Merge remote-tracking branch 'origin/main' into fix-thought-signature-pruning

Author: its-amann

src/google/adk/telemetry/_instrumentation.py

@@ -125,7 +125,6 @@ def record_llm_response(
 def _record_agent_metrics(
     agent_name: str,
     elapsed_s: float,
-    events: object,
     caught_error: Exception | None,
 ) -> None:
   try:
@@ -134,7 +133,6 @@ def _record_agent_metrics(
         elapsed_s,
         caught_error,
     )
-    _metrics.record_agent_workflow_steps(agent_name, events)
   except Exception:  # pylint: disable=broad-exception-caught
     logger.exception("Failed to record agent metrics for agent %s", agent_name)
 
@@ -196,7 +194,6 @@ async def record_agent_invocation(
     _record_agent_metrics(
         agent.name,
         _metrics.get_elapsed_s(span, start_time),
-        getattr(getattr(ctx, "session", None), "events", []),
         caught_error,
     )
     _flush_invoke_agent_metrics(tel_ctx, agent.name)

src/google/adk/telemetry/_metrics.py

@@ -27,7 +27,6 @@
 from opentelemetry.semconv.attributes import error_attributes
 
 if TYPE_CHECKING:
-  from google.adk.events.event import Event
   from google.adk.models.llm_request import LlmRequest
   from google.adk.models.llm_response import LlmResponse
   from opentelemetry.trace import Span
@@ -91,24 +90,6 @@
         81.92,
     ],
 )
-_agent_workflow_steps = meter.create_histogram(
-    "gen_ai.agent.workflow.steps",
-    unit="1",
-    description="Length of agentic workflow (# of events).",
-    explicit_bucket_boundaries_advisory=[
-        1,
-        2,
-        4,
-        8,
-        16,
-        32,
-        64,
-        128,
-        256,
-        512,
-        1024,
-    ],
-)
 _client_operation_duration = (
     gen_ai_metrics.create_gen_ai_client_operation_duration(meter)
 )
@@ -198,13 +179,6 @@ def record_invoke_agent_tool_calls(agent_name: str, count: int) -> None:
   _invoke_agent_tool_calls.record(count, attributes=attrs)
 
 
-def record_agent_workflow_steps(agent_name: str, events: list[Event]):
-  """Records the number of steps in the agent workflow by counting the number of events."""
-  attrs = {gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name}
-  count = sum(1 for event in events if event.author == agent_name)
-  _agent_workflow_steps.record(count, attributes=attrs)
-
-
 def record_tool_execution_duration(
     tool_name: str,
     tool_type: str,

src/google/adk/telemetry/node_tracing.py

@@ -186,10 +186,6 @@ def _use_invoke_workflow_span(
   # The flag rides along the otel_context propagated to child nodes, so nested
   # workflows see it set.
   nested = bool(context_api.get_value(_ENTRYPOINT_WORKFLOW_KEY, otel_context))
-  if not nested:
-    otel_context = context_api.set_value(
-        _ENTRYPOINT_WORKFLOW_KEY, True, otel_context
-    )
   attributes: dict[str, AttributeValue] = {
       GEN_AI_OPERATION_NAME: "invoke_workflow",
       GEN_AI_CONVERSATION_ID: conversation_id,
@@ -207,11 +203,14 @@ def _use_invoke_workflow_span(
   start_s = time.monotonic()
   workflow_span: Span | None = None
   try:
-    with tracer.start_as_current_span(
-        name=span_name,
-        attributes=attributes,
-        context=otel_context,
-    ) as span:
+    with (
+        tracer.start_as_current_span(
+            name=span_name,
+            attributes=attributes,
+            context=otel_context,
+        ) as span,
+        _mark_nested_workflows(),
+    ):
       workflow_span = span
       yield span
   finally:
@@ -221,3 +220,14 @@ def _use_invoke_workflow_span(
         nested=nested,
         error=sys.exc_info()[1],
     )
+
+
+@contextmanager
+def _mark_nested_workflows() -> Iterator[None]:
+  token = context_api.attach(
+      context_api.set_value(_ENTRYPOINT_WORKFLOW_KEY, True)
+  )
+  try:
+    yield
+  finally:
+    context_api.detach(token)