fix: handle malformed request inputs in GenAI extractors by sangkyoonnam · Pull Request #866 · open-telemetry/opentelemetry-python-genai

sangkyoonnam · 2026-10-09T07:07:58Z

Description

Fixes request extraction in bedrock, openai, anthropic, portkey and google-genai that raised on malformed input, so the request reaches the SDK and the SDK returns its own error. Each fix sits at the line that raised and uses a type check, safe field access or narrowly guarded numeric conversion: a Converse document.format that isn't a string, chat messages=None, a Responses or interactions.create input item whose type isn't a string, an Anthropic message without role or a system list of plain strings, Portkey sampling or token parameters that don't convert to numbers, and generate_content with contents of None or an empty list.

Known gaps: this covers the listed extraction failures; other malformed input paths and objects whose __str__ raises are out of scope.

Type of change

Bug fix (non-breaking change which fixes an issue)

How has this been tested?

Regression cases pass on this branch and fail against upstream/main source. tox latest and oldest for the five packages, plus typecheck.

Checklist

Followed the style guidelines of this project
Changelog updated if the change requires an entry
Unit tests added

opentelemetry-pr-dashboard · 2026-10-09T07:08:53Z

Pull request dashboard status

Merged · refreshed 2026-10-16 15:55 UTC

Status above doesn't look right?

Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

Copilot

🟢 Approval recommended

The implementation preserves SDK exceptions and invocation finalization with comprehensive regression coverage.

0 open findings

What changed in this PR

Prevents Bedrock request-extraction failures from altering SDK behavior or leaving telemetry invocations unfinished.

Changes:

Adds guarded request extraction with debug logging.
Covers sync/async handlers and regression scenarios.

File	Description
`patch.py`	Safely invokes Bedrock request extractors.
`tests/test_patch.py`	Tests extraction failures, SDK errors, and recovery.

🧠 Review effort: Balanced

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

DylanRussell · 2026-10-09T15:02:21Z

 BEDROCK_AGENT_RUNTIME = "bedrock-agent-runtime"


+def _extract_request(


i wonder if this method can be in utils instead, and all instrumentations can be updated to use it ? I think we have the same issue in all instrumentations ?

Yes, I think the same shape is elsewhere. In the OpenAI instrumentation, for example, create_chat_invocation runs before the try in patch.py, and malformed message input can raise there when content capture is enabled. I kept this one Bedrock-only on purpose: a shared helper would be new util-genai API, so I'd rather agree on the approach first. Would you prefer an issue for a util-level helper, with this PR staying as the Bedrock fix, or should I fold it in here? Either way, I'm happy to take the follow-up for the other instrumentations.

lets try putting it in utils and updating all instrumentations to use it..

Moved it to util-genai as suppress_extraction_errors(), and bedrock, openai, anthropic, portkey, google-genai and smolagents use it now. The description lists the remaining gaps.

Assisted-by: Claude Opus 5.5

… google-genai Assisted-by: Claude Opus 5.5

Assisted-by: Claude Opus 5.5

lmolkova

Please fix the extraction paths that can throw. Do not add blanket exception suppression. Extractors should handle malformed inputs defensively at the point of extraction. Instrumentation code is written in defensive way and in most cases cannot throw. Adding blanket suppression logic pollutes the code and make it impossible to understand where exceptions should be handled (in the code that calls something that can throw)

lmolkova · 2026-10-11T04:23:38Z

+    with suppress_extraction_errors():
+        raw_agent_id = api_params.get("agentId")
+        if raw_agent_id:
+            invocation.agent_id = str(raw_agent_id)


We should not guard against arbitrary objects whose str raises where the SDK expects a string. This is outside the expected input contract, and the same kind of failure could occur anywhere, including in the exception handler or exporting pipeline

sangkyoonnam · 2026-10-11T05:10:27Z

Thanks for the review and the direction. I'll drop suppress_extraction_errors() and fix the extraction paths that actually throw at the point of extraction, and I'll remove the __str__ guards along with their tests.

Assisted-by: Claude Opus 5.5

sangkyoonnam · 2026-10-12T07:40:53Z

@lmolkova Thanks, done. I dropped suppress_extraction_errors() entirely (c126a75) and fixed each extraction path that could raise at the line that raised (e68114b, e91eed5): type checks, safe field access, and narrowly guarded numeric conversion. The __str__ guards and their tests are gone too. The description lists each case. Ready for another look.

lmolkova · 2026-10-15T22:28:09Z

+def _safe_float(value: Any) -> float | None:
+    try:
+        return float(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def _safe_int(value: Any) -> int | None:
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return None


These still raise: float(10**400) and int(float("inf")) throw OverflowError, and float(obj) runs arbitrary __float__. Could we check types up front instead of catching, so the code can't throw (same idea as _coerce_float in google-genai interactions.py)?

Suggested change

def _safe_float(value: Any) -> float | None:

try:

return float(value)

except (TypeError, ValueError):

return None

def _safe_int(value: Any) -> int | None:

try:

return int(value)

except (TypeError, ValueError):

return None

def _as_float(value: Any) -> float | None:

if isinstance(value, float):

return value

if (

isinstance(value, int)

and not isinstance(value, bool)

and abs(value) <= sys.float_info.max

):

return float(value)

return None

def _as_int(value: Any) -> int | None:

if isinstance(value, int) and not isinstance(value, bool):

return value

return None

(needs import sys; a test case with temperature=10**400 would cover it.)

Applied your suggestion in 9e48fd9 and added the 10**400 and inf cases; numeric strings are now dropped by the type check.

…ng conversion errors

sangkyoonnam · 2026-10-16T07:32:40Z

The instrumentation-genai-qwen-agent-conformance 3.14 failure is a uv download error ("I/O operation failed during extraction"), unrelated to this PR. Could someone re-run it?

Resolve the single content conflict in tests/test_responses.py by keeping both sides: the Responses.parse() coverage added on this branch and the malformed-input test from open-telemetry#866 on main. Production files auto-merged; the parse() wiring in __init__.py and response_extractors.py is intact. Assisted-by: Doubao

fix(bedrock): keep request extraction errors from failing the call

cf0423c

sangkyoonnam requested a review from a team as a code owner October 9, 2026 07:07

Copilot AI balanced review requested due to automatic review settings October 9, 2026 07:08

github-project-automation Bot added this to GenAI Semantic Conventions and Instrumentation libraries Oct 9, 2026

chore(bedrock): add changelog fragment for open-telemetry#866

bdc6ef3

Copilot started reviewing on behalf of sangkyoonnam October 9, 2026 07:08

View session

opentelemetry-pr-dashboard Bot mentioned this pull request Oct 9, 2026

Pull Request Dashboard #171

Open

Copilot AI reviewed Oct 9, 2026

View reviewed changes

DylanRussell reviewed Oct 9, 2026

View reviewed changes

sangkyoonnam added 3 commits October 10, 2026 11:49

feat(util-genai): add suppress_extraction_errors and use it in bedrock

c962074

Assisted-by: Claude Opus 5.5

fix: use suppress_extraction_errors in openai, anthropic, portkey and…

4d2f473

… google-genai Assisted-by: Claude Opus 5.5

chore: add changelog fragments for open-telemetry#866

8c83e45

sangkyoonnam changed the title ~~fix(bedrock): keep request extraction errors from failing the call~~ fix(util-genai): keep request extraction errors from failing instrumented calls Oct 10, 2026

fix(smolagents): keep request extraction errors from failing the call

8be9860

Assisted-by: Claude Opus 5.5

lmolkova requested changes Oct 11, 2026

View reviewed changes

lmolkova reviewed Oct 11, 2026

View reviewed changes

sangkyoonnam added 5 commits October 11, 2026 14:11

chore: merge main

2bc35ca

refactor: drop suppress_extraction_errors

c126a75

Assisted-by: Claude Opus 5.5

fix: handle malformed request inputs in extractors

e68114b

Assisted-by: Claude Opus 5.5

chore: add changelog fragments for open-telemetry#866

5ecc21f

fix: tighten interactions fallback, empty contents and numeric parsing

e91eed5

Assisted-by: Claude Opus 5.5

sangkyoonnam changed the title ~~fix(util-genai): keep request extraction errors from failing instrumented calls~~ fix: handle malformed request inputs in GenAI extractors Oct 11, 2026

chore: merge main

3387cd7

sangkyoonnam requested a review from lmolkova October 12, 2026 07:40

DylanRussell approved these changes Oct 12, 2026

View reviewed changes

sangkyoonnam added 2 commits October 14, 2026 01:57

chore: merge main

070608f

chore: narrow the anthropic changelog fragment after open-telemetry#873

e9b51b3

lmolkova approved these changes Oct 16, 2026

View reviewed changes

fix(portkey): check numeric request parameter types instead of catchi…

9e48fd9

…ng conversion errors

Merge branch 'main' into fix/bedrock-request-extraction-errors

2a449ce

lmolkova added this pull request to the merge queue Oct 16, 2026

Merged via the queue into open-telemetry:main with commit ab95823 Oct 16, 2026
82 checks passed

github-project-automation Bot moved this to Done in GenAI Semantic Conventions and Instrumentation libraries Oct 16, 2026

		BEDROCK_AGENT_RUNTIME = "bedrock-agent-runtime"


		def _extract_request(

Conversation

sangkyoonnam commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Type of change

How has this been tested?

Checklist

Uh oh!

opentelemetry-pr-dashboard Bot commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pull request dashboard status

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

DylanRussell Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

sangkyoonnam Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

DylanRussell Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

sangkyoonnam Oct 10, 2026

Choose a reason for hiding this comment

Uh oh!

lmolkova left a comment

Choose a reason for hiding this comment

Uh oh!

lmolkova Oct 11, 2026

Choose a reason for hiding this comment

Uh oh!

sangkyoonnam commented Oct 11, 2026

Uh oh!

sangkyoonnam commented Oct 12, 2026

Uh oh!

lmolkova Oct 15, 2026

Choose a reason for hiding this comment

Uh oh!

sangkyoonnam Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

sangkyoonnam commented Oct 16, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

sangkyoonnam commented Oct 9, 2026 •

edited

Loading

opentelemetry-pr-dashboard Bot commented Oct 9, 2026 •

edited

Loading