feat(instrumentation): Set peer.service as the traced object name for client spans to mark the destination services

[LakshmiPriyaSujith]

• I have added tests that cover my changes.
• If adding a new instrumentation or changing an existing one, I've added screenshots from some observability platform showing the change.
• PR name follows conventional commits format: feat(instrumentation): ... or fix(instrumentation): ....
• (If applicable) I have updated the documentation accordingly.

---

Important

Adds peer.service attribute to client spans across multiple instrumentation packages to mark destination services.

• Behavior:
  • Adds SpanAttributes.PEER_SERVICE to client spans in _wrap() and _awrap() functions in anthropic/__init__.py, groq/__init__.py, and watsonx/__init__.py.
  • Adds SpanAttributes.PEER_SERVICE to with_instrumentation() in bedrock/__init__.py and sagemaker/__init__.py.
  • Adds SpanAttributes.PEER_SERVICE to wrap_agent_execute_task(), wrap_task_execute(), and wrap_llm_call() in crewai/instrumentation.py.
  • Adds SpanAttributes.PEER_SYSTEM to _create_llm_span() in langchain/callback_handler.py.
• Misc:
  • Adds PEER_SERVICE to SpanAttributes in semconv_ai/__init__.py.

^{This description was created by for 8256dd5. You can customize this summary. It will automatically update as commits are pushed.}

[CLAassistant]

All committers have signed the CLA.

[ellipsis-dev]

Caution

Changes requested ❌

Reviewed everything up to 8256dd5 in 1 minute and 37 seconds. Click for details.

• Reviewed 303 lines of code in 8 files
• Skipped 0 files when reviewing.
• Skipped posting 3 draft 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-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py:49

• Draft comment:
  Consider handling the commented-out LLM_RESPONSE_ID. Either implement it or remove the commented code if not needed.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

2. packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py:35

• Draft comment:
  Check naming consistency: some attributes use 'gen_ai.' while others use 'llm.' prefixes (e.g., LLM_USAGE_PROMPT_TOKENS vs LLM_USAGE_TOTAL_TOKENS). Consider standardizing the naming.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

3. packages/opentelemetry-semantic-conventions-ai/opentelemetry/semconv_ai/__init__.py:24

• Draft comment:
  Review the value for LLM_COMPLETIONS_EXCEPTIONS which references OpenAI conventions. Ensure it aligns with overall semantic conventions for all supported LLMs.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

Workflow ID: wflow_nUIVNKvyONunEykS

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

[nirga]

Hey @LakshmiPriyaSujith thanks for this - can you explain this change? I'm not sure why the object is the peer service that you want to set

[LakshmiPriyaSujith]

Hey @LakshmiPriyaSujith thanks for this - can you explain this change? I'm not sure why the object is the peer service that you want to set

Hi @nirga! This is my thought. Since LLM traces have only client spans, the destinations of LLM calls are unspecified. I thought it would be better to use the instrumented object as the remote service rather can using a constant value like Watsonx or Bedrock. For example, in the above screenshot, the span bedrock.completion represents a call from the client application to Bedrock implemented via ClientCreator. Do you think there's a better way to mark the destination service?

[nirga]

Can you give an example of when it won't be bedrock?

[LakshmiPriyaSujith]

Can you give an example of when it won't be bedrock?

Here's an example using Langchain instrumentation.