ref(grouping): Skip exception filtering for new Python SDK events (#95272)

The new version of the Python SDK (version 3) sends `mechanism` data with exceptions which the old version didn't used to. As a result, certain exception groups are hashing differently because they can now use the `filter_exceptions_for_exception_groups` helper in the newstyle grouping config.

In the long run, we need to figure out how to compensate for this on the server side, so that we can match up old hashes with new hashes. In the short run, in order not to block the release of the SDK, we're just going to pretend the change didn't happen, by bailing out of the helper early for events from the new version of the SDK.

Author: lobsterkatie

src/sentry/grouping/strategies/newstyle.py

@@ -35,6 +35,7 @@
 from sentry.interfaces.stacktrace import Frame, Stacktrace
 from sentry.interfaces.threads import Threads
 from sentry.stacktraces.platform import get_behavior_family_for_platform
+from sentry.utils.safe import get_path
 
 if TYPE_CHECKING:
     from sentry.eventstore.models import Event
@@ -679,6 +680,17 @@ def filter_exceptions_for_exception_groups(
     if len(exceptions) <= 1:
         return exceptions
 
+    # TODO: Get rid of this hack!
+    #
+    # A change in the python SDK between version 2 and version 3 means that suddenly this function
+    # applies where it didn't used to, which in turn changes how some exception groups are hashed.
+    # As a temporary stopgap, until we can build a system akin to the grouping config transition
+    # system to compensate for the change, we're just emulating the old behavior.
+    if event.platform == "python" and get_path(event.data, "sdk", "version", default="").startswith(
+        "3"
+    ):
+        return exceptions
+
     # Reconstruct the tree of exceptions if the required data is present.
     class ExceptionTreeNode:
         def __init__(

tests/sentry/grouping/test_strategies.py

@@ -0,0 +1,68 @@
+from typing import Any
+
+from sentry.eventstore.models import Event
+from sentry.testutils.cases import TestCase
+
+
+class ChainedExceptionTest(TestCase):
+    def test_ignores_mechanism_in_python_sdk_version_3_chained_exception_events(self):
+        # First, get hashes for an event with no `mechanism` data
+        event_data: dict[str, Any] = {
+            "platform": "python",
+            "sdk": {"name": "python", "version": "3.1"},
+            "exception": {
+                "values": [
+                    {"type": "FetchError", "value": "Charlie didn't bring the ball back"},
+                    {"type": "ShoeError", "value": "Oh, no! Charlie ate the flip-flops!"},
+                    {"type": "AggregateException", "value": "She's a very good dog, but..."},
+                ]
+            },
+        }
+
+        no_mechanism_hashes = Event(
+            event_id="1121123104150908",
+            project_id=self.project.id,
+            data=event_data,
+        ).get_hashes()
+
+        # Now add in `mechanism` data, and we'll see that the hash doesn't change
+        event_data["exception"]["values"][0]["mechanism"] = {
+            "type": "chained",
+            "handled": True,
+            "source": "InnerExceptions[1]",
+            "exception_id": 2,
+            "parent_id": 0,
+        }
+        event_data["exception"]["values"][1]["mechanism"] = {
+            "type": "chained",
+            "handled": True,
+            "source": "InnerExceptions[0]",
+            "exception_id": 1,
+            "parent_id": 0,
+        }
+        event_data["exception"]["values"][2]["mechanism"] = {
+            "type": "AppDomain.UnhandledException",
+            "handled": False,
+            "is_exception_group": True,
+            "exception_id": 0,
+        }
+
+        with_mechanism_hashes = Event(
+            event_id="1231112109080415",
+            project_id=self.project.id,
+            data=event_data,
+        ).get_hashes()
+
+        assert no_mechanism_hashes == with_mechanism_hashes
+
+        # Just to prove that were it not for the hack, the grouping *would* change with the addition
+        # of mechanism data, we switch the platform
+        event_data["platform"] = "javascript"
+
+        js_with_mechanism_hashes = Event(
+            event_id="1121201212312012",
+            project_id=self.project.id,
+            data=event_data,
+        ).get_hashes()
+
+        assert js_with_mechanism_hashes != no_mechanism_hashes