feat(instrumentaiton): OpenAI audio transcription #3514
Conversation
* instruments OpenAI().audio.transcriptions.create() * captures audio duration * captures model name and other standard request/response parameters * capturing audio duration depends on `mutagen`, included as optional dependency. * user should import instrumentation with flag `["audio"]` if he wishes to capture audio duration. This way mutagen will be included. * Still WIP, need to finish tests and consult with maintainers on semconv and overall review.
WalkthroughThe changes introduce OpenTelemetry instrumentation for OpenAI audio transcription calls. A new audio_wrappers module provides synchronous and asynchronous wrapper functions for transcription methods, capturing span attributes, duration metrics, and error details. Supporting utilities and integration hooks are added to the instrumentation layer, with optional mutagen dependency for extracting audio duration metadata. Changes
Sequence DiagramsequenceDiagram
participant Client
participant Wrapper as Wrapper Layer
participant Tracer
participant Span
participant Histogram as Metrics<br/>(Histogram)
participant Counter as Metrics<br/>(Counter)
participant OpenAI as OpenAI API
Client->>Wrapper: call transcription_wrapper()
activate Wrapper
Note over Wrapper: Start timing
Wrapper->>Tracer: start_as_current_span("openai.audio.transcriptions")
activate Tracer
Tracer->>Span: create span
activate Span
Wrapper->>Span: set request attributes<br/>(model, duration, api base)
Wrapper->>OpenAI: execute transcription call
activate OpenAI
OpenAI-->>Wrapper: response
deactivate OpenAI
Note over Wrapper: End timing
Wrapper->>Span: set response attributes
Wrapper->>Histogram: record duration
activate Histogram
Histogram-->>Wrapper: metric recorded
deactivate Histogram
Wrapper->>Span: mark success
Span-->>Wrapper: span ends
deactivate Span
deactivate Tracer
Wrapper-->>Client: return response
deactivate Wrapper
rect rgb(200, 150, 255)
Note over Wrapper,Counter: Error Path (on exception)
Wrapper->>Counter: increment exception_counter
activate Counter
Counter-->>Wrapper: counted
deactivate Counter
Wrapper->>Histogram: record duration
Wrapper->>Span: set error status + exception info
Wrapper-->>Client: re-raise exception
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Areas requiring attention:
Poem
Pre-merge checks and finishing touches❌ Failed checks (1 warning, 1 inconclusive)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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 |
|
Hi @nirga It's still a draft, but I need some consultancy on the questions that I described in that PR description. |
|
Hey @ataha322 - can you mark the PR as ready to review and did the PR issues? |
![ellipsis-dev[bot]](https://avatars.githubusercontent.com/in/64358?s=60&v=4){kind=small}
There was a problem hiding this comment.
Caution
Changes requested ❌
Reviewed everything up to b6db65e in 2 minutes and 27 seconds. Click for details.
- Reviewed
434lines of code in5files - Skipped
0files when reviewing. - Skipped posting
4draft comments. View those below. - Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.
1. packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py:195
- Draft comment:
Remove or update the commented-out span attribute code once the semconv is published. - Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 10% vs. threshold = 50% The code already has a TODO comment on line 198 that says "TODO(Ata): come back here later when semconv is published". The automated tool's comment essentially says the same thing: "Remove or update the commented-out span attribute code once the semconv is published." This is redundant with the existing TODO. According to the rules, I should not keep comments that are obvious or unimportant, and this comment doesn't add any new information beyond what the developer already documented. The developer is clearly aware of this issue and has already marked it for future work. Perhaps the automated comment is more visible than an inline TODO and could serve as a reminder in the PR review process. Also, having it as a PR comment might ensure it gets tracked in a review system rather than just being buried in code. While PR comments can be more visible, the developer has already explicitly documented this with a TODO comment that includes their name and a clear explanation. The PR review process should catch obvious TODOs in new code if they're a concern. Adding a redundant comment doesn't provide additional value and violates the rule about not making obvious comments. The comment should be deleted because it's redundant with the existing TODO comment in the code (line 198). The developer has already documented this issue clearly, and the automated comment doesn't add any new information or actionable guidance.
2. packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py:99
- Draft comment:
There's a typographical error in the comment on this line. It reads "if there are legal duration, record it". Consider revising it to something like "if there is a valid duration, record it". - Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 20% vs. threshold = 50% This is a comment about fixing a typo/grammatical error in a code comment. While it's technically correct that "legal duration" is awkward phrasing and "valid duration" would be better, this falls under the category of "obvious or unimportant" comments. The meaning is clear enough despite the grammatical issue. The rules specifically state "Do NOT make comments that are obvious or unimportant." A minor grammatical improvement to a comment is not a critical code change. Additionally, the word "legal" might be intentional (though awkward) to mean "valid/acceptable". This is a very minor style/grammar issue that doesn't affect functionality. The comment does point out a legitimate grammatical issue that would improve code readability. "Legal duration" is indeed awkward English, and "valid duration" is clearer. Code comments should be clear and well-written. This is an actionable suggestion that improves code quality. While the suggestion does improve clarity, the rules explicitly state not to make comments that are "obvious or unimportant." A minor grammatical improvement to a comment (not even to code) falls into this category. The current comment is understandable despite the awkward phrasing, and this is not a critical issue that requires attention. This comment should be deleted. It's pointing out a minor grammatical issue in a code comment that doesn't affect functionality or significantly impact readability. This falls under "obvious or unimportant" comments that should not be made according to the rules.
3. packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.py:22
- Draft comment:
It appears that the nameatranscription_wrappermight be a typographical error. If this is intended to handle audio transcription, consider renaming it to something likeaudio_transcription_wrapperfor clarity. - Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 0% vs. threshold = 50% The comment is clearly incorrect. The codebase has a consistent pattern of prefixing async wrappers with 'a' -achat_wrapper,acompletion_wrapper,aembeddings_wrapper, and nowatranscription_wrapper. This is not a typo but a deliberate naming convention. The comment shows a lack of understanding of the codebase's naming patterns. This is exactly the type of comment that should be deleted - it's suggesting a change that would actually break the consistency of the codebase. Could there be a reason why this particular wrapper should deviate from the established pattern? Maybe the 'a' prefix is confusing in this specific context? No, there's no reason to deviate from the pattern. The 'a' prefix for async is used consistently throughout the file for all async wrappers. Changing this one would make it inconsistent with the rest of the codebase. The comment is objectively wrong. This comment should be deleted. It incorrectly identifies a deliberate naming convention as a typo. The 'a' prefix for async wrappers is used consistently throughout the codebase (achat_wrapper, acompletion_wrapper, aembeddings_wrapper), and atranscription_wrapper follows this same pattern.
4. packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.py:277
- Draft comment:
The function name 'atranscription_wrapper' looks like it might be a typographical error. If its purpose is to wrap asynchronous transcription calls, consider renaming it to 'async_transcription_wrapper' for clarity and consistency. - Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 0% vs. threshold = 50% This comment is clearly incorrect. The codebase has an established naming convention where async wrappers are prefixed with 'a' rather than 'async_'. Examples includeachat_wrapper,acompletion_wrapper, andaembeddings_wrapper. The newatranscription_wrapperfollows this exact same pattern and is therefore consistent with the codebase style. The comment suggests changing it toasync_transcription_wrapper, which would actually make it inconsistent with the rest of the codebase. This is a code quality suggestion that fails to recognize the existing conventions. Could there be a reason why the author might want to change the naming convention for this particular wrapper? Perhaps the 'a' prefix is considered unclear and they want to start using more explicit names going forward? Even if there were a desire to change naming conventions, that should be done consistently across the entire codebase, not just for one new function. The comment doesn't suggest refactoring all the other wrappers, so it's asking for inconsistency. This is clearly not understanding the existing pattern. The comment should be deleted because it incorrectly suggests thatatranscription_wrapperis a typo when it actually follows the established naming convention in the codebase (e.g.,achat_wrapper,acompletion_wrapper,aembeddings_wrapper). The suggestion would make the code inconsistent.
Workflow ID: wflow_OSgCpYeNswuAXV6E
You can customize 
| span, 'gen_ai.openai.audio.input.duration_seconds', audio_duration | ||
| ) | ||
| else: | ||
| print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN") |
There was a problem hiding this comment.
Remove the debug print statement and use the logger at an appropriate level instead.
| print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN") | |
| logger.debug("COULD NOT READ AUDIO FILE WITH MUTAGEN") |
| end_time = time.time() | ||
| except Exception as e: # pylint: disable=broad-except | ||
| end_time = time.time() | ||
| duration = end_time - start_time if "start_time" in locals() else 0 |
There was a problem hiding this comment.
Consider initializing start_time before the try block to avoid checking if "start_time" in locals() in the exception handler.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (1)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py (1)
33-64: Consider supporting string/bytes paths in_get_audio_durationRight now, any
filethat is astr/bytespath returnsNonewithout attempting duration extraction:elif isinstance(file, (str, bytes)): # If it's a path string or bytes return NoneIf you expect callers to pass a path (e.g.,
file="path/to/file.mp3"), you could let mutagen handle those as well:- try: - # First check if it's a file-like object with a name attribute - if hasattr(file, "name"): - file_path = file.name - elif isinstance(file, (str, bytes)): - # If it's a path string or bytes - return None - else: - # If it's a file-like object without name, we can't easily determine duration - return None + try: + # Resolve a filesystem path from the input + if hasattr(file, "name"): + file_path = file.name + elif isinstance(file, (str, bytes)): + file_path = file + else: + # If it's a file-like object without name, we can't easily determine duration + return NoneThe existing broad exception handling already ensures failures here don’t impact user code.
If you adopt this, please verify it behaves correctly for both open file objects and plain path strings on your supported platforms.
📜 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 (5)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py(1 hunks)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py(1 hunks)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.py(3 hunks)packages/opentelemetry-instrumentation-openai/pyproject.toml(2 hunks)packages/opentelemetry-instrumentation-openai/tests/traces/test_audio_transcription.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/tests/traces/test_audio_transcription.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py
🧠 Learnings (2)
📚 Learning: 2025-12-02T21:09:48.690Z
Learnt from: duanyutong
Repo: traceloop/openllmetry PR: 3487
File: packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py:177-178
Timestamp: 2025-12-02T21:09:48.690Z
Learning: The opentelemetry-instrumentation-openai and opentelemetry-instrumentation-openai-agents packages must remain independent and not share code, so code duplication between them is acceptable.
Applied to files:
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.pypackages/opentelemetry-instrumentation-openai/pyproject.toml
📚 Learning: 2025-08-22T14:41:26.962Z
Learnt from: prane-eth
Repo: traceloop/openllmetry PR: 3336
File: packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py:8-8
Timestamp: 2025-08-22T14:41:26.962Z
Learning: In the openllmetry project, the `packaging` library is available as a transitive dependency through other packages (visible in poetry.lock) and doesn't need to be explicitly declared in pyproject.toml dependencies.
Applied to files:
packages/opentelemetry-instrumentation-openai/pyproject.toml
🧬 Code graph analysis (3)
packages/opentelemetry-instrumentation-openai/tests/traces/test_audio_transcription.py (2)
packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py (1)
SpanAttributes(64-245)packages/traceloop-sdk/traceloop/sdk/utils/in_memory_span_exporter.py (1)
get_finished_spans(40-43)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.py (2)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py (2)
atranscription_wrapper(126-182)transcription_wrapper(67-122)packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py (1)
is_metrics_enabled(39-40)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py (2)
packages/opentelemetry-instrumentation-anthropic/opentelemetry/instrumentation/anthropic/__init__.py (1)
wrapper(421-434)packages/opentelemetry-instrumentation-llamaindex/opentelemetry/instrumentation/llamaindex/utils.py (2)
wrapper(18-19)wrapper(48-58)
🪛 Flake8 (7.3.0)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py
[error] 22-22: 'opentelemetry.semconv_ai.SpanAttributes' imported but unused
(F401)
🪛 Ruff (0.14.8)
packages/opentelemetry-instrumentation-openai/tests/traces/test_audio_transcription.py
10-10: Unused function argument: instrument_legacy
(ARG001)
37-37: Unused function argument: instrument_legacy
(ARG001)
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py
57-58: try-except-pass detected, consider logging the exception
(S110)
57-57: Do not catch blind exception: Exception
(BLE001)
60-60: Do not catch blind exception: Exception
(BLE001)
🔇 Additional comments (3)
packages/opentelemetry-instrumentation-openai/pyproject.toml (1)
31-31: Optional mutagen dependency andaudioextra wiring look correctDefining
mutagenasoptional = trueand exposing it via theaudioextra is aligned with how Poetry handles optional features.Please confirm
^1.47.0matches the minimum version you want to support for duration extraction across your supported Python/OpenAI versions.Also applies to: 51-54
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py (1)
86-106: Audio telemetry wrapper pattern matches existing utilitiesThe
_with_audio_telemetry_wrapperhelper is consistent with the embeddings/chat wrappers and correctly threadstracer,duration_histogram, andexception_counterinto the inner wrapper; this looks idiomatic for this codebase.packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/__init__.py (1)
21-24: Audio transcription wrapping is wired correctly; metric naming ready for future semconvThe sync/async
Transcriptions.createhooks and correspondingunwrapcalls are symmetrical and follow the same pattern as chat/embeddings/image instrumentation, reusingduration_histogramand an exception counter, which looks correct.For the counter:
audio_transcription_exception_counter = meter.create_counter( # name=Meters.LLM_AUDIO_TRANSCRIPTIONS_EXCEPTIONS, # TODO name="llm.openai.audio.transcriptions.exceptions", ... )using a provisional vendor-specific name is reasonable; just ensure you align it with the eventual
Meters.LLM_AUDIO_TRANSCRIPTIONS_EXCEPTIONS(or equivalent) once the semconv metric is published to avoid a breaking rename later.Please double‑check the intended final metric name in
opentelemetry-semconv-ai(or its design doc) so the provisional string matches or can be aliased cleanly when the constant lands.Also applies to: 254-282, 375-376
| from opentelemetry.semconv_ai import ( | ||
| SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY, | ||
| SpanAttributes, | ||
| ) |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, locate and read the audio_wrappers.py file
find . -type f -name "audio_wrappers.py" | head -20Repository: traceloop/openllmetry
Length of output: 174
🏁 Script executed:
cat -n ./packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.pyRepository: traceloop/openllmetry
Length of output: 8977
Remove unused SpanAttributes import to satisfy Flake8
SpanAttributes is imported but never used in this file—it only appears in commented-out code (line 196), which will trigger a Flake8 F401 violation. Remove it from the import:
from opentelemetry.semconv_ai import (
SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY,
- SpanAttributes,
)Additionally, remove the debug print statement on line 203:
- print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN")Re-introduce SpanAttributes once the audio duration attribute is published in semconv_ai.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| from opentelemetry.semconv_ai import ( | |
| SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY, | |
| SpanAttributes, | |
| ) | |
| from opentelemetry.semconv_ai import ( | |
| SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY, | |
| ) |
🧰 Tools
🪛 Flake8 (7.3.0)
[error] 22-22: 'opentelemetry.semconv_ai.SpanAttributes' imported but unused
(F401)
🤖 Prompt for AI Agents
In
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py
around lines 22–25, remove the unused SpanAttributes import from
opentelemetry.semconv_ai to fix the Flake8 F401 violation and delete the debug
print statement at line 203; keep the rest of the imports intact and re-add
SpanAttributes later if/when the audio duration attribute is published in
semconv_ai.
| @dont_throw | ||
| def _handle_request(span, kwargs, instance): | ||
| _set_request_attributes(span, kwargs, instance) | ||
| _set_client_attributes(span, instance) | ||
|
|
||
| # Extract and set audio duration | ||
| file_param = kwargs.get("file") | ||
| if file_param: | ||
| audio_duration = _get_audio_duration(file_param) | ||
| if audio_duration is not None: | ||
| # _set_span_attribute( | ||
| # span, SpanAttributes.LLM_OPENAI_AUDIO_INPUT_DURATION_SECONDS, audio_duration | ||
| # ) | ||
| # TODO(Ata): come back here later when semconv is published | ||
| _set_span_attribute( | ||
| span, 'gen_ai.openai.audio.input.duration_seconds', audio_duration | ||
| ) | ||
| else: | ||
| print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN") | ||
|
|
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
cat -n packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py | sed -n '185,210p'Repository: traceloop/openllmetry
Length of output: 1110
🏁 Script executed:
# Check if the file exists and explore structure
fd audio_wrappers.pyRepository: traceloop/openllmetry
Length of output: 172
🏁 Script executed:
# Search for SpanAttributes to understand naming conventions
rg "SpanAttributes" packages/opentelemetry-instrumentation-openai --type py -A 3 -B 1Repository: traceloop/openllmetry
Length of output: 50378
Remove debug print statement and refactor audio duration attribute naming
Two issues require attention:
- Debug output on line 203
The print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN") statement will spam stdout when audio duration cannot be read. Remove it entirely or replace with a structured log:
else:
- print("REMOVE ME : ATA-DBG : COULD NOT READ AUDIO FILE WITH MUTAGEN")- Hardcoded attribute key on line 200
Replace the hardcoded string 'gen_ai.openai.audio.input.duration_seconds' with a module-level constant. This will simplify the migration to SpanAttributes.LLM_OPENAI_AUDIO_INPUT_DURATION_SECONDS once it is published:
# At module top
AUDIO_INPUT_DURATION_ATTR = "gen_ai.openai.audio.input.duration_seconds"
# Then use
_set_span_attribute(span, AUDIO_INPUT_DURATION_ATTR, audio_duration)This approach avoids hardcoding magic strings and makes future refactoring to the official semconv attribute straightforward.
🤖 Prompt for AI Agents
In
packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/audio_wrappers.py
around lines 185 to 204, remove the debug print statement that spams stdout when
mutagen cannot read the audio file and instead either drop the message or
replace it with a structured logger call; also introduce a module-level constant
(e.g., AUDIO_INPUT_DURATION_ATTR) at the top of the file for the attribute key
and use that constant in the _set_span_attribute call instead of the hardcoded
string so future migration to
SpanAttributes.LLM_OPENAI_AUDIO_INPUT_DURATION_SECONDS is simple.
| @pytest.mark.vcr | ||
| def test_audio_transcription(instrument_legacy, span_exporter, openai_client): | ||
| # Create a mock audio file (in real test, use VCR cassette) | ||
| audio_file = io.BytesIO(b"fake audio content") | ||
| audio_file.name = "test_audio.mp3" | ||
|
|
||
| openai_client.audio.transcriptions.create( | ||
| file=audio_file, | ||
| model="whisper-1", | ||
| ) | ||
|
|
||
| spans = span_exporter.get_finished_spans() | ||
| assert len(spans) == 1 | ||
| assert spans[0].name == "openai.audio.transcriptions" | ||
|
|
||
| transcription_span = spans[0] | ||
| assert ( | ||
| transcription_span.attributes[GenAIAttributes.GEN_AI_REQUEST_MODEL] | ||
| == "whisper-1" | ||
| ) | ||
| assert ( | ||
| transcription_span.attributes[SpanAttributes.LLM_OPENAI_API_BASE] | ||
| == "https://api.openai.com/v1/" | ||
| ) |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
find . -type f -name "test_audio_transcription.py" | head -20Repository: traceloop/openllmetry
Length of output: 153
🏁 Script executed:
fd "test_audio_transcription.py"Repository: traceloop/openllmetry
Length of output: 151
🏁 Script executed:
git ls-files | grep -i "audio_transcription"Repository: traceloop/openllmetry
Length of output: 151
🏁 Script executed:
cat -n packages/opentelemetry-instrumentation-openai/tests/traces/test_audio_transcription.pyRepository: traceloop/openllmetry
Length of output: 2220
🏁 Script executed:
# Check conftest.py for VCR configuration in the same directory
cat -n packages/opentelemetry-instrumentation-openai/tests/traces/conftest.py 2>/dev/null || echo "conftest.py not found in traces dir"Repository: traceloop/openllmetry
Length of output: 332
🏁 Script executed:
# Check parent conftest.py
cat -n packages/opentelemetry-instrumentation-openai/tests/conftest.py 2>/dev/null || echo "conftest.py not found in tests dir"Repository: traceloop/openllmetry
Length of output: 6770
🏁 Script executed:
fd "cassettes" -type d packages/opentelemetry-instrumentation-openai/tests/Repository: traceloop/openllmetry
Length of output: 235
🏁 Script executed:
find packages/opentelemetry-instrumentation-openai/tests -type d -name "*cassette*" -o -type d -name "*fixtures*"Repository: traceloop/openllmetry
Length of output: 203
🏁 Script executed:
# Look for YAML or JSON cassette files that might be VCR recordings
find packages/opentelemetry-instrumentation-openai/tests -type f \( -name "*.yaml" -o -name "*.yml" -o -name "*.json" \) | head -20Repository: traceloop/openllmetry
Length of output: 2678
🏁 Script executed:
find packages/opentelemetry-instrumentation-openai/tests/traces/cassettes -name "*audio*"Repository: traceloop/openllmetry
Length of output: 47
🏁 Script executed:
ls -la packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/ | grep -i audioRepository: traceloop/openllmetry
Length of output: 47
🏁 Script executed:
# List all cassettes in the traces cassettes directory
ls packages/opentelemetry-instrumentation-openai/tests/traces/cassettes/Repository: traceloop/openllmetry
Length of output: 283
🏁 Script executed:
# Check if there's a test_audio_transcription.py file in traces directory listing
ls packages/opentelemetry-instrumentation-openai/tests/traces/ | grep -i audioRepository: traceloop/openllmetry
Length of output: 92
🏁 Script executed:
# Get complete file listing to see if test_audio_transcription is listed
ls packages/opentelemetry-instrumentation-openai/tests/traces/test_*.py | head -20Repository: traceloop/openllmetry
Length of output: 1361
🏁 Script executed:
# Check if ruff or flake8 is configured and if ARG001 is enabled
find packages/opentelemetry-instrumentation-openai -name "pyproject.toml" -o -name "ruff.toml" -o -name ".flake8" | xargs grep -l "ARG001\|ARG\|flake8\|ruff" 2>/dev/null | head -5Repository: traceloop/openllmetry
Length of output: 179
🏁 Script executed:
# Check root pyproject.toml for linting config
cat pyproject.toml 2>/dev/null | grep -A 20 "\[tool.ruff\]\|\[tool.flake8\]" | head -40Repository: traceloop/openllmetry
Length of output: 47
🏁 Script executed:
cat packages/opentelemetry-instrumentation-openai/.flake8Repository: traceloop/openllmetry
Length of output: 178
🏁 Script executed:
cat packages/opentelemetry-instrumentation-openai/pyproject.toml | head -60Repository: traceloop/openllmetry
Length of output: 1597
VCR cassettes must be recorded before these tests can run; fake audio content will cause API errors
Both test_audio_transcription and test_audio_transcription_async are decorated with @pytest.mark.vcr but have no cassette files recorded. Without cassettes, the tests will either make live API calls or fail immediately since the fake audio content (b"fake audio content") cannot be processed by the real OpenAI Whisper API. You must:
- Record VCR cassettes with real audio files (e.g., a valid MP3) against the actual OpenAI API, then commit them to the cassettes directory
- Once cassettes are in place, run tests in playback mode to verify they pass without hitting the live API
Additionally, the instrument_legacy fixture parameter is unused in both test bodies but required for instrumentation side effects. Consider renaming to _instrument_legacy (with # noqa: ARG001) to signal intentional non-use, though this may not be enforced by your current linting setup.
Also applies to: 35-60
🧰 Tools
🪛 Ruff (0.14.8)
10-10: Unused function argument: instrument_legacy
(ARG001)
|
I see the issues, will fix them next week. |
2 participants
mutagen, included as optional dependency.["audio"]if he wishes to capture audio duration. This way mutagen will be included.Hello maintainers!
This is still a draft, but I still would like to ask for a review because this is a lot of changes and I'm not as familiar with this code base. Mainly, these questions:
Here's how attributes look like in OTEL collector:
Important
Adds OpenAI audio transcription instrumentation capturing audio duration and other parameters, with optional
mutagendependency and tests.transcription_wrapperandatranscription_wrapperinaudio_wrappers.pyto instrumentTranscriptions.createandAsyncTranscriptions.create.mutagen, model name, and other request/response parameters.mutagenas an optional dependency inpyproject.toml.extras = ["audio"]to includemutagen.test_audio_transcription.pywith tests for both sync and async transcription using mock audio files._instrumentand_uninstrumentmethods in__init__.pyto wrap and unwrap transcription functions.This description was created by
for b6db65e. You can customize this summary. It will automatically update as commits are pushed.
Summary by CodeRabbit
New Features
Tests
✏️ Tip: You can customize this high-level summary in your review settings.