fix: Don't hang when capturing long stacktrace (#4191)

Fixes #2764

---------

Co-authored-by: Anton Pirker <[email protected]>

Author: szokeasaurusrex

sentry_sdk/_types.py

@@ -47,11 +47,14 @@ def removed_because_raw_data(cls):
         )
 
     @classmethod
-    def removed_because_over_size_limit(cls):
-        # type: () -> AnnotatedValue
-        """The actual value was removed because the size of the field exceeded the configured maximum size (specified with the max_request_body_size sdk option)"""
+    def removed_because_over_size_limit(cls, value=""):
+        # type: (Any) -> AnnotatedValue
+        """
+        The actual value was removed because the size of the field exceeded the configured maximum size,
+        for example specified with the max_request_body_size sdk option.
+        """
         return AnnotatedValue(
-            value="",
+            value=value,
             metadata={
                 "rem": [  # Remark
                     [

sentry_sdk/client.py

@@ -755,6 +755,8 @@ def _update_session_from_event(
         if exceptions:
             errored = True
             for error in exceptions:
+                if isinstance(error, AnnotatedValue):
+                    error = error.value or {}
                 mechanism = error.get("mechanism")
                 if isinstance(mechanism, Mapping) and mechanism.get("handled") is False:
                     crashed = True

sentry_sdk/utils.py

@@ -77,6 +77,15 @@
 FALSY_ENV_VALUES = frozenset(("false", "f", "n", "no", "off", "0"))
 TRUTHY_ENV_VALUES = frozenset(("true", "t", "y", "yes", "on", "1"))
 
+MAX_STACK_FRAMES = 2000
+"""Maximum number of stack frames to send to Sentry.
+
+If we have more than this number of stack frames, we will stop processing
+the stacktrace to avoid getting stuck in a long-lasting loop. This value
+exceeds the default sys.getrecursionlimit() of 1000, so users will only
+be affected by this limit if they have a custom recursion limit.
+"""
+
 
 def env_to_bool(value, *, strict=False):
     # type: (Any, Optional[bool]) -> bool | None
@@ -732,10 +741,23 @@ def single_exception_from_error_tuple(
             max_value_length=max_value_length,
             custom_repr=custom_repr,
         )
-        for tb in iter_stacks(tb)
+        # Process at most MAX_STACK_FRAMES + 1 frames, to avoid hanging on
+        # processing a super-long stacktrace.
+        for tb, _ in zip(iter_stacks(tb), range(MAX_STACK_FRAMES + 1))
     ]  # type: List[Dict[str, Any]]
 
-    if frames:
+    if len(frames) > MAX_STACK_FRAMES:
+        # If we have more frames than the limit, we remove the stacktrace completely.
+        # We don't trim the stacktrace here because we have not processed the whole
+        # thing (see above, we stop at MAX_STACK_FRAMES + 1). Normally, Relay would
+        # intelligently trim by removing frames in the middle of the stacktrace, but
+        # since we don't have the whole stacktrace, we can't do that. Instead, we
+        # drop the entire stacktrace.
+        exception_value["stacktrace"] = AnnotatedValue.removed_because_over_size_limit(
+            value=None
+        )
+
+    elif frames:
         if not full_stack:
             new_frames = frames
         else:
@@ -941,7 +963,7 @@ def to_string(value):
 
 
 def iter_event_stacktraces(event):
-    # type: (Event) -> Iterator[Dict[str, Any]]
+    # type: (Event) -> Iterator[Annotated[Dict[str, Any]]]
     if "stacktrace" in event:
         yield event["stacktrace"]
     if "threads" in event:
@@ -950,20 +972,26 @@ def iter_event_stacktraces(event):
                 yield thread["stacktrace"]
     if "exception" in event:
         for exception in event["exception"].get("values") or ():
-            if "stacktrace" in exception:
+            if isinstance(exception, dict) and "stacktrace" in exception:
                 yield exception["stacktrace"]
 
 
 def iter_event_frames(event):
     # type: (Event) -> Iterator[Dict[str, Any]]
     for stacktrace in iter_event_stacktraces(event):
+        if isinstance(stacktrace, AnnotatedValue):
+            stacktrace = stacktrace.value or {}
+
         for frame in stacktrace.get("frames") or ():
             yield frame
 
 
 def handle_in_app(event, in_app_exclude=None, in_app_include=None, project_root=None):
     # type: (Event, Optional[List[str]], Optional[List[str]], Optional[str]) -> Event
     for stacktrace in iter_event_stacktraces(event):
+        if isinstance(stacktrace, AnnotatedValue):
+            stacktrace = stacktrace.value or {}
+
         set_in_app_in_frames(
             stacktrace.get("frames"),
             in_app_exclude=in_app_exclude,

tests/test_basics.py

@@ -1065,3 +1065,47 @@ def __str__(self):
     (event,) = events
 
     assert event["exception"]["values"][0]["value"] == "aha!\nnote 1\nnote 3"
+
+
+@pytest.mark.skipif(
+    sys.version_info < (3, 11),
+    reason="this test appears to cause a segfault on Python < 3.11",
+)
+def test_stacktrace_big_recursion(sentry_init, capture_events):
+    """
+    Ensure that if the recursion limit is increased, the full stacktrace is not captured,
+    as it would take too long to process the entire stack trace.
+    Also, ensure that the capturing does not take too long.
+    """
+    sentry_init()
+    events = capture_events()
+
+    def recurse():
+        recurse()
+
+    old_recursion_limit = sys.getrecursionlimit()
+
+    try:
+        sys.setrecursionlimit(100_000)
+        recurse()
+    except RecursionError as e:
+        capture_start_time = time.perf_counter_ns()
+        sentry_sdk.capture_exception(e)
+        capture_end_time = time.perf_counter_ns()
+    finally:
+        sys.setrecursionlimit(old_recursion_limit)
+
+    (event,) = events
+
+    assert event["exception"]["values"][0]["stacktrace"] is None
+    assert event["_meta"] == {
+        "exception": {
+            "values": {"0": {"stacktrace": {"": {"rem": [["!config", "x"]]}}}}
+        }
+    }
+
+    # On my machine, it takes about 100-200ms to capture the exception,
+    # so this limit should be generous enough.
+    assert (
+        capture_end_time - capture_start_time < 10**9
+    ), "stacktrace capture took too long, check that frame limit is set correctly"