fix(openai): record service_tier attribute #3458
Conversation
|
|
WalkthroughAdds propagation of OpenAI service_tier by importing OpenAIAttributes, adding Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant Client
participant Instrumentation
participant TracedData
participant SpanExporter
Client->>Instrumentation: call API (kwargs include service_tier)
Instrumentation->>TracedData: construct traced_data (set request_service_tier from kwargs)
Instrumentation->>Client: send HTTP request
Client-->>Instrumentation: HTTP response (parsed, may include service_tier)
Instrumentation->>TracedData: set response_service_tier from parsed response
Instrumentation->>SpanExporter: set_data_attributes(traced_data) — write OPENAI_REQUEST/RESPONSE_SERVICE_TIER
SpanExporter->>Instrumentation: span finished
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes
Possibly related PRs
Suggested reviewers
Poem
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Tip 📝 Customizable high-level summaries are now available in beta!You can now customize how CodeRabbit generates the high-level summary in your pull requests — including its content, structure, tone, and formatting.
Example instruction:
Note: This feature is currently in beta for Pro-tier users, and pricing will be announced later. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (1)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (1)
136-138: Consider the Literal type constraint for future service tier values.The
Literal["auto", "default", "flex", "scale", "priority"]constraint provides type safety but may cause issues if OpenAI introduces new service_tier values in the future. Consider whether a broader type likeOptional[str]might be more maintainable, or add a comment noting that this list should be updated when OpenAI adds new tiers.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
Disabled knowledge base sources:
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (4)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py(3 hunks)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py(8 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (1)
**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/*.py: Store API keys only in environment variables/secure vaults; never hardcode secrets in code
Use Flake8 for code linting and adhere to its rules
Files:
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_chat.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_responses.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py
🧠 Learnings (1)
📚 Learning: 2025-08-17T15:06:48.109Z
Learnt from: CR
Repo: traceloop/openllmetry PR: 0
File: CLAUDE.md:0-0
Timestamp: 2025-08-17T15:06:48.109Z
Learning: Semantic conventions must follow the OpenTelemetry GenAI specification
Applied to files:
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py
🧬 Code graph analysis (4)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (1)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (1)
_set_span_attribute(31-38)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (2)
packages/opentelemetry-instrumentation-openai/tests/conftest.py (2)
instrument_legacy(134-149)openai_client(41-42)packages/traceloop-sdk/traceloop/sdk/utils/in_memory_span_exporter.py (1)
get_finished_spans(40-43)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py (1)
packages/traceloop-sdk/traceloop/sdk/utils/in_memory_span_exporter.py (2)
InMemorySpanExporter(22-61)get_finished_spans(40-43)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (1)
packages/opentelemetry-instrumentation-alephalpha/opentelemetry/instrumentation/alephalpha/__init__.py (1)
_set_span_attribute(63-67)
🪛 Ruff (0.14.5)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py
1501-1501: Unused function argument: instrument_legacy
(ARG001)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py
34-34: Unused function argument: instrument_legacy
(ARG001)
🔇 Additional comments (9)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py (1)
32-47: LGTM! Test coverage for service_tier is appropriate.The test validates that both
openai.request.service_tierandopenai.response.service_tierattributes are correctly propagated through the instrumentation for the responses API.Note: The static analysis hint about
instrument_legacybeing unused is a false positive—it's a pytest fixture that enables instrumentation for the test.packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (1)
1500-1519: LGTM! Service tier test for chat completions is correct.The test properly validates that service_tier propagation works for chat completions, mirroring the coverage provided in test_responses.py for the responses API.
Note: The static analysis hint about
instrument_legacyis a false positive—it's a pytest fixture required for instrumentation.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (3)
145-147: Request service_tier attribute setting is correct.The implementation properly captures the
service_tierparameter from kwargs and sets it as a span attribute using the shared_set_span_attributehelper, which handles None, empty, and NOT_GIVEN values appropriately.
217-221: Response service_tier attribute setting is correct.The implementation properly extracts
service_tierfrom the response and sets it as a span attribute. The formatting is consistent with other response attributes in this function.
15-15: Verify OpenAIAttributes constants are accessible at runtime.The OpenTelemetry semantic conventions define OpenAI attributes including openai.request.service_tier and openai.response.service_tier. The dependency
opentelemetry-semantic-conventions >= 0.59b0is properly declared inpyproject.toml, and the code successfully importsOpenAIAttributesalongsideGenAIAttributesfrom the same source package.However, the exact Python module structure for
opentelemetry.semconv._incubating.attributes.openai_attributesand the presence of constantsOPENAI_REQUEST_SERVICE_TIERandOPENAI_RESPONSE_SERVICE_TIERcould not be confirmed from public documentation. While the constants are used in multiple locations (lines 146, 219 inshared/__init__.pyand lines 197-198 inv1/responses_wrappers.py), you should verify that these constants are properly exported and accessible at runtime, particularly given they are part of the incubating (_incubating) module which may still be evolving.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (4)
197-198: Service tier span attributes are correctly set.The implementation properly propagates both request and response service_tier values to span attributes using the appropriate OpenAIAttributes constants and the
_set_span_attributehelper.
493-494: Sync path service_tier propagation is correct.The implementation consistently handles service_tier in both error and normal execution paths:
- Error path (lines 493-494): Captures request service_tier from kwargs and preserves response service_tier from existing_data
- Normal path (lines 558-559): Prioritizes existing request service_tier but falls back to kwargs, and captures response service_tier from the parsed_response
Also applies to: 558-559
635-636: Async path service_tier propagation mirrors sync path correctly.The async implementation maintains the same logic as the sync path for service_tier handling, ensuring consistency across both execution modes.
Also applies to: 701-702
809-810: ResponseStream initialization handles service_tier appropriately.The streaming initialization correctly captures the request service_tier from request_kwargs and initializes response_service_tier to None, which will be populated when the complete response is received.
ataha322
force-pushed
the
main
branch
2 times, most recently
from
e23bdce to
b980863
Compare
|
Hey @nirga, done |
* `openai.request.service_tier` is captured * `openai.response.service_tier` is captured * Supported APIs: Responses, Chat * Corresponding tests are added
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (3)
packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yaml (1)
1-110: Cassette captures service_tier correctly; consider scrubbing cookies/IDs if neededThe cassette correctly records a
responsescall withservice_tier: "priority"and no API keys or auth headers. If your cassette policy treats Cloudflare cookies and OpenAI org/project identifiers as sensitive, consider scrubbing those header values in line with your other fixtures.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (2)
40-51: Remove unusedLiteralimport
Literalis imported but never used, which Flake8 flags. It’s safe to drop it from the import list.-from opentelemetry.semconv_ai import SpanAttributes -from opentelemetry.semconv.attributes.error_attributes import ERROR_TYPE -from opentelemetry.trace import SpanKind, Span, StatusCode, Tracer -from typing import Any, Optional, Union, Literal +from opentelemetry.semconv_ai import SpanAttributes +from opentelemetry.semconv.attributes.error_attributes import ERROR_TYPE +from opentelemetry.trace import SpanKind, Span, StatusCode, Tracer +from typing import Any, Optional, Union
136-139: Streaming path doesn’t populateopenai.response.service_tieryetThe new
request_service_tier/response_service_tierfields onTracedDataare wired through the sync and async wrappers and ultimately emitted inset_data_attributes, so non‑streaming calls correctly get bothopenai.request.service_tierandopenai.response.service_tier.For streaming (
ResponseStream), you capturerequest_service_tierfromrequest_kwargs, but_process_complete_responsenever setsself._traced_data.response_service_tierfrom the finalparsed_response. As a result,openai.response.service_tierwill remain unset on spans produced via streaming, even when the response includes a service tier.Consider setting
response_service_tierwhen you have the complete response:@dont_throw def _process_complete_response(self): """Process the complete response and emit span""" with self._cleanup_lock: if self._cleanup_completed: return try: if self._complete_response_data: parsed_response = parse_response(self._complete_response_data) self._traced_data.response_id = parsed_response.id self._traced_data.response_model = parsed_response.model self._traced_data.output_text = self._output_text if parsed_response.usage: self._traced_data.usage = parsed_response.usage if parsed_response.output: self._traced_data.output_blocks = { block.id: block for block in parsed_response.output } + + # Capture service tier from the final response, if available + service_tier = getattr(parsed_response, "service_tier", None) + if service_tier is None and isinstance(parsed_response, dict): + service_tier = parsed_response.get("service_tier") + self._traced_data.response_service_tier = service_tier responses[parsed_response.id] = self._traced_data set_data_attributes(self._traced_data, self._span)This keeps streaming behavior consistent with non‑streaming calls for the new
openai.response.service_tierattribute while remaining defensive for both object and dict response shapes.Also applies to: 191-199, 493-495, 558-560, 635-637, 701-703, 793-811
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
Disabled knowledge base sources:
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (6)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py(3 hunks)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py(8 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_chat/test_chat_with_service_tier.yaml(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yaml(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py(1 hunks)
✅ Files skipped from review due to trivial changes (1)
- packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_chat/test_chat_with_service_tier.yaml
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/*.py: Store API keys only in environment variables/secure vaults; never hardcode secrets in code
Use Flake8 for code linting and adhere to its rules
Files:
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py
**/cassettes/**/*.{yaml,yml,json}
📄 CodeRabbit inference engine (CLAUDE.md)
Never commit secrets or PII in VCR cassettes; scrub sensitive data
Files:
packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yaml
🧠 Learnings (1)
📚 Learning: 2025-08-17T15:06:48.109Z
Learnt from: CR
Repo: traceloop/openllmetry PR: 0
File: CLAUDE.md:0-0
Timestamp: 2025-08-17T15:06:48.109Z
Learning: Semantic conventions must follow the OpenTelemetry GenAI specification
Applied to files:
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py
🧬 Code graph analysis (2)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (8)
packages/opentelemetry-instrumentation-alephalpha/opentelemetry/instrumentation/alephalpha/__init__.py (1)
_set_span_attribute(63-67)packages/opentelemetry-instrumentation-mistralai/opentelemetry/instrumentation/mistralai/__init__.py (1)
_set_span_attribute(74-78)packages/opentelemetry-instrumentation-watsonx/opentelemetry/instrumentation/watsonx/__init__.py (1)
_set_span_attribute(105-109)packages/opentelemetry-instrumentation-llamaindex/opentelemetry/instrumentation/llamaindex/custom_llm_instrumentor.py (1)
_set_span_attribute(71-75)packages/opentelemetry-instrumentation-milvus/opentelemetry/instrumentation/milvus/wrapper.py (1)
_set_span_attribute(53-57)packages/opentelemetry-instrumentation-chromadb/opentelemetry/instrumentation/chromadb/wrapper.py (1)
_set_span_attribute(26-30)packages/opentelemetry-instrumentation-qdrant/opentelemetry/instrumentation/qdrant/wrapper.py (1)
_set_span_attribute(11-15)packages/opentelemetry-instrumentation-weaviate/opentelemetry/instrumentation/weaviate/wrapper.py (1)
_set_span_attribute(26-30)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (1)
packages/opentelemetry-instrumentation-openai/tests/conftest.py (2)
instrument_legacy(134-149)openai_client(41-42)
🪛 Flake8 (7.3.0)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py
[error] 50-50: 'typing.Literal' imported but unused
(F401)
🪛 Ruff (0.14.5)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py
34-34: Unused function argument: instrument_legacy
(ARG001)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py
1501-1501: Unused function argument: instrument_legacy
(ARG001)
🔇 Additional comments (3)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py (1)
32-46: Nice focused coverage for Responses service_tier propagationThe test cleanly exercises the Responses path with
service_tier="priority"and verifies both request/response attributes on the span; this aligns with the new semantics and mirrors existing test patterns in this file.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (1)
15-16: Service tier attributes are wired in cleanly via OpenAIAttributesImporting
openai_attributes as OpenAIAttributesand setting request/response service_tier in_set_request_attributesand_set_response_attributesvia_set_span_attributeis consistent with the existing pattern and keeps the attribute keys aligned with the OpenTelemetry semconv definitions. Based on learnings, this preserves GenAI semantic‑convention compliance.Also applies to: 145-147, 217-221
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (1)
1500-1519: Good addition to cover Chat service_tier propagationThis test follows the existing chat patterns (using
spans[-1]) and validates both request/response service_tier attributes, giving direct coverage of the new behavior for the Chat API.
There was a problem hiding this comment.
Actionable comments posted: 1
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
Disabled knowledge base sources:
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (6)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py(3 hunks)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py(8 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_chat/test_chat_with_service_tier.yaml(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yaml(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py(1 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py(1 hunks)
🚧 Files skipped from review as they are similar to previous changes (1)
- packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/init.py
🧰 Additional context used
📓 Path-based instructions (2)
**/cassettes/**/*.{yaml,yml,json}
📄 CodeRabbit inference engine (CLAUDE.md)
Never commit secrets or PII in VCR cassettes; scrub sensitive data
Files:
packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yamlpackages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_chat/test_chat_with_service_tier.yaml
**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/*.py: Store API keys only in environment variables/secure vaults; never hardcode secrets in code
Use Flake8 for code linting and adhere to its rules
Files:
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_chat.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py
🧬 Code graph analysis (3)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (1)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py (1)
_set_span_attribute(31-38)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (1)
packages/opentelemetry-instrumentation-openai/tests/conftest.py (2)
instrument_legacy(134-149)openai_client(41-42)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py (1)
packages/traceloop-sdk/traceloop/sdk/utils/in_memory_span_exporter.py (2)
InMemorySpanExporter(22-61)get_finished_spans(40-43)
🪛 Ruff (0.14.5)
packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py
1501-1501: Unused function argument: instrument_legacy
(ARG001)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py
34-34: Unused function argument: instrument_legacy
(ARG001)
🔇 Additional comments (9)
packages/opentelemetry-instrumentation-openai/tests/traces/test_responses.py (1)
32-47: LGTM! Service tier test correctly validates attribute propagation.The test properly exercises the service_tier parameter and verifies both request and response attributes are captured in the span.
Note: The static analysis warning about unused
instrument_legacyis a false positive - this is a pytest fixture used for instrumentation setup.packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_responses/test_responses_with_service_tier.yaml (1)
1-110: LGTM! Cassette properly records service tier interaction.The cassette correctly captures the request with
service_tier: "priority"and the corresponding response. No secrets or PII detected in the recording.As per coding guidelines, cassette data appears properly scrubbed.
packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/test_chat/test_chat_with_service_tier.yaml (1)
1-113: LGTM! Chat cassette properly captures service tier.The cassette correctly records the chat completions request with
service_tier: "priority". No secrets or PII detected.packages/opentelemetry-instrumentation-openai/tests/traces/test_chat.py (1)
1500-1519: LGTM! Chat service tier test provides good coverage.The test properly validates service_tier propagation for chat completions, complementing the responses test.
Note: The static analysis warning about unused
instrument_legacyis a false positive - this is a pytest fixture.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py (5)
45-45: LGTM! Import correctly adds OpenAI attributes.The import follows the existing pattern and enables access to the new service tier attribute constants.
136-138: LGTM! TracedData fields properly defined.The new service tier fields follow the established pattern for request/response attributes and use appropriate types.
197-198: LGTM! Span attributes correctly written.The service tier attributes are properly written to the span using the OpenAI attribute constants. The
_set_span_attributehelper handles None values appropriately.
493-494: LGTM! Service tier propagation handles both success and error paths.The implementation correctly:
- Captures request service tier from kwargs in error scenarios
- Uses fallback logic in success path (existing_data → current request/response)
- Handles multi-turn responses where existing_data may already contain service tier
Also applies to: 558-559
635-636: LGTM! Async wrapper maintains consistency with sync implementation.The async wrapper correctly implements the same service tier propagation logic as the synchronous version, maintaining consistency across both code paths.
Also applies to: 701-702
...lemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.py
Show resolved
Hide resolved
nirga
changed the title
| span, SpanAttributes.LLM_IS_STREAMING, kwargs.get("stream") or False | ||
| ) | ||
| _set_span_attribute( | ||
| span, OpenAIAttributes.OPENAI_REQUEST_SERVICE_TIER, kwargs.get("service_tier") |
There was a problem hiding this comment.
@nirga Are you sure? It's already guarded in the setter:
def _set_span_attribute(span, name, value):
if value is None or value == "":
returnThere was a problem hiding this comment.
You’re right! We don’t have it in all instrumentations unfortunately so I forgot about that here. Thanks!
|
tysm for quick cooperation! |
|
@nirga could you please make it a release |
3 participants
openai.request.service_tieris capturedopenai.response.service_tieris capturedfeat(instrumentation): ...orfix(instrumentation): ....Summary by CodeRabbit
New Features
Tests
✏️ Tip: You can customize this high-level summary in your review settings.