opentelemetry-instrumentation-botocore fails to parse tool use input

### Describe your environment

OS: MacOs
Python version: 3.12.10
Package version: opentelemetry-instrumentation 0.59b0


### What happened?

What happened?

Getting `Type Error JSON object must be str, bytes or bytearray, not dict` while parsing tool use input in [_decode_tool_use](https://github.com/open-telemetry/opentelemetry-python-contrib/blob/428af02fdc4e3150d6f409e61cffbc7edd051710/instrumentation/opentelemetry-instrumentation-botocore/src/opentelemetry/instrumentation/botocore/extensions/bedrock_utils.py#L39)

### Steps to Reproduce

1. Create `run_with_otel.sh` (see below)
2. Create `bedrock_empty_tool_args.py` (see below)
3. In one terminal, run `./run_with_otel.sh`
4. In second terminal, run `curl -X POST http://localhost:8080/invocations -H "Content-Type: application/json" -d '{"input": "what are my accounts?"}'`

Callout: If I run with no instrumentation (`python bedrock_empty_tool_args.py`), call works just fine.


### Scripts:
* `run_with_otel.sh` :
```
#!/bin/bash

# Install dependencies
pip freeze | xargs pip uninstall -y
pip install boto3 langchain-aws langgraph bedrock-agentcore 
pip install opentelemetry-distro==0.59b0
opentelemetry-bootstrap -a install

# Set AWS credentials
export AWS_ACCESS_KEY_ID=*********
export AWS_SECRET_ACCESS_KEY=*********
export AWS_SESSION_TOKEN=*********

# Set environment variables
OTEL_TRACES_EXPORTER=console \
OTEL_METRICS_EXPORTER=none \
OTEL_LOGS_EXPORTER=none \
opentelemetry-instrument python bedrock_empty_tool_args.py
```
* `bedrock_empty_tool_args.py`:
```
import asyncio
import os

from bedrock_agentcore import BedrockAgentCoreApp
from langchain_aws import ChatBedrock
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.tools import tool
from langgraph.graph import MessagesState, StateGraph
from langgraph.prebuilt import ToolNode, tools_condition

from langchain_core.runnables import RunnableConfig
from langgraph.graph.message import BaseMessage
from langgraph.graph.state import CompiledStateGraph
from langgraph.types import Command
from langchain_core.messages.ai import AIMessageChunk
from langgraph.typing import ContextT, OutputT, StateT, InputT
import logging

logger = logging.getLogger(__name__)


class StreamingQueue:
    """Simple async queue for streaming responses."""

    def __init__(self):
        self._queue = asyncio.Queue()
        self._finished = False

    async def put(self, data: str) -> None:
        """Add an item to the queue."""
        await self._queue.put({"event": "message", "data": data})

    async def finish(self) -> None:
        """Mark the queue as finished and add sentinel value."""
        self._finished = True
        await self._queue.put(None)

    async def stream(self):
        """Stream items from the queue until finished."""
        while True:
            item = await self._queue.get()
            if item is None and self._finished:
                break
            yield item

async def stream_messages(
    graph: CompiledStateGraph[StateT, ContextT, InputT, OutputT],
    streaming_queue: StreamingQueue,
    graph_input: InputT | Command,
    graph_config: RunnableConfig | None = None,
) -> None:
    """Stream graph response with stream mode 'messages'"""
    async for chunk in graph.astream(
        input=graph_input,
        stream_mode="messages",
        config=graph_config,
    ):
        # Handle the streaming chunk properly - it's a tuple with (message, metadata)
        if chunk and len(chunk) > 0:
            message_chunk: BaseMessage = chunk[0]  # type: ignore
            logger.info("Chunk: %s", message_chunk)
            if isinstance(message_chunk, AIMessageChunk):
                # Do not send the tool messages to the user
                await streaming_queue.put(message_chunk.text())
        else:
            logger.info("Invalid chunk: %s", chunk)
            await streaming_queue.put("We cannot get an answer at this moment. Try again later.")

app = BedrockAgentCoreApp()
queue = StreamingQueue()


@tool
def salesforce_search():
    """Tool used when you need to search for an account name in our Salesforce system"""
    search_results = {"records": [{"account_name": "Test account", "account_id": "ABC123XYZ"}]}
    return search_results


@tool
def salesforce_account_details(account_id: str):
    """Tool used when you need to search for an account id in our Salesforce system"""
    search_results = {
        "account_name": "Test account",
        "account_id": f"{account_id}",
        "arr": 1312.00,
    }
    return search_results


def create_agent():
    """Create and configure the LangGraph agent"""
    # Initialize your LLM (adjust model and parameters as needed)
    llm = ChatBedrock(
        model="us.anthropic.claude-sonnet-4-20250514-v1:0",
        model_kwargs={"temperature": 0.1},
        region=os.environ.get("AWS_REGION_NAME", "us-east-1"),
    )

    # Bind tools to the LLM
    tools = [salesforce_search, salesforce_account_details]
    llm_with_tools = llm.bind_tools(tools)

    # System message
    system_message = """You are the Salesforce Search Assistant, you can help users find the right details for their accounts.
                        CAPABILITIES:
                        - You can search for matching accounts with the salesforce_search tool
                        - You can look up account details with the salesforce_account_details tool
                        """

    # Define the chatbot node
    async def chatbot(state: MessagesState):
        raw_messages = state["messages"]

        # Remove any existing system messages to avoid duplicates or misplacement
        non_system_messages = [msg for msg in raw_messages if not isinstance(msg, SystemMessage)]

        # Always ensure SystemMessage is first
        messages = [SystemMessage(content=system_message), *non_system_messages]

        # Get response from model with tools bound
        response = await llm_with_tools.ainvoke(messages)

        # Append response to full message history
        return {"messages": [*raw_messages, response]}

    # Create the graph
    graph_builder = StateGraph(MessagesState)

    # Add nodes
    graph_builder.add_node("chatbot", chatbot)
    graph_builder.add_node("tools", ToolNode(tools))

    # Add edges
    graph_builder.add_conditional_edges(
        "chatbot",
        tools_condition,
    )
    graph_builder.add_edge("tools", "chatbot")

    # Set entry point
    graph_builder.set_entry_point("chatbot")

    # Compile the graph
    return graph_builder.compile()


async def agent_task(payload):
    """Invoke the agent with a payload"""
    await queue.put("Begin agent execution")

    try:
        user_input = payload.get(
            "input",
            "No user input provided",
        )

        agent = create_agent()
        #response = await agent.ainvoke({"messages": [HumanMessage(content=user_input)]})
        await stream_messages(agent, queue, {"messages": [HumanMessage(content=user_input)]})
        # await queue.put(response["messages"][-1].content)
    except Exception as e:
        logger.exception(e)
        await queue.put(f"An exception occurred: {e!s}")
    finally:
        await queue.finish()


@app.entrypoint
async def agent_invocation(payload, context):
    """Main entrypoint for agent invocation."""

    # Create and start the agent task
    task = asyncio.create_task(agent_task(payload=payload))

    async def stream_with_task():
        """Stream results while ensuring task completion."""
        async for item in queue.stream():
            yield item
        await task  # Ensure task completes

    return stream_with_task()


if __name__ == "__main__":
    logger.info("Starting Agent")
    app.run()
```

### Expected Result

* In second terminal:
```
thp@842f577dc4b0 Downloads % curl -X POST http://localhost:8080/invocations -H "Content-Type: application/json" -d '{"input": "what are my accounts?"}'
data: {"event": "message", "data": "Begin agent execution"}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": "I"}

data: {"event": "message", "data": "'ll help you find your accounts in Sales"}

data: {"event": "message", "data": "force. Let me search for them"}

data: {"event": "message", "data": " now."}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": "Base"}

data: {"event": "message", "data": "d on the search results, I"}

data: {"event": "message", "data": " found one account associated with you:\n\n**"}

data: {"event": "message", "data": "Account Name:** Test account  "}

data: {"event": "message", "data": "\n**Account ID:** ABC123X"}

data: {"event": "message", "data": "YZ\n\nWould you like me to"}

data: {"event": "message", "data": " get more detailed information about this account,"}

data: {"event": "message", "data": " or do you need help with anything else regarding"}

data: {"event": "message", "data": " your Salesforce accounts?"}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}
```

### Actual Result

* In second terminal:
```
curl -X POST http://localhost:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": "what are my accounts?"}'
data: {"event": "message", "data": "Begin agent execution"}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": "I'll help you find your accounts in"}

data: {"event": "message", "data": " Salesforce. Let me search for"}

data: {"event": "message", "data": " them now."}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": ""}

data: {"event": "message", "data": "An exception occurred: the JSON object must be str, bytes or bytearray, not dict"}
```
* In first terminal:
```
the JSON object must be str, bytes or bytearray, not dict
Traceback (most recent call last):
  File "<sanitized>/bedrock_empty_tool_args.py", line 160, in agent_task
    await stream_messages(agent, queue, {"messages": [HumanMessage(content=user_input)]})
  File "<sanitized>/bedrock_empty_tool_args.py", line 53, in stream_messages
    async for chunk in graph.astream(
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/pregel/main.py", line 2993, in astream
    async for _ in runner.atick(
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/pregel/_runner.py", line 401, in atick
    _panic_or_proceed(
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/pregel/_runner.py", line 511, in _panic_or_proceed
    raise exc
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/pregel/_retry.py", line 137, in arun_with_retry
    return await task.proc.ainvoke(task.input, config)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/_internal/_runnable.py", line 706, in ainvoke
    input = await asyncio.create_task(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/asyncio/__init__.py", line 299, in trace_coroutine
    return await coro
           ^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langgraph/_internal/_runnable.py", line 474, in ainvoke
    ret = await self.afunc(*args, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>/bedrock_empty_tool_args.py", line 122, in chatbot
    response = await llm_with_tools.ainvoke(messages)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/runnables/base.py", line 5724, in ainvoke
    return await self.bound.ainvoke(
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 417, in ainvoke
    llm_result = await self.agenerate_prompt(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 1036, in agenerate_prompt
    return await self.agenerate(
           ^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 994, in agenerate
    raise exceptions[0]
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/asyncio/__init__.py", line 299, in trace_coroutine
    return await coro
           ^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 1153, in _agenerate_with_cache
    async for chunk in self._astream(messages, stop=stop, **kwargs):
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 1281, in _astream
    item = await run_in_executor(
           ^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/runnables/config.py", line 611, in run_in_executor
    return await asyncio.get_running_loop().run_in_executor(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/concurrent/futures/thread.py", line 59, in run
    result = self.fn(*self.args, **self.kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/threading/__init__.py", line 171, in wrapped_func
    return original_func(*func_args, **func_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_core/runnables/config.py", line 602, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_aws/chat_models/bedrock.py", line 869, in _stream
    for chunk in self._prepare_input_and_invoke_stream(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_aws/llms/bedrock.py", line 1165, in _prepare_input_and_invoke_stream
    for chunk in LLMInputOutputAdapter.prepare_output_stream(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/langchain_aws/llms/bedrock.py", line 509, in prepare_output_stream
    for event in stream:
                 ^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/botocore/extensions/bedrock_utils.py", line 211, in __iter__
    self._process_event(event)
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/botocore/extensions/bedrock_utils.py", line 233, in _process_event
    self._process_anthropic_claude_chunk(chunk)
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/botocore/extensions/bedrock_utils.py", line 373, in _process_anthropic_claude_chunk
    _decode_tool_use(self._content_block)
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/site-packages/opentelemetry/instrumentation/botocore/extensions/bedrock_utils.py", line 43, in _decode_tool_use
    tool_use["input"] = json.loads(tool_use["input"])
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<sanitized>.local/share/mise/installs/python/3.12.10/lib/python3.12/json/__init__.py", line 339, in loads
    raise TypeError(f'the JSON object must be str, bytes or bytearray, '
TypeError: the JSON object must be str, bytes or bytearray, not dict
During task with name 'chatbot' and id '5ca8992d-7886-79b4-af0e-86f7e4e8518b'
```

### Additional context

See https://github.com/open-telemetry/opentelemetry-python-contrib/issues/3843

### Would you like to implement a fix?

Yes

### Tip

<sub>[React](https://github.blog/news-insights/product-news/add-reactions-to-pull-requests-issues-and-comments/) with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding `+1` or `me too`, to help us triage it. Learn more [here](https://opentelemetry.io/community/end-user/issue-participation/).</sub>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

opentelemetry-instrumentation-botocore fails to parse tool use input #3874

Describe your environment

What happened?

Steps to Reproduce

Scripts:

Expected Result

Actual Result

Additional context

Would you like to implement a fix?

Tip

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

opentelemetry-instrumentation-botocore fails to parse tool use input #3874

Description

Describe your environment

What happened?

Steps to Reproduce

Scripts:

Expected Result

Actual Result

Additional context

Would you like to implement a fix?

Tip

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions