feat(otlp): Optionally capture exceptions from otel's Span.record_exception api (#5235)

#### Issues

* resolves: #5097
* resolves: PY-1973

Author: sl0thentr0py

sentry_sdk/integrations/otlp.py

@@ -1,7 +1,12 @@
-from sentry_sdk import get_client
+from sentry_sdk import get_client, capture_event
 from sentry_sdk.integrations import Integration, DidNotEnable
 from sentry_sdk.scope import register_external_propagation_context
-from sentry_sdk.utils import logger, Dsn
+from sentry_sdk.utils import (
+    Dsn,
+    logger,
+    event_from_exception,
+    capture_internal_exceptions,
+)
 from sentry_sdk.consts import VERSION, EndpointType
 from sentry_sdk.tracing_utils import Baggage
 from sentry_sdk.tracing import (
@@ -11,7 +16,7 @@
 
 try:
     from opentelemetry.propagate import set_global_textmap
-    from opentelemetry.sdk.trace import TracerProvider
+    from opentelemetry.sdk.trace import TracerProvider, Span
     from opentelemetry.sdk.trace.export import BatchSpanProcessor
     from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
 
@@ -82,6 +87,38 @@ def setup_otlp_traces_exporter(dsn: "Optional[str]" = None) -> None:
     tracer_provider.add_span_processor(span_processor)
 
 
+_sentry_patched_exception = False
+
+
+def setup_capture_exceptions() -> None:
+    """
+    Intercept otel's Span.record_exception to automatically capture those exceptions in Sentry.
+    """
+    global _sentry_patched_exception
+    _original_record_exception = Span.record_exception
+
+    if _sentry_patched_exception:
+        return
+
+    def _sentry_patched_record_exception(
+        self: "Span", exception: "BaseException", *args: "Any", **kwargs: "Any"
+    ) -> None:
+        otlp_integration = get_client().get_integration(OTLPIntegration)
+        if otlp_integration and otlp_integration.capture_exceptions:
+            with capture_internal_exceptions():
+                event, hint = event_from_exception(
+                    exception,
+                    client_options=get_client().options,
+                    mechanism={"type": OTLPIntegration.identifier, "handled": False},
+                )
+                capture_event(event, hint=hint)
+
+        _original_record_exception(self, exception, *args, **kwargs)
+
+    Span.record_exception = _sentry_patched_record_exception  # type: ignore[method-assign]
+    _sentry_patched_exception = True
+
+
 class SentryOTLPPropagator(SentryPropagator):
     """
     We need to override the inject of the older propagator since that
@@ -136,13 +173,28 @@ def _to_traceparent(span_context: "SpanContext") -> str:
 
 
 class OTLPIntegration(Integration):
+    """
+    Automatically setup OTLP ingestion from the DSN.
+
+    :param setup_otlp_traces_exporter: Automatically configure an Exporter to send OTLP traces from the DSN, defaults to True.
+        Set to False if using a custom collector or to setup the TracerProvider manually.
+    :param setup_propagator: Automatically configure the Sentry Propagator for Distributed Tracing, defaults to True.
+        Set to False to configure propagators manually or to disable propagation.
+    :param capture_exceptions: Intercept and capture exceptions on the OpenTelemetry Span in Sentry as well, defaults to False.
+        Set to True to turn on capturing but be aware that since Sentry captures most exceptions, duplicate exceptions might be dropped by DedupeIntegration in many cases.
+    """
+
     identifier = "otlp"
 
     def __init__(
-        self, setup_otlp_traces_exporter: bool = True, setup_propagator: bool = True
+        self,
+        setup_otlp_traces_exporter: bool = True,
+        setup_propagator: bool = True,
+        capture_exceptions: bool = False,
     ) -> None:
         self.setup_otlp_traces_exporter = setup_otlp_traces_exporter
         self.setup_propagator = setup_propagator
+        self.capture_exceptions = capture_exceptions
 
     @staticmethod
     def setup_once() -> None:
@@ -161,3 +213,5 @@ def setup_once_with_options(
             logger.debug("[OTLP] Setting up propagator for distributed tracing")
             # TODO-neel better propagator support, chain with existing ones if possible instead of replacing
             set_global_textmap(SentryOTLPPropagator())
+
+        setup_capture_exceptions()

tests/integrations/otlp/test_otlp.py

@@ -24,6 +24,25 @@
 original_propagator = get_global_textmap()
 
 
+@pytest.fixture(autouse=True)
+def mock_otlp_ingest():
+    responses.start()
+    responses.add(
+        responses.POST,
+        url="https://bla.ingest.sentry.io/api/12312012/integration/otlp/v1/traces/",
+        status=200,
+    )
+
+    yield
+
+    tracer_provider = get_tracer_provider()
+    if isinstance(tracer_provider, TracerProvider):
+        tracer_provider.force_flush()
+
+    responses.stop()
+    responses.reset()
+
+
 @pytest.fixture(autouse=True)
 def reset_otlp(uninstall_integration):
     trace._TRACER_PROVIDER_SET_ONCE = Once()
@@ -127,14 +146,7 @@ def test_does_not_set_propagator_if_disabled(sentry_init):
     assert propagator is original_propagator
 
 
-@responses.activate
 def test_otel_propagation_context(sentry_init):
-    responses.add(
-        responses.POST,
-        url="https://bla.ingest.sentry.io/api/12312012/integration/otlp/v1/traces/",
-        status=200,
-    )
-
     sentry_init(
         dsn="https://[email protected]/12312012",
         integrations=[OTLPIntegration()],
@@ -145,9 +157,6 @@ def test_otel_propagation_context(sentry_init):
         with tracer.start_as_current_span("bar") as span:
             external_propagation_context = get_external_propagation_context()
 
-    # Force flush to ensure spans are exported while mock is active
-    get_tracer_provider().force_flush()
-
     assert external_propagation_context is not None
     (trace_id, span_id) = external_propagation_context
     assert trace_id == format_trace_id(root_span.get_span_context().trace_id)
@@ -222,3 +231,74 @@ def test_propagator_inject_continue_trace(sentry_init):
         assert carrier["baggage"] == incoming_headers["baggage"]
 
     detach(token)
+
+
+def test_capture_exceptions_enabled(sentry_init, capture_events):
+    sentry_init(
+        dsn="https://[email protected]/12312012",
+        integrations=[OTLPIntegration(capture_exceptions=True)],
+    )
+
+    events = capture_events()
+
+    tracer = trace.get_tracer(__name__)
+    with tracer.start_as_current_span("test_span") as span:
+        try:
+            raise ValueError("Test exception")
+        except ValueError as e:
+            span.record_exception(e)
+
+    (event,) = events
+    assert event["exception"]["values"][0]["type"] == "ValueError"
+    assert event["exception"]["values"][0]["value"] == "Test exception"
+    assert event["exception"]["values"][0]["mechanism"]["type"] == "otlp"
+    assert event["exception"]["values"][0]["mechanism"]["handled"] is False
+
+    trace_context = event["contexts"]["trace"]
+    assert trace_context["trace_id"] == format_trace_id(
+        span.get_span_context().trace_id
+    )
+    assert trace_context["span_id"] == format_span_id(span.get_span_context().span_id)
+
+
+def test_capture_exceptions_disabled(sentry_init, capture_events):
+    sentry_init(
+        dsn="https://[email protected]/12312012",
+        integrations=[OTLPIntegration(capture_exceptions=False)],
+    )
+
+    events = capture_events()
+
+    tracer = trace.get_tracer(__name__)
+    with tracer.start_as_current_span("test_span") as span:
+        try:
+            raise ValueError("Test exception")
+        except ValueError as e:
+            span.record_exception(e)
+
+    assert len(events) == 0
+
+
+def test_capture_exceptions_preserves_otel_behavior(sentry_init, capture_events):
+    sentry_init(
+        dsn="https://[email protected]/12312012",
+        integrations=[OTLPIntegration(capture_exceptions=True)],
+    )
+
+    events = capture_events()
+
+    tracer = trace.get_tracer(__name__)
+    with tracer.start_as_current_span("test_span") as span:
+        try:
+            raise ValueError("Test exception")
+        except ValueError as e:
+            span.record_exception(e, attributes={"foo": "bar"})
+
+        # Verify the span recorded the exception (OpenTelemetry behavior)
+        # The span should have events with the exception information
+        (otel_event,) = span._events
+        assert otel_event.name == "exception"
+        assert otel_event.attributes["foo"] == "bar"
+
+    # verify sentry also captured it
+    assert len(events) == 1