feat(ai): Add `python-genai` integration #4891

vgrozdanic · 2025-10-06T12:51:40Z

Adds support for python-genai integrations. It supports both sync and async clients, and both regular and streaming modes for interacting with models and building agents.

Closes PY-1733: Add agent monitoring support for google-genai

linear · 2025-10-07T11:01:12Z

PY-1733 Add agent monitoring support for `google-genai`

seer-by-sentry · 2025-10-07T11:05:56Z

sentry_sdk/integrations/google_genai/utils.py

+        ("frequency_penalty", SPANDATA.GEN_AI_REQUEST_FREQUENCY_PENALTY),
+        ("seed", SPANDATA.GEN_AI_REQUEST_SEED),
+    ]:
+        if hasattr(config, param):


Potential bug: The function set_span_data_for_request calls hasattr(config, ...) without checking if config is None, causing an AttributeError if the optional config parameter is omitted.

Description: In set_span_data_for_request, the config object is retrieved from kwargs using kwargs.get("config"). If a user does not provide a config object in their API call, which is a valid use case, config will be None. The subsequent loop attempts to access attributes on this None object with hasattr(config, param), which raises an AttributeError. Because this integration code runs within a user's application, this unhandled exception will crash the application.

Suggested fix: Before the loop that iterates over configuration parameters in set_span_data_for_request, add a check to ensure config is not None. If config is None, the entire loop that sets span data from the config should be skipped.
_{severity: 0.85, confidence: 0.98}

_{Did we get this right? 👍 / 👎 to inform future reviews.}

hasattr(None, 'something') won't raise an AttributeError, it'll simply be False

sentrivana

TY for the PR Vjeran!

I'll do a full review shortly, just a couple of things I noticed on the fly:

The new test suite is not running in CI. Check https://github.com/getsentry/sentry-python/blob/master/scripts/populate_tox/README.md#add-a-new-test-suite (the Add a new test suite section specifically) for what's missing in order to make this happen
Have a look at what mypy's complaining about in the Lint Sources step in CI

Lmk if anything is unclear.

sentrivana

👋🏻 Back with a big review. Overall looks nice and clean! Don't be fooled by the number of comments, it's just that it's a big PR, and I'm pointing out some stuff multiple times.

sentrivana · 2025-10-07T11:43:29Z

sentry_sdk/integrations/__init__.py

    "sentry_sdk.integrations.flask.FlaskIntegration",
    "sentry_sdk.integrations.gql.GQLIntegration",
    "sentry_sdk.integrations.graphene.GrapheneIntegration",
+    "sentry_sdk.integrations.google_genai.GoogleGenAIIntegration",


Adding the integration here will make it auto-enabling (the user wouldn't have to add GoogleGenAIIntegration manually to their init) -- however for new integrations, I strongly prefer to make them opt-in first, in order to iron out issues, and only make them auto-enabling after some time.

sentrivana · 2025-10-07T11:44:31Z

setup.py

        "statsig": ["statsig>=0.55.3"],
        "tornado": ["tornado>=6"],
        "unleash": ["UnleashClient>=6.0.1"],
+        "google-genai": ["google-genai"],


Suggested change

"google-genai": ["google-genai"],

"google-genai": ["google-genai>=1.0.0"],

We should also have the lower bound here, to make sure it's not possible to install the sdk with an unsupported version of google-genai.

sentrivana · 2025-10-08T07:07:07Z

sentry_sdk/integrations/google_genai/streaming.py

+    from sentry_sdk.tracing import Span
+
+
+def prepare_generate_content_args(args, kwargs):


Can we move this function to google_genai/utils.py since it's also used by the non-streaming wrappers?

sentrivana · 2025-10-08T07:19:55Z

sentry_sdk/integrations/google_genai/utils.py

+        ("frequency_penalty", SPANDATA.GEN_AI_REQUEST_FREQUENCY_PENALTY),
+        ("seed", SPANDATA.GEN_AI_REQUEST_SEED),
+    ]:
+        if hasattr(config, param):


hasattr(None, 'something') won't raise an AttributeError, it'll simply be False

sentry_sdk/integrations/google_genai/__init__.py

sentrivana · 2025-10-08T09:02:14Z

sentry_sdk/integrations/google_genai/utils.py

+    if kwargs.get("stream", False):
+        span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)


Do we also want to add the attribute for non-streaming responses?

Suggested change

if kwargs.get("stream", False):

span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)

span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, kwargs.get("stream") or False)

sentrivana · 2025-10-08T09:07:22Z

sentry_sdk/integrations/google_genai/utils.py

+
+                    return result
+                except Exception as exc:
+                    capture_exception(exc)


Should probably set span status to error here, also in the sync version below.

sentrivana · 2025-10-08T09:09:40Z

tests/integrations/google_genai/test_google_genai.py

+try:
+    from google import genai
+    from google.genai import types as genai_types
+    from google.genai.models import Models
+except ImportError:
+    # If google.genai is not installed, skip the tests
+    pytest.skip("google-genai not installed", allow_module_level=True)


The usual way of checking whether the correct package is installed in a test suite is to put an importorskip in __init__.py, see e.g. https://github.com/getsentry/sentry-python/blob/master/tests/integrations/celery/__init__.py, and then you don't need to guard the imports in the test file.

sentrivana · 2025-10-08T09:21:07Z

tests/integrations/google_genai/test_google_genai.py

+    yield models_instance
+
+
+def setup_mock_generate_content(mock_instance, return_value=None, side_effect=None):


Ideally the tests would test that the integration does the wrapping correctly. This util seems to apply the integration wrapper on its own?

sentrivana · 2025-10-08T09:26:01Z

tests/integrations/google_genai/test_google_genai.py

+
+
+@pytest.fixture
+def mock_models_instance():


I see we're mocking/populating a lot of internal google genai stuff in the tests. Generally, in tests we try to be as end-to-end as possible, with as little mocking as possible -- mocking takes away our ability to detect regressions/incompatibilities if something changes upstream.

I know these AI integrations are a pain to test in general, because we can't make actual requests to the API. However, even in such cases we should generally aim for mocking as early in the chain as possible (e.g., just mocking the actual json response from the API, but leaving the actual machinery of the library that uses that response intact).

Do you think it's possible to set the tests up to mock less of the internals?

Hmm, let me try to do that, I haven't even thought of it, but definitely makes sense to do it

cursor

Bug: Token Count Inflation in Streaming Responses

The accumulate_streaming_response function incorrectly sums token counts from streaming chunks. Since these counts (especially input and cached tokens) are often cumulative or constant across chunks, summing them leads to inflated and inaccurate usage metrics.

sentry_sdk/integrations/google_genai/streaming.py#L507-L526

https://github.com/getsentry/sentry-python/blob/1c5aa1c08955f7271a7d8b5397874e616b2f3819/sentry_sdk/integrations/google_genai/streaming.py#L507-L526

sentry_sdk/integrations/google_genai/streaming.py#L69-L76

sentry-python/sentry_sdk/integrations/google_genai/streaming.py

Lines 69 to 76 in 1c5aa1c

    
           # Accumulate token usage 
        
           extracted_usage_data = extract_usage_data(chunk) 
        
           total_input_tokens += extracted_usage_data["input_tokens"] 
        
           total_output_tokens += extracted_usage_data["output_tokens"] 
        
           total_cached_tokens += extracted_usage_data["input_tokens_cached"] 
        
           total_reasoning_tokens += extracted_usage_data["output_tokens_reasoning"] 
        
           total_tokens += extracted_usage_data["total_tokens"]

sentry_sdk/integrations/google_genai/streaming.py

cursor · 2025-10-09T11:43:29Z

sentry_sdk/integrations/google_genai/utils.py

+    texts = []
+    for candidate in response.candidates:
+        if not hasattr(candidate, "content") or not hasattr(candidate.content, "parts"):
+            continue


Bug: Nested Attribute Access Fails

The code can raise an AttributeError when accessing nested attributes, such as candidate.content.parts. This occurs if an intermediate attribute (like content) is missing, or if the final attribute (parts) is None when an iterable is expected. This pattern is visible in _extract_response_text.

Additional Locations (1)

sentry_sdk/integrations/google_genai/utils.py#L255-L257

sentry_sdk/integrations/google_genai/streaming.py

vgrozdanic · 2025-10-09T12:39:43Z

@sentrivana i have made the changes regarding all the feedback, and did some more refactoring, to make the code more consistent and a bit easier to read.

Also, now the pipeline is finally green 🥳

vgrozdanic force-pushed the vg/add-python-genai-integration branch 7 times, most recently from e265d62 to 48a0c71 Compare October 7, 2025 10:57

vgrozdanic marked this pull request as ready for review October 7, 2025 10:58

vgrozdanic requested a review from a team as a code owner October 7, 2025 10:58

seer-by-sentry bot reviewed Oct 7, 2025

View reviewed changes

sentrivana reviewed Oct 7, 2025

View reviewed changes

This comment was marked as outdated.

Sign in to view

sentrivana reviewed Oct 8, 2025

View reviewed changes

vgrozdanic added 6 commits October 8, 2025 16:41

feat(ai): Add python-genai integration

fb1bc54

fix test running

37919fb

refactor

a5b517c

mock API response for tests

1492349

feedback

3d165f3

apply all the feedback

9e03a64

vgrozdanic force-pushed the vg/add-python-genai-integration branch from f0f32d2 to 9e03a64 Compare October 8, 2025 15:37

This comment was marked as outdated.

Sign in to view

vgrozdanic added 3 commits October 9, 2025 11:06

refactor streaming

a34799c

regenerate test files

e003535

mypy fixes

dc0a147

This comment was marked as outdated.

Sign in to view

fix linter

2010a6c

This comment was marked as outdated.

Sign in to view

vgrozdanic added 2 commits October 9, 2025 11:48

more fixes

bc2a62f

more fixes

bbcb2a0

This comment was marked as outdated.

Sign in to view

vgrozdanic added 2 commits October 9, 2025 12:16

comment clean up

174198d

fix test execution

8270e0d

This comment was marked as outdated.

Sign in to view

fix ci

f33afb2

This comment was marked as outdated.

Sign in to view

fix ci attempt 2

1c5aa1c

cursor bot reviewed Oct 9, 2025

View reviewed changes

vgrozdanic requested a review from sentrivana October 9, 2025 12:39

	"google-genai": ["google-genai"],
	"google-genai": ["google-genai>=1.0.0"],

		from sentry_sdk.tracing import Span


		def prepare_generate_content_args(args, kwargs):

		if kwargs.get("stream", False):
		span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)

		yield models_instance


		def setup_mock_generate_content(mock_instance, return_value=None, side_effect=None):


	# Accumulate token usage
	extracted_usage_data = extract_usage_data(chunk)
	total_input_tokens += extracted_usage_data["input_tokens"]
	total_output_tokens += extracted_usage_data["output_tokens"]
	total_cached_tokens += extracted_usage_data["input_tokens_cached"]
	total_reasoning_tokens += extracted_usage_data["output_tokens_reasoning"]
	total_tokens += extracted_usage_data["total_tokens"]

feat(ai): Add python-genai integration #4891

Are you sure you want to change the base?

feat(ai): Add python-genai integration #4891

Conversation

vgrozdanic commented Oct 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

linear bot commented Oct 7, 2025

Uh oh!

seer-by-sentry bot Oct 7, 2025

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sentrivana left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment was marked as outdated.

Uh oh!

sentrivana left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Bug: Token Count Inflation in Streaming Responses

Uh oh!

Uh oh!

cursor bot Oct 9, 2025

Choose a reason for hiding this comment

Bug: Nested Attribute Access Fails

Uh oh!

Uh oh!

vgrozdanic commented Oct 9, 2025

Uh oh!

Uh oh!

feat(ai): Add `python-genai` integration #4891

feat(ai): Add `python-genai` integration #4891

vgrozdanic commented Oct 6, 2025 •

edited

Loading

sentrivana left a comment •

edited

Loading