Allow for tracing via context provider (#5889)

These changes allow for 2 important use-cases:
1. Add a span for tool calls which will enable tracing of all tool calls
in agent_chat
2. Allow runtimes to pick up global `tracer_providers` if they are
available. This is very helpful because it allows for nested teams/agent
to all use the same tracer.
---------

Co-authored-by: Eric Zhu <ekzhu@users.noreply.github.com>

python/packages/autogen-core/src/autogen_core/_telemetry/_tracing.py

@@ -1,7 +1,7 @@
 import contextlib
 from typing import Dict, Generic, Iterator, Optional
 
-from opentelemetry.trace import NoOpTracerProvider, Span, SpanKind, TracerProvider
+from opentelemetry.trace import NoOpTracerProvider, Span, SpanKind, TracerProvider, get_tracer_provider
 from opentelemetry.util import types
 
 from ._propagation import TelemetryMetadataContainer, get_telemetry_links
@@ -22,9 +22,10 @@ class TraceHelper(Generic[Operation, Destination, ExtraAttributes]):
         tracer_provider: TracerProvider | None,
         instrumentation_builder_config: TracingConfig[Operation, Destination, ExtraAttributes],
     ) -> None:
-        self.tracer = (tracer_provider if tracer_provider else NoOpTracerProvider()).get_tracer(
-            f"autogen {instrumentation_builder_config.name}"
-        )
+        # Evaluate in order: first try tracer_provider param, then get_tracer_provider(), finally fallback to NoOp
+        # This allows for nested tracing with a default tracer provided by the user
+        self.tracer_provider = tracer_provider or get_tracer_provider() or NoOpTracerProvider()
+        self.tracer = self.tracer_provider.get_tracer(f"autogen {instrumentation_builder_config.name}")
         self.instrumentation_builder_config = instrumentation_builder_config
 
     @contextlib.contextmanager

python/packages/autogen-core/src/autogen_core/tools/_base.py

@@ -5,6 +5,7 @@ from collections.abc import Sequence
 from typing import Any, Dict, Generic, Mapping, Protocol, Type, TypedDict, TypeVar, cast, runtime_checkable
 
 import jsonref
+from opentelemetry.trace import get_tracer
 from pydantic import BaseModel
 from typing_extensions import NotRequired
 
@@ -147,8 +148,16 @@ class BaseTool(ABC, Tool, Generic[ArgsT, ReturnT], ComponentBase[BaseModel]):
     async def run(self, args: ArgsT, cancellation_token: CancellationToken) -> ReturnT: ...
 
     async def run_json(self, args: Mapping[str, Any], cancellation_token: CancellationToken) -> Any:
-        # Execute the tool's run method
-        return_value = await self.run(self._args_type.model_validate(args), cancellation_token)
+        with get_tracer("base_tool").start_as_current_span(
+            self._name,
+            attributes={
+                "tool_name": self._name,
+                "tool_description": self._description,
+                "tool_args": json.dumps(args),
+            },
+        ):
+            # Execute the tool's run method
+            return_value = await self.run(self._args_type.model_validate(args), cancellation_token)
 
         # Log the tool call event
         event = ToolCallEvent(