fix: don't log @validate_call function exceptions as validation errors

logfire/integrations/pydantic.py

@@ -75,6 +75,7 @@ class _ValidateWrapper:
         'schema_name',
         '_record',
         '_logfire',
+        '_schema_kind',
     )
 
     def __init__(
@@ -85,13 +86,15 @@ def __init__(
         _plugin_settings: PluginSettings | dict[str, Any],
         schema_type_path: SchemaTypePath,
         record: Literal['all', 'failure', 'metrics'],
+        schema_kind: SchemaKind,
     ) -> None:
         self.validation_method = validation_method
 
         # We accept the schema, config, and plugin_settings in the init since these are the things
         # that are currently exposed by the plugin to potentially configure the validation handlers.
         self.schema_name = get_schema_name(schema)
         self._record = record
+        self._schema_kind = schema_kind
 
         self._logfire = logfire.DEFAULT_LOGFIRE_INSTANCE
         # trace_sample_rate = _plugin_settings.get('logfire', {}).get('trace_sample_rate')
@@ -132,6 +135,15 @@ def wrapped_validator(input_data: Any, *args: Any, **kwargs: Any) -> Any:
                     self._count_validation(success=False)
                     raise
                 except Exception as exception:
+                    # For validate_call, non-ValidationError exceptions are likely from the decorated
+                    # function's execution, not from validation itself. We should not record these
+                    # as validation errors since the arguments validated successfully.
+                    # See https://github.com/pydantic/logfire/issues/1516
+                    if self._schema_kind == 'validate_call':
+                        # End the span successfully since validation passed
+                        self._on_success(span, None)
+                        self._count_validation(success=True)
+                        raise
                     self._on_exception_span(span, exception)
                     self._count_validation(success=False)
                     raise
@@ -156,6 +168,13 @@ def wrapped_validator(input_data: Any, *args: Any, **kwargs: Any) -> Any:
                     self._on_error_log(error)
                     raise
                 except Exception as exception:
+                    # For validate_call, non-ValidationError exceptions are likely from the decorated
+                    # function's execution, not from validation itself. We should not record these
+                    # as validation errors since the arguments validated successfully.
+                    # See https://github.com/pydantic/logfire/issues/1516
+                    if self._schema_kind == 'validate_call':
+                        self._count_validation(success=True)
+                        raise
                     self._count_validation(success=False)
                     self._on_exception_log(exception)
                     raise
@@ -173,7 +192,17 @@ def wrapped_validator(input_data: Any, *args: Any, **kwargs: Any) -> Any:
 
                 try:
                     result = validator(input_data, *args, **kwargs)
+                except ValidationError:
+                    self._count_validation(success=False)
+                    raise
                 except Exception:
+                    # For validate_call, non-ValidationError exceptions are likely from the decorated
+                    # function's execution, not from validation itself. We should count this as
+                    # successful validation since the arguments validated successfully.
+                    # See https://github.com/pydantic/logfire/issues/1516
+                    if self._schema_kind == 'validate_call':
+                        self._count_validation(success=True)
+                        raise
                     self._count_validation(success=False)
                     raise
                 else:
@@ -348,9 +377,9 @@ def new_schema_validator(
             if _include_model(schema_type_path):
                 _patch_build_wrapper()
                 return (
-                    _ValidateWrapper('validate_python', schema, config, plugin_settings, schema_type_path, record),
-                    _ValidateWrapper('validate_json', schema, config, plugin_settings, schema_type_path, record),
-                    _ValidateWrapper('validate_strings', schema, config, plugin_settings, schema_type_path, record),
+                    _ValidateWrapper('validate_python', schema, config, plugin_settings, schema_type_path, record, schema_kind),
+                    _ValidateWrapper('validate_json', schema, config, plugin_settings, schema_type_path, record, schema_kind),
+                    _ValidateWrapper('validate_strings', schema, config, plugin_settings, schema_type_path, record, schema_kind),
                 )
 
             return None, None, None

tests/test_pydantic_plugin.py

@@ -1301,3 +1301,125 @@ class Hero(sqlmodel.SQLModel, table=True):
             }
         ]
     )
+
+
+def test_validate_call_exception_not_logged_as_validation_error(exporter: TestExporter) -> None:
+    """Test that exceptions raised inside @validate_call decorated functions are NOT logged as validation errors.
+
+    This is a regression test for https://github.com/pydantic/logfire/issues/1516
+    """
+    from pydantic import validate_call
+
+    @validate_call(config=ConfigDict(plugin_settings={'logfire': {'record': 'all'}}))
+    def my_test_function(value: int) -> None:
+        raise RuntimeError('Test error from function')
+
+    with pytest.raises(RuntimeError, match='Test error from function'):
+        my_test_function(1)
+
+    # The span should show successful validation, not a validation error
+    spans = exporter.exported_spans_as_dict()
+    assert len(spans) == 1
+    span = spans[0]
+
+    # The validation should be marked as successful since arguments validated correctly
+    assert span['attributes']['success'] is True
+    # Should not have error-related attributes
+    assert 'error_count' not in span['attributes']
+    assert 'errors' not in span['attributes']
+    # Should NOT have exception events (unlike when a validator raises an exception)
+    assert 'events' not in span or len(span.get('events', [])) == 0
+    # Message should show success, not an error
+    assert 'succeeded' in span['attributes']['logfire.msg']
+
+
+def test_validate_call_exception_record_failure_mode(exporter: TestExporter, metrics_reader: InMemoryMetricReader) -> None:
+    """Test that @validate_call exceptions are not logged in 'failure' mode.
+
+    In failure mode, only validation failures should be logged, not exceptions
+    from the decorated function's execution.
+    """
+    from pydantic import validate_call
+
+    @validate_call(config=ConfigDict(plugin_settings={'logfire': {'record': 'failure'}}))
+    def my_test_function(value: int) -> None:
+        raise RuntimeError('Test error from function')
+
+    with pytest.raises(RuntimeError, match='Test error from function'):
+        my_test_function(1)
+
+    # No spans should be exported since this is not a validation failure
+    assert exporter.exported_spans_as_dict(_include_pending_spans=True) == []
+
+    # Metrics should show successful validation
+    metrics_collected = get_collected_metrics(metrics_reader)
+    assert len(metrics_collected) == 1
+    data_points = metrics_collected[0]['data']['data_points']
+    assert len(data_points) == 1
+    assert data_points[0]['attributes']['success'] is True
+
+
+def test_validate_call_validation_error_still_logged(exporter: TestExporter) -> None:
+    """Test that actual validation errors in @validate_call are still logged correctly."""
+    from pydantic import validate_call
+
+    @validate_call(config=ConfigDict(plugin_settings={'logfire': {'record': 'all'}}))
+    def my_test_function(value: int) -> None:
+        pass
+
+    with pytest.raises(ValidationError):
+        my_test_function('not an int')  # type: ignore
+
+    # The span should show a validation failure
+    spans = exporter.exported_spans_as_dict()
+    assert len(spans) == 1
+    span = spans[0]
+
+    # The validation should be marked as failed
+    assert span['attributes']['success'] is False
+    # Should have error attributes
+    assert span['attributes']['error_count'] == 1
+    assert 'errors' in span['attributes']
+    # Message should show failure
+    assert 'failed' in span['attributes']['logfire.msg']
+
+
+def test_validate_call_validator_exception_not_logged_as_validation_error(exporter: TestExporter) -> None:
+    """Test that validator exceptions in @validate_call are NOT logged as validation errors.
+
+    Due to Pydantic's current design, we cannot distinguish between exceptions raised
+    by validators and exceptions raised by the decorated function. Both are raised
+    without being wrapped in ValidationError for @validate_call.
+
+    Since we cannot distinguish, we take the conservative approach of not logging
+    any non-ValidationError exceptions as validation errors for @validate_call.
+    This prevents function exceptions from being incorrectly logged as validation
+    errors (the issue reported in #1516), at the cost of also not logging validator
+    exceptions.
+
+    The proper fix would be in Pydantic itself, to separate argument validation
+    from function execution.
+    """
+
+    from pydantic import validate_call
+
+    def raise_error(v: int) -> int:
+        raise TypeError('Validator error')
+
+    @validate_call(config=ConfigDict(plugin_settings={'logfire': {'record': 'all'}}))
+    def my_test_function(value: Annotated[int, AfterValidator(raise_error)]) -> None:
+        pass
+
+    with pytest.raises(TypeError, match='Validator error'):
+        my_test_function(1)
+
+    # The span should show successful validation because we can't distinguish
+    # validator exceptions from function exceptions for @validate_call
+    spans = exporter.exported_spans_as_dict()
+    assert len(spans) == 1
+    span = spans[0]
+
+    # The validation is marked as successful (conservative approach)
+    assert span['attributes']['success'] is True
+    # Should NOT have exception events
+    assert 'events' not in span or len(span.get('events', [])) == 0