flask: Clear context at the end of the request in non-streaming scenarios. (#4341)

* Initial plan

* Clean up _ENVIRON_ACTIVATION_KEY and _ENVIRON_TOKEN in _teardown_request to prevent duplicate execution

Add a finally block in _teardown_request to pop _ENVIRON_ACTIVATION_KEY and
_ENVIRON_TOKEN from flask.request.environ after teardown completes. This
prevents issues when _teardown_request is called multiple times, as the
second call will see activation as None and return early.

Co-authored-by: 123liuziming <32130965+123liuziming@users.noreply.github.com>

* Improve test to explicitly verify duplicate teardown calls don't cause errors

Co-authored-by: 123liuziming <32130965+123liuziming@users.noreply.github.com>

* fix lint

Change-Id: Iea7711d86dcd6f4da7a3af92cb68db2fe3db345f
Co-developed-by: Cursor <noreply@cursor.com>

* polish CHANGELOG

Change-Id: Ie89cac244ade8d04768503112db8673f91d81756
Co-developed-by: Cursor <noreply@cursor.com>

* Refactor request teardown to avoid duplicate cleanup

Remove redundant environment variable cleanup in finally block.

* Update test_flask_compatibility.py

* Uncomment FlaskInstrumentor import in test file

* Initial plan

* Fix ruff format errors in docker test files

Co-authored-by: 123liuziming <32130965+123liuziming@users.noreply.github.com>
Agent-Logs-Url: https://github.com/123liuziming/opentelemetry-python-contrib/sessions/9165702c-5648-474d-9291-dfa57008b0b6

* Initial plan

* Revert docker-test formatting changes that undo merged #4339 cleanup

Agent-Logs-Url: https://github.com/123liuziming/opentelemetry-python-contrib/sessions/25f73137-1606-4050-b879-ffc01d7ab803

Co-authored-by: 123liuziming <32130965+123liuziming@users.noreply.github.com>

* Update CHANGELOG.md

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Emídio Neto <9735060+emdneto@users.noreply.github.com>
Co-authored-by: Mike Goldsmith <goldsmith.mike@gmail.com>
Co-authored-by: Riccardo Magliocchetti <riccardo.magliocchetti@gmail.com>

Author: 5 people

CHANGELOG.md

@@ -24,6 +24,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   ([#4461](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4461))
 - `opentelemetry-instrumentation-dbapi` Use `ObjectProxy` instead of `BaseObjectProxy` for `TracedCursorProxy` to restore iterability with wrapt 2.x
   ([#4427](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4427))
+- `opentelemetry-instrumentation-flask`: Clean up environ keys in `_teardown_request` to prevent duplicate execution
+  ([#4341](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/4341))
 
 ### Breaking changes
 

instrumentation/opentelemetry-instrumentation-flask/src/opentelemetry/instrumentation/flask/__init__.py

@@ -636,6 +636,8 @@ def _teardown_request(exc):
 
             if token:
                 context.detach(token)
+                flask.request.environ.pop(_ENVIRON_ACTIVATION_KEY, None)
+                flask.request.environ.pop(_ENVIRON_TOKEN, None)
 
         except (RuntimeError, AttributeError, ValueError) as teardown_exc:
             # Log the error but don't raise it to avoid breaking the request handling

instrumentation/opentelemetry-instrumentation-flask/tests/test_flask_compatibility.py

@@ -26,6 +26,8 @@
 
 from opentelemetry import trace
 from opentelemetry.instrumentation.flask import (
+    _ENVIRON_ACTIVATION_KEY,
+    _ENVIRON_TOKEN,
     FlaskInstrumentor,
     _request_ctx_ref,
 )
@@ -354,3 +356,59 @@ def generate():
         # This ensures OpenTelemetry contexts are properly cleaned up for streaming responses
         # following Logfire's recommendations (see open-telemetry/opentelemetry-python#2606)
         # If we reach this point, the Flask 3.1+ streaming context cleanup is working
+
+    def test_duplicate_teardown_request_does_not_cause_errors(self):
+        """Test that _teardown_request can be called multiple times without errors.
+
+        Flask may call teardown_request multiple times in some scenarios.
+        After the first call, _ENVIRON_ACTIVATION_KEY and _ENVIRON_TOKEN
+        should be cleaned up so subsequent calls are no-ops.
+        """
+        app = flask.Flask(__name__)
+
+        # Register the check handler BEFORE instrumenting so it runs AFTER
+        # the instrumentation's teardown (Flask uses LIFO order).
+        cleaned_up = {}
+        call_count = {"teardown_calls": 0}
+
+        @app.teardown_request
+        def check_cleanup(exc):
+            cleaned_up["activation_present"] = (
+                _ENVIRON_ACTIVATION_KEY in flask.request.environ
+            )
+            cleaned_up["token_present"] = (
+                _ENVIRON_TOKEN in flask.request.environ
+            )
+
+        FlaskInstrumentor().instrument_app(app)
+
+        # Wrap the instrumentation's teardown to count calls and invoke it
+        # a second time to simulate duplicate teardown.
+        instrumentation_teardown = app.teardown_request_funcs[None][-1]
+
+        def counting_teardown(exc):
+            call_count["teardown_calls"] += 1
+            instrumentation_teardown(exc)
+            # Call it again to simulate duplicate teardown - should not raise
+            instrumentation_teardown(exc)
+
+        app.teardown_request_funcs[None][-1] = counting_teardown
+
+        @app.route("/test")
+        def test_endpoint():
+            return "OK"
+
+        client = app.test_client()
+        response = client.get("/test")
+        self.assertEqual(response.status_code, 200)
+        # Verify the teardown was actually called
+        self.assertGreater(call_count["teardown_calls"], 0)
+        # Verify env keys are cleaned up after teardown
+        self.assertFalse(
+            cleaned_up.get("activation_present", True),
+            "_ENVIRON_ACTIVATION_KEY should be cleaned up after teardown",
+        )
+        self.assertFalse(
+            cleaned_up.get("token_present", True),
+            "_ENVIRON_TOKEN should be cleaned up after teardown",
+        )