fix(openai-agents): emit cache_read.input_tokens and reasoning_tokens#4130
fix(openai-agents): emit cache_read.input_tokens and reasoning_tokens#4130dvirski wants to merge 4 commits into
Conversation
dvirski
requested review from
doronkopit5,
galzilber,
max-deygin-traceloop,
netanel-tl and
nina-kollman
|
|
📝 WalkthroughWalkthroughExtended OpenAI Agents instrumentation to extract cached input and reasoning tokens from response usage details. Updated dependency constraint to ChangesCached and Reasoning Token Extraction for OpenAI Agents
Sequence DiagramsequenceDiagram
participant ResponseUsage as Response.usage
participant ExtractHook as _extract_response_attributes
participant OTel_Span as OTel_Span
ResponseUsage->>ExtractHook: pass usage object
ExtractHook->>ExtractHook: read input_tokens_details.cached_tokens
ExtractHook->>OTel_Span: set_attribute(GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS, cached)
ExtractHook->>ExtractHook: read output_tokens_details.reasoning_tokens
ExtractHook->>OTel_Span: set_attribute(GEN_AI_USAGE_REASONING_TOKENS, reasoning)
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🧹 Nitpick comments (1)
packages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py (1)
554-557: ⚡ Quick winSelect the target span by attribute instead of positional index.
Using
response_spans[0]is brittle when multipleopenai.responsespans are emitted; pick the span that actually has the asserted attribute.More deterministic span selection
- response_span = response_spans[0] + response_span = next( + (s for s in response_spans if GenAIAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS in s.attributes), + None, + ) + assert response_span is not None- response_span = response_spans[0] + response_span = next( + (s for s in response_spans if SpanAttributes.GEN_AI_USAGE_REASONING_OUTPUT_TOKENS in s.attributes), + None, + ) + assert response_span is not NoneAlso applies to: 574-577
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@packages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py` around lines 554 - 557, The test currently picks a span by position (response_spans[0]) which is brittle; replace that with a deterministic selection that finds the span by the attribute you assert later (use response_spans and the specific attribute key used in the assertions). For example, replace uses of response_spans[0] in the test with code that selects the span via next(s for s in response_spans if '<ATTRIBUTE_KEY>' in s.attributes or s.attributes.get('<ATTRIBUTE_KEY>') == '<EXPECTED_VALUE>'), where <ATTRIBUTE_KEY> is the attribute name asserted later in the test; do this for both occurrences (around the response_spans usage at the two ranges noted) so the test picks the span by attribute, not index.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents/test_cache_read_input_tokens_recorded.yaml`:
- Around line 90-93: Remove all Cookie and set-cookie header values from the
cassette test_openai_agents/test_cache_read_input_tokens_recorded.yaml (and the
other occurrences noted) by redacting those header lines and re-recording the
cassette; ensure the test harness that generates cassettes filters or normalizes
"Cookie" request headers and "set-cookie" response headers before saving (update
the cassette recording/filtering logic to strip or replace these headers with a
stable placeholder) so future recordings do not include session cookies and
re-run the tests to commit the sanitized cassette.
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents/test_reasoning_output_tokens_recorded.yaml`:
- Around line 16-17: The cassette
test_openai_agents/test_reasoning_output_tokens_recorded.yaml contains
sensitive/volatile headers (e.g., Cookie and request/infra identifiers) that
must be filtered from VCR recordings; update the test recording configuration
used for this suite to strip or replace Cookie and any request/infra ID headers
(e.g., CF identifiers, X-Request-Id) during capture, re-record the cassette, and
commit the regenerated YAML so the cassette no longer contains those sensitive
values.
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py`:
- Around line 546-547: The test's cache-hit assertion currently uses the
expression ">= 0" which is too weak; locate the assertion that checks for a
cache hit (the line asserting ">= 0") in tests/test_openai_agents.py and change
it to use "> 0" so the second call is enforced as a cache read per the test
docstring; update any variable names involved (e.g., cache_hit_count or similar)
accordingly and run the test to confirm it now fails when there is no cache hit.
- Around line 544-545: The VCR configuration in conftest.py only filters
["authorization", "api-key"] and misses CloudFlare cookies, which are present in
cassettes like test_reasoning_output_tokens_recorded.yaml; update the VCR
config's filter_headers list to include "cookie" (i.e., modify the
filter_headers variable/setting in the VCR setup in conftest.py to add "cookie")
and then re-run the test suite to re-record the affected cassettes so
session/tracking cookies are scrubbed from the YAML files.
In
`@packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py`:
- Line 82: Add a deprecated alias for the renamed constant so existing imports
still work: in opentelemetry.semconv_ai.__init__.py (next to
GEN_AI_USAGE_REASONING_OUTPUT_TOKENS) define the old symbol
SpanAttributes.GEN_AI_USAGE_REASONING_TOKENS (or simply
GEN_AI_USAGE_REASONING_TOKENS if constants are module-level) and assign it the
same string value as GEN_AI_USAGE_REASONING_OUTPUT_TOKENS, include a brief
comment noting it is deprecated and optionally emit a DeprecationWarning when
accessed to guide users to the new constant name.
---
Nitpick comments:
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py`:
- Around line 554-557: The test currently picks a span by position
(response_spans[0]) which is brittle; replace that with a deterministic
selection that finds the span by the attribute you assert later (use
response_spans and the specific attribute key used in the assertions). For
example, replace uses of response_spans[0] in the test with code that selects
the span via next(s for s in response_spans if '<ATTRIBUTE_KEY>' in s.attributes
or s.attributes.get('<ATTRIBUTE_KEY>') == '<EXPECTED_VALUE>'), where
<ATTRIBUTE_KEY> is the attribute name asserted later in the test; do this for
both occurrences (around the response_spans usage at the two ranges noted) so
the test picks the span by attribute, not index.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 21e8e29b-0fa9-4aff-8aaf-b153db91fba1
⛔ Files ignored due to path filters (28)
packages/opentelemetry-instrumentation-agno/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-anthropic/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-bedrock/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-chromadb/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-cohere/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-crewai/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-groq/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-haystack/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-lancedb/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-marqo/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-mcp/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-milvus/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-mistralai/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-ollama/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-openai-agents/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-openai/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-pinecone/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-qdrant/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-replicate/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-sagemaker/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-together/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-transformers/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-vertexai/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-voyageai/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-watsonx/uv.lockis excluded by!**/*.lockpackages/opentelemetry-instrumentation-writer/uv.lockis excluded by!**/*.lockpackages/sample-app/uv.lockis excluded by!**/*.lockpackages/traceloop-sdk/uv.lockis excluded by!**/*.lock
📒 Files selected for processing (13)
packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/_hooks.pypackages/opentelemetry-instrumentation-openai-agents/pyproject.tomlpackages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents/test_cache_read_input_tokens_recorded.yamlpackages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents/test_reasoning_output_tokens_recorded.yamlpackages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/chat_wrappers.pypackages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/v1/responses_wrappers.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_azure.pypackages/opentelemetry-instrumentation-openai/tests/traces/test_chat.pypackages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.pypackages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/_testing.pypackages/traceloop-sdk/tests/test_manual.pypackages/traceloop-sdk/traceloop/sdk/tracing/manual.py
| Cookie: | ||
| - __cf_bm=GqvMTTwxqvHlYO_zMGr15vhngiHBKowc4W9cytloB1U-1778513729.5012145-1.0.1.1-TuWmsWma1J3OWFenDizlhbf_5qCYt89CjxNLLDDm.zLzVxhm2ep.2mEgG6rExy5xHOM0Quwrvd09cIn.yYTv.TO2jCa5bit4sQJ6ZIeHevWLiejQyIXZYqEUyXQItPzY |
There was a problem hiding this comment.
Scrub sensitive/volatile headers from cassette recordings.
Cookie and request/infra identifiers should not be committed in cassettes; they are sensitive/ephemeral and make fixtures less stable. Please filter these headers during VCR recording and regenerate this cassette.
Also applies to: 60-64, 104-105
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents/test_reasoning_output_tokens_recorded.yaml`
around lines 16 - 17, The cassette
test_openai_agents/test_reasoning_output_tokens_recorded.yaml contains
sensitive/volatile headers (e.g., Cookie and request/infra identifiers) that
must be filtered from VCR recordings; update the test recording configuration
used for this suite to strip or replace Cookie and any request/infra ID headers
(e.g., CF identifiers, X-Request-Id) during capture, re-record the cassette, and
commit the regenerated YAML so the cassette no longer contains those sensitive
values.
| @pytest.mark.vcr | ||
| def test_cache_read_input_tokens_recorded(exporter, test_agent): |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Locate the new cassettes referenced by these tests
fd -a 'test_(cache_read_input_tokens_recorded|reasoning_output_tokens_recorded)\.yaml' \
packages/opentelemetry-instrumentation-openai-agents/tests/cassettes
# Scan cassettes for common secret/PII indicators (review any hits; placeholders/masked values are acceptable)
rg -n --iglob '*.yaml' -e 'Authorization|Bearer\s+[A-Za-z0-9\-_\.]+|api[_-]?key|sk-[A-Za-z0-9]|@|email' \
packages/opentelemetry-instrumentation-openai-agents/tests/cassettesRepository: traceloop/openllmetry
Length of output: 365
🏁 Script executed:
#!/bin/bash
# Check the cassette files for any sensitive content
cd packages/opentelemetry-instrumentation-openai-agents/tests/cassettes/test_openai_agents
# Show file sizes and examine key sections of the cassettes
for cassette in test_cache_read_input_tokens_recorded.yaml test_reasoning_output_tokens_recorded.yaml; do
if [ -f "$cassette" ]; then
echo "=== $cassette ==="
wc -l "$cassette"
echo "--- First 50 lines ---"
head -50 "$cassette"
echo "--- Looking for potential secrets (detailed check) ---"
rg -n 'sk_|Bearer|Authorization|api.?key|token|secret|password' "$cassette" || echo "No matches found"
fi
doneRepository: traceloop/openllmetry
Length of output: 6564
🏁 Script executed:
#!/bin/bash
# Check the test file to see if VCR filters are configured
cd packages/opentelemetry-instrumentation-openai-agents/tests
# Search for VCR decorator and filter configuration around the test methods
rg -B5 -A10 'def test_(cache_read_input_tokens_recorded|reasoning_output_tokens_recorded)' test_openai_agents.pyRepository: traceloop/openllmetry
Length of output: 1274
🏁 Script executed:
#!/bin/bash
# Look for VCR configuration in conftest files
fd conftest.py packages/opentelemetry-instrumentation-openai-agents/tests | head -5
# Check the conftest for VCR filter setup
if [ -f packages/opentelemetry-instrumentation-openai-agents/tests/conftest.py ]; then
echo "=== conftest.py content ==="
cat -n packages/opentelemetry-instrumentation-openai-agents/tests/conftest.py
fiRepository: traceloop/openllmetry
Length of output: 11337
🏁 Script executed:
#!/bin/bash
# Also check for pytest.ini or setup.cfg with VCR config
find packages/opentelemetry-instrumentation-openai-agents/tests -maxdepth 1 \( -name "pytest.ini" -o -name "setup.cfg" -o -name "pyproject.toml" \) -exec echo "=== {} ===" \; -exec cat {} \;Repository: traceloop/openllmetry
Length of output: 47
VCR cassettes need improved filtering configuration and re-recording.
The conftest.py VCR config (lines 284–286) only filters ["authorization", "api-key"] headers, but the cassettes contain unfiltered CloudFlare bot management cookies (e.g., __cf_bm=... in test_reasoning_output_tokens_recorded.yaml). Add "cookie" to the filter_headers list and re-record the cassettes to ensure all session/tracking data is properly scrubbed.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@packages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py`
around lines 544 - 545, The VCR configuration in conftest.py only filters
["authorization", "api-key"] and misses CloudFlare cookies, which are present in
cassettes like test_reasoning_output_tokens_recorded.yaml; update the VCR
config's filter_headers list to include "cookie" (i.e., modify the
filter_headers variable/setting in the VCR setup in conftest.py to add "cookie")
and then re-run the test suite to re-record the affected cassettes so
session/tracking cookies are scrubbed from the YAML files.
| GEN_AI_CONTENT_COMPLETION_CHUNK = "gen_ai.content.completion.chunk" | ||
| GEN_AI_REQUEST_REASONING_EFFORT = "gen_ai.request.reasoning_effort" | ||
| GEN_AI_USAGE_REASONING_TOKENS = "gen_ai.usage.reasoning_tokens" | ||
| GEN_AI_USAGE_REASONING_OUTPUT_TOKENS = "gen_ai.usage.reasoning.output_tokens" |
There was a problem hiding this comment.
Preserve backward compatibility for the renamed reasoning constant.
Dropping SpanAttributes.GEN_AI_USAGE_REASONING_TOKENS can break consumers importing that symbol. Keep a deprecated alias pointing to the new key to avoid runtime import breaks.
Proposed compatibility patch
class SpanAttributes:
@@
GEN_AI_USAGE_REASONING_OUTPUT_TOKENS = "gen_ai.usage.reasoning.output_tokens"
+ # Backward-compatible alias (deprecated): remove in next major version
+ GEN_AI_USAGE_REASONING_TOKENS = GEN_AI_USAGE_REASONING_OUTPUT_TOKENS📝 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.
| GEN_AI_USAGE_REASONING_OUTPUT_TOKENS = "gen_ai.usage.reasoning.output_tokens" | |
| GEN_AI_USAGE_REASONING_OUTPUT_TOKENS = "gen_ai.usage.reasoning.output_tokens" | |
| # Backward-compatible alias (deprecated): remove in next major version | |
| GEN_AI_USAGE_REASONING_TOKENS = GEN_AI_USAGE_REASONING_OUTPUT_TOKENS |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py`
at line 82, Add a deprecated alias for the renamed constant so existing imports
still work: in opentelemetry.semconv_ai.__init__.py (next to
GEN_AI_USAGE_REASONING_OUTPUT_TOKENS) define the old symbol
SpanAttributes.GEN_AI_USAGE_REASONING_TOKENS (or simply
GEN_AI_USAGE_REASONING_TOKENS if constants are module-level) and assign it the
same string value as GEN_AI_USAGE_REASONING_OUTPUT_TOKENS, include a brief
comment noting it is deprecated and optionally emit a DeprecationWarning when
accessed to guide users to the new constant name.
|
Reviewed the change — direction is right (aligning attribute names with the OTel GenAI spec). A few things worth addressing: 1. Two sources of truth for the cache constantsThe same constants exist in both places, resolving to the same string today:
The PR mixes both: I'd lean toward standardizing on the upstream OTel-native 2.
|
|
Changes in response to review feedback
|
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In
`@packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/_hooks.py`:
- Around line 591-595: The current code sets the
GenAIAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS attribute even when
cached_tokens == 0; change the guard so you only set the attribute when
cached_tokens > 0. Locate the block in _hooks.py where cached_tokens =
getattr(input_details, "cached_tokens", None) and replace the if cached_tokens
is not None check with an explicit positive check (e.g., cached_tokens is not
None and cached_tokens > 0) before calling otel_span.set_attribute to ensure
only real cache hits are recorded.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: da513cee-a70f-4b55-983f-b65d5991a3f9
⛔ Files ignored due to path filters (1)
packages/traceloop-sdk/uv.lockis excluded by!**/*.lock
📒 Files selected for processing (2)
packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/_hooks.pypackages/opentelemetry-instrumentation-openai-agents/tests/test_openai_agents.py
| cached_tokens = getattr(input_details, "cached_tokens", None) | ||
| if cached_tokens is not None: | ||
| otel_span.set_attribute( | ||
| GenAIAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS, cached_tokens | ||
| ) |
There was a problem hiding this comment.
Only set cache-read usage on actual cache hits.
Line 592 currently records cache-read tokens even when cached_tokens == 0. Gate this to > 0 to match cache-hit semantics and avoid noisy attributes.
Suggested fix
- if cached_tokens is not None:
+ if cached_tokens is not None and cached_tokens > 0:
otel_span.set_attribute(
GenAIAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS, cached_tokens
)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/_hooks.py`
around lines 591 - 595, The current code sets the
GenAIAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS attribute even when
cached_tokens == 0; change the guard so you only set the attribute when
cached_tokens > 0. Locate the block in _hooks.py where cached_tokens =
getattr(input_details, "cached_tokens", None) and replace the if cached_tokens
is not None check with an explicit positive check (e.g., cached_tokens is not
None and cached_tokens > 0) before calling otel_span.set_attribute to ensure
only real cache hits are recorded.
895aa3c to
8d827c4
Compare
dvirski
force-pushed
the
dr/feat(semconv)-align-reasoning-and-cache-token-attributes-with-OTel-GenAI-spec
branch
from
8d827c4 to
d77ffe4
Compare
|
Re-reviewed after the latest push ( ✅ Resolved
|
doronkopit5
left a comment
doronkopit5
left a comment
There was a problem hiding this comment.
just fix the last comment form code rabbit about the non zero attribute
dvirski
changed the title
isn't zero cache token also meaningful? like it doesn't mean "no cache hit". Adding a check summary from otel spec: Spec check (semconv reference (https://opentelemetry.io/docs/specs/semconv/registry/attributes/gen-ai/)): gen_ai.usage.cache_read.input_tokens is classified as Recommended with no "when applicable" qualifier — the same level as gen_ai.usage.input_tokens and gen_ai.usage.output_tokens. The spec |
3 participants
Fixes #4060
Problem:
gen_ai.usage.reasoning_tokens was the wrong attribute name — the OTel GenAI spec defines it as gen_ai.usage.reasoning.output_tokens. Similarly, the LLM_USAGE_CACHE_READ_INPUT_TOKENS constant resolved to gen_ai.usage.cache_read_input_tokens (underscore) instead of
the spec-correct gen_ai.usage.cache_read.input_tokens (dot). The openai-agents instrumentor also wasn't recording cache read or reasoning tokens at all. The opentelemetry-semantic-conventions floor was too old (0.60b1) to provide the official upstream cache token constants.
Fix:
Bumped opentelemetry-semantic-conventions to >=0.62b1 so cache token constants come from the upstream OTel package. Added GEN_AI_USAGE_REASONING_OUTPUT_TOKENS = "gen_ai.usage.reasoning.output_tokens" to semconv_ai and removed the old incorrect constant. Updated the
openai and openai-agents instrumentors to use the corrected constants. Added two new tests with recorded cassettes for cache read and reasoning token recording in openai-agents.
Summary by CodeRabbit
New Features
Dependencies
Tests