🐛 Bug Report: Incomplete cache token migration after #4243 breaks Bedrock and Groq instrumentation

Open Beginner friendly
#4,665 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@Quantum-Enigma-dot is already working on this.

Since Oct 29, 2026.

  • #4667 by @Quantum-Enigma-dot — open

Assessment

Difficulty
2/5
Estimated time
1-3 hours
Newbie friendliness
76/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
python

Research direction

Start with the Bedrock and Groq span_utils.py files named in the issue, then search the repository for the two removed cache-token constants and check the current GenAIAttributes definitions. Run the prompt-caching test and the Groq span-utils tests listed in the report. Done means cache usage records successfully and remaining runtime and test references no longer fail against the workspace semantic-conventions source.

Written by the indexing model from the issue text.

Description

Which component is this bug for?

Bedrock Instrumentation

📜 Description

The semantic-convention migration in #4243 removed the following attributes from SpanAttributes:

  • SpanAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS
  • SpanAttributes.GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS

However, the Bedrock and Groq instrumentations still reference these attributes in runtime code.

When a response contains prompt-cache usage data, accessing the removed attributes raises AttributeError. Several tests in Bedrock, Groq, Google Generative AI, and Mistral AI also continue to reference the removed attributes.

This appears to be an incomplete migration from SpanAttributes to GenAIAttributes.

👟 Reproduction steps
  1. Clone the repository and enter the Bedrock instrumentation package:
cd packages/opentelemetry-instrumentation-bedrock

2.Run the prompt-caching test using the workspace version of opentelemetry-semantic-conventions-ai:
uv run --group test python -c "import sys; sys.path.insert(0, '../opentelemetry-semantic-conventions-ai'); import pytest; sys.exit(pytest.main(['tests/traces/test_prompt_caching.py']))"

3.Alternatively, verify directly that the attributes are absent:
```python
from opentelemetry.semconv_ai import SpanAttributes

print(
    hasattr(
        SpanAttributes,
        "GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS",
    )
)
# False

print(
    hasattr(
        SpanAttributes,
        "GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS",
    )
)
# False

4.Bedrock still accesses the missing attributes inside converse_usage_record():

SpanAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS
SpanAttributes.GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS

5.When cache usage is present, Python raises an AttributeError.

👍 Expected behavior
Bedrock and Groq should record prompt-cache token usage without raising an exception. All remaining references should use the current `GenAIAttributes` constants, or compatible aliases should remain available on `SpanAttributes` during the migration.

### 👎 Actual Behavior with Screenshots

When a cache-token code path accesses the removed constant, Python raises:

```text
AttributeError: type object 'SpanAttributes' has no attribute 'GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS'
The Bedrock prompt-caching test also fails when run against the current workspace version of opentelemetry-semantic-conventions-ai:
```text
FAILED tests/traces/test_prompt_caching.py::test_invoke_model_cache_tokens

The Groq span-utils suite produces six failures involving the same missing cache-read attribute.

🤖 Python Version

3.13.7

📃 Provide any additional context for the Bug.

I traced the regression to #4243.

That migration moved the cache-token constants to GenAIAttributes and removed them from SpanAttributes. Several packages were migrated, but remaining runtime references in Bedrock and Groq were not updated.

The issue can be masked when package-specific environments resolve the previously published opentelemetry-semantic-conventions-ai==0.5.1, where the attributes are still available. Running the tests against the current workspace semantic-conventions source exposes the regression.

Relevant runtime locations include:

  • packages/opentelemetry-instrumentation-bedrock/opentelemetry/instrumentation/bedrock/span_utils.py
  • packages/opentelemetry-instrumentation-groq/opentelemetry/instrumentation/groq/span_utils.py

A possible fix is to migrate the remaining runtime and test references to GenAIAttributes. Alternatively, compatibility aliases could be restored on SpanAttributes if backward compatibility is intended.

I searched the repository history and existing reports but did not find a later fix for these remaining references. I am willing to submit a PR.

👀 Have you spent some time to check if this bug has been raised before?
  • I checked and didn't find similar issue
Are you willing to submit PR?

Yes I am willing to submit a PR!

Dominant language
Python
Stars
7.5k
Forks
1.2k
Avg merge
2d 6h
Merged PRs (30d)
2

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from traceloop/openllmetry

All issues in traceloop/openllmetry

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.