fix(llmobs): fix parsing errors for bedrock converse stream (#13238)

Fixing a noop test discovered a couple bugs for parsing out the response
from converse stream
1. We were adding a `[{"role": "assistant"}]` message that didn't
contain any content blocks to be parsed out - the fix is to only try to
extract a message if it contains associated content block indices
2. `toolName` -> `name`

## Checklist
- [x] PR author has checked that all the criteria below are met
- The PR description includes an overview of the change
- The PR description articulates the motivation for the change
- The change includes tests OR the PR description describes a testing
strategy
- The PR description notes risks associated with the change, if any
- Newly-added code is easy to change
- The change follows the [library release note
guidelines](https://ddtrace.readthedocs.io/en/stable/releasenotes.html)
- The change includes or references documentation updates if necessary
- Backport labels are set (if
[applicable](https://ddtrace.readthedocs.io/en/latest/contributing.html#backporting))

## Reviewer Checklist
- [x] Reviewer has checked that all the criteria below are met 
- Title is accurate
- All changes are related to the pull request's stated goal
- Avoids breaking
[API](https://ddtrace.readthedocs.io/en/stable/versioning.html#interfaces)
changes
- Testing strategy adequately addresses listed risks
- Newly-added code is easy to change
- Release note makes sense to a user of the library
- If necessary, author has acknowledged and discussed the performance
implications of this PR as reported in the benchmarks PR comment
- Backport labels are set in a manner that is consistent with the
[release branch maintenance
policy](https://ddtrace.readthedocs.io/en/latest/contributing.html#backporting)

---------

Co-authored-by: lievan <[email protected]>

Author: lievan

ddtrace/llmobs/_integrations/bedrock.py

@@ -186,18 +186,18 @@ def _extract_output_message_for_converse_stream(
 
             if "messageStart" in chunk:
                 message_data = chunk["messageStart"]
-                current_message = {"role": message_data.get("role", "assistant"), "context_block_indices": []}
+                current_message = {"role": message_data.get("role", "assistant"), "content_block_indicies": []}
 
             # always make sure we have a current message
             if current_message is None:
-                current_message = {"role": "assistant", "context_block_indices": []}
+                current_message = {"role": "assistant", "content_block_indicies": []}
 
             if "contentBlockStart" in chunk:
                 block_start = chunk["contentBlockStart"]
                 index = block_start.get("contentBlockIndex")
 
                 if index is not None:
-                    current_message["context_block_indices"].append(index)
+                    current_message["content_block_indicies"].append(index)
                     if "start" in block_start and "toolUse" in block_start["start"]:
                         tool_content_blocks[index] = block_start["start"]["toolUse"]
 
@@ -206,8 +206,8 @@ def _extract_output_message_for_converse_stream(
                 index = content_block_delta.get("contentBlockIndex")
 
                 if index is not None and "delta" in content_block_delta:
-                    if index not in current_message.get("context_block_indices", []):
-                        current_message["context_block_indices"].append(index)
+                    if index not in current_message.get("content_block_indicies", []):
+                        current_message["content_block_indicies"].append(index)
 
                     delta_content = content_block_delta["delta"]
                     text_content_blocks[index] = text_content_blocks.get(index, "") + delta_content.get("text", "")
@@ -225,7 +225,7 @@ def _extract_output_message_for_converse_stream(
                 current_message = None
 
         # Handle the case where we didn't receive an explicit message stop event
-        if current_message is not None:
+        if current_message is not None and current_message.get("content_block_indicies"):
             messages.append(
                 get_final_message_converse_stream_message(current_message, text_content_blocks, tool_content_blocks)
             )

ddtrace/llmobs/_integrations/utils.py

@@ -908,7 +908,7 @@ def get_final_message_converse_stream_message(
     Returns:
         Dict containing the processed message with content and optional tool calls
     """
-    indices = sorted(message.get("context_block_indices", []))
+    indices = sorted(message.get("content_block_indicies", []))
     message_output = {"role": message["role"]}
 
     text_contents = [text_blocks[idx] for idx in indices if idx in text_blocks]
@@ -920,7 +920,7 @@ def get_final_message_converse_stream_message(
         if not tool_block:
             continue
         tool_call = {
-            "name": tool_block.get("toolName", ""),
+            "name": tool_block.get("name", ""),
             "tool_id": tool_block.get("toolUseId", ""),
         }
         tool_input = tool_block.get("input")

releasenotes/notes/converse-parsing-cd9ebb9506f8bbe1.yaml

@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    LLM Observability: This fix resolves an issue where tool call names were not being captured for bedrock `converse_stream` calls.
+  - |
+    LLM Observability: This fix resolves an issue where bedrock `converse_stream` calls contained an extra empty output message.

tests/contrib/botocore/test_bedrock_llmobs.py

@@ -735,12 +735,12 @@ def test_llmobs_converse(cls, bedrock_client, request_vcr, mock_tracer, llmobs_e
         span = mock_tracer.pop_traces()[0][0]
         assert len(llmobs_events) == 1
 
-        llmobs_events[0] == _expected_llmobs_llm_span_event(
+        assert llmobs_events[0] == _expected_llmobs_llm_span_event(
             span,
             model_name="claude-3-sonnet-20240229-v1:0",
             model_provider="anthropic",
             input_messages=[
-                {"role": "system", "content": request_params.get("system")},
+                {"role": "system", "content": request_params.get("system")[0]["text"]},
                 {"role": "user", "content": request_params.get("messages")[0].get("content")[0].get("text")},
             ],
             output_messages=[
@@ -813,12 +813,12 @@ def test_llmobs_converse_stream(cls, bedrock_client, request_vcr, mock_tracer, l
         span = mock_tracer.pop_traces()[0][0]
         assert len(llmobs_events) == 1
 
-        llmobs_events[0] == _expected_llmobs_llm_span_event(
+        assert llmobs_events[0] == _expected_llmobs_llm_span_event(
             span,
             model_name="claude-3-sonnet-20240229-v1:0",
             model_provider="anthropic",
             input_messages=[
-                {"role": "system", "content": request_params.get("system")},
+                {"role": "system", "content": request_params.get("system")[0]["text"]},
                 {"role": "user", "content": request_params.get("messages")[0].get("content")[0].get("text")},
             ],
             output_messages=[
@@ -835,7 +835,6 @@ def test_llmobs_converse_stream(cls, bedrock_client, request_vcr, mock_tracer, l
                 }
             ],
             metadata={
-                "stop_reason": "tool_use",
                 "temperature": request_params.get("inferenceConfig", {}).get("temperature"),
                 "max_tokens": request_params.get("inferenceConfig", {}).get("maxTokens"),
             },