[opentelemetry-instrumentation-genai-bedrock] Add instrumentation for InvokeModel and InvokeModelWithResponseStream

[DylanRussell]

Description

Adds instrumentation for InvokeModel and InvokeModelWithResponseStream

Add support for bedrock specific sem convs: Add aws.bedrock.guardrail.id, add gen_ai.prompt.variable.<key>, add gen_ai.output.type.

• New feature (non-breaking change which adds functionality)

How has this been tested?

Unit tests

Checklist

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

[Copilot]

Pull request overview

Adds Amazon Bedrock Runtime (boto3/botocore) operation-level GenAI instrumentation to the existing opentelemetry-instrumentation-genai-bedrock package, including span/attribute extraction for InvokeModel and InvokeModelWithResponseStream plus corresponding tests and conformance wiring.

Changes:

• Patch botocore.client.BaseClient._make_api_call to instrument Bedrock Converse* and InvokeModel* operations, including stream wrappers for streaming responses.
• Add Bedrock-specific request/response/stream-chunk extractors mapping provider payloads into opentelemetry-util-genai invocation types.
• Add unit + VCR-backed tests and conformance scenarios; wire a Bedrock conformance tox env.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tox.ini	Adds Bedrock conformance env and adjusts Bedrock test commands; introduces global passenv patterns.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_invoke_model.py	New unit tests for invoke_model request/response extraction and span attributes.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_invoke_model_stream.py	New unit tests for streaming InvokeModelWithResponseStream wrapper behavior and error paths.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_converse.py	New VCR-backed tests for converse instrumentation, content capture, tool calls, and errors.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_converse_stream.py	New VCR-backed tests for converse_stream instrumentation, early close, and error paths.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_conformance.py	Adds conformance test runner that executes Bedrock scenarios with weaver live-check.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/test_completion_hook.py	Adds tests validating completion-hook wiring behavior for Bedrock.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/requirements.oldest.txt	Updates oldest test requirements (editable util-genai note).
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/requirements.latest.txt	Keeps latest test requirements aligned with editable workspace installs.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/conftest.py	Adds boto3 Bedrock client fixture, VCR config/scrubbing, env setup, and content-capture fixtures.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/conformance/tool_calling.py	Adds tool-calling conformance scenario for Bedrock chat.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/conformance/inference.py	Adds non-streaming inference conformance scenario for Bedrock chat.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/conformance/inference_streaming.py	Adds streaming inference conformance scenario and validates gen_ai.request.stream=true.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/conformance/init.py	Marks conformance scenario package.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_with_invalid_model.yaml	VCR cassette for Converse error case.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_with_content.yaml	VCR cassette for Converse success with content capture.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_tool_call_with_content.yaml	VCR cassette for Converse tool call flow (content).
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_tool_call_no_content.yaml	VCR cassette for Converse tool call flow (no content).
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_with_invalid_model.yaml	VCR cassette for ConverseStream error case.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_with_content.yaml	VCR cassette for ConverseStream success with content capture.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_with_content_tool_call.yaml	VCR cassette for ConverseStream tool call flow (content).
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_no_content.yaml	VCR cassette for ConverseStream success (no content).
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_handles_event_stream_error.yaml	VCR cassette used for stream error-path test coverage.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_close_before_consumption.yaml	VCR cassette used to validate early stream close handling.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_stream_caller_side_error.yaml	VCR cassette used for caller-side exception path coverage.
instrumentation/opentelemetry-instrumentation-genai-bedrock/tests/cassettes/test_converse_no_content.yaml	VCR cassette for Converse success (no content).
instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/stream.py	Adds SyncStreamWrapper subclasses for ConverseStream and InvokeModelWithResponseStream.
instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/patch.py	Adds BaseClient._make_api_call wrapper and handlers for Converse/InvokeModel paths.
instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/extractors.py	Adds request/response extraction + finish-reason mapping for Bedrock payload formats.
instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/init.py	Wires patch/unpatch into BedrockInstrumentor.
instrumentation/opentelemetry-instrumentation-genai-bedrock/README.rst	Documents supported Bedrock operations and basic usage.
instrumentation/opentelemetry-instrumentation-genai-bedrock/pyproject.toml	Updates util-genai dependency floor for Bedrock instrumentation.
instrumentation/opentelemetry-instrumentation-genai-bedrock/examples/zero-code/main.py	Updates zero-code example to use converse.
instrumentation/opentelemetry-instrumentation-genai-bedrock/examples/manual/main.py	Updates manual example to use converse.
instrumentation/opentelemetry-instrumentation-genai-bedrock/examples/manual/custom_hook.py	Adds custom completion-hook example for Bedrock.
instrumentation/opentelemetry-instrumentation-genai-bedrock/.changelog/488.added	Adds a changelog fragment describing the feature addition.

Suppressed comments (2)

instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/extractors.py:574

• int(...) conversion for output_tokens from the response payload is unguarded and can raise, which would cause the instrumentation wrapper to fail the SDK call.

        if invocation.output_tokens is None:
            out_tok = usage.get("output_tokens") or usage.get("outputTokens")
            if out_tok is not None:
                invocation.output_tokens = int(out_tok)

instrumentation/opentelemetry-instrumentation-genai-bedrock/src/opentelemetry/instrumentation/genai/bedrock/extractors.py:584

• cacheReadInputTokens/cacheWriteInputTokens payload parsing uses unguarded int(...) conversions; if these values are absent or non-integer the instrumentation will raise and fail the request.

        cache_read = usage.get("cache_read_input_tokens") or usage.get(
            "cacheReadInputTokens"
        )
        if cache_read is not None:
            invocation.cache_read_input_tokens = int(cache_read)
        cache_write = usage.get("cache_creation_input_tokens") or usage.get(
            "cacheWriteInputTokens"
        )
        if cache_write is not None:
            invocation.cache_creation_input_tokens = int(cache_write)

---

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

[opentelemetry-pr-dashboard]

Pull request dashboard status

Waiting on the author · refreshed 2026-09-02 21:15 UTC

Respond to 2 review items (e.g. link a commit, explain why not, ask a follow-up):

• Inline threads: 1, 2

Status above doesn't look right?

• Just replied or pushed? Anything around or after the refresh time above may not be picked up yet — give it a few minutes.
• Should this be with reviewers? Comment /dashboard route:reviewers to route it to them.
• Anything wrong — including the routing? Report it with what you expected; it helps us improve the dashboard.

[lmolkova]

This change breaks OpenAI instrumentation when choices have tool calls (OutputMessage expects FinishReason, but OpenAI returns "tool_calls"). Please update OpenAI to map "tool_calls" to "tool_call".

[lmolkova]

self._self_accumulated_reasoning is never populated in _process_chunk, did you want to populate reasoning part with it?