard with kwargs.get("headers") or {} before dict().

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At drift/instrumentation/httpx/instrumentation.py, line 412:

<comment>Potential TypeError when headers=None in async mock lookup; guard with `kwargs.get("headers") or {}` before dict().</comment>

<file context>
@@ -0,0 +1,992 @@
+            parsed_url = urlparse(url)
+
+            # Extract request data
+            headers = dict(kwargs.get("headers", {}))
+            params = dict(kwargs.get("params", {})) if kwargs.get("params") else {}
+
</file context>

Suggested change

	headers = dict(kwargs.get("headers", {}))
	headers = dict(kwargs.get("headers") or {})