refactor: move exception handling from metric emission into instrumentation handlers

Metric instrumentation code should never disrupt the regular execution flow, so with this we capture any errors one level higher and make instrumentation more robust (especially against tests that use incomplete mocks of adk components).

PiperOrigin-RevId: 904412190

Author: google-genai-bot

src/google/adk/telemetry/_instrumentation.py

@@ -83,23 +83,28 @@ async def record_agent_invocation(
     with tracing.tracer.start_as_current_span(span_name) as s:
       span = s
       tracing.trace_agent_invocation(span, agent, ctx)
-      _metrics.record_agent_request_size(agent.name, ctx.user_content)
       tel_ctx = TelemetryContext(otel_context=context_api.get_current())
       yield tel_ctx
   except Exception as e:
     caught_error = e
     raise
   finally:
     elapsed_ms = _get_elapsed_ms(span, start_time)
-    _metrics.record_agent_invocation_duration(
-        agent.name,
-        elapsed_ms,
-        ctx.user_content,
-        ctx.session.events,
-        caught_error,
-    )
-    _metrics.record_agent_response_size(agent.name, ctx.session.events)
-    _metrics.record_agent_workflow_steps(agent.name, len(ctx.session.events))
+    try:
+      _metrics.record_agent_invocation_duration(
+          agent.name,
+          elapsed_ms,
+          ctx.user_content,
+          ctx.session.events,
+          caught_error,
+      )
+      _metrics.record_agent_request_size(agent.name, ctx.user_content)
+      _metrics.record_agent_response_size(agent.name, ctx.session.events)
+      _metrics.record_agent_workflow_steps(agent.name, len(ctx.session.events))
+    except Exception:  # pylint: disable=broad-exception-caught
+      logger.exception(
+          "Failed to record agent metrics for agent %s", agent.name
+      )
 
 
 @contextlib.asynccontextmanager
@@ -144,11 +149,16 @@ async def record_tool_execution(
         if isinstance(result_event, event_lib.Event)
         else None
     )
-    _metrics.record_tool_execution_duration(
-        tool_name=tool.name,
-        agent_name=agent.name,
-        elapsed_ms=elapsed_ms,
-        input_content=invocation_context.user_content,
-        output_content=output_content,
-        error=caught_error,
-    )
+    try:
+      _metrics.record_tool_execution_duration(
+          tool_name=tool.name,
+          agent_name=agent.name,
+          elapsed_ms=elapsed_ms,
+          input_content=invocation_context.user_content,
+          output_content=output_content,
+          error=caught_error,
+      )
+    except Exception:  # pylint: disable=broad-exception-caught
+      logger.exception(
+          "Failed to record tool execution duration for tool %s", tool.name
+      )

src/google/adk/telemetry/_metrics.py

@@ -69,17 +69,12 @@ def record_agent_request_size(
     agent_name: str, user_content: types.Content | None
 ):
   """Records the size of the agent request."""
-  try:
-    size = _get_content_size(user_content)
-    attrs = {
-        gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
-        GEN_AI_INPUT_TYPE: _get_modality_from_content(user_content),
-    }
-    _agent_request_size.record(size, attributes=attrs)
-  except Exception:  # pylint: disable=broad-exception-caught
-    logger.exception(
-        "Failed to record agent request size for agent %s", agent_name
-    )
+  size = _get_content_size(user_content)
+  attrs = {
+      gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
+      GEN_AI_INPUT_TYPE: _get_modality_from_content(user_content),
+  }
+  _agent_request_size.record(size, attributes=attrs)
 
 
 def record_agent_invocation_duration(
@@ -90,64 +85,49 @@ def record_agent_invocation_duration(
     error: Exception | None = None,
 ):
   """Records the duration of the agent invocation."""
-  try:
-    response_content: types.Content | None = None
-    for event in reversed(events):
-      if event.author == agent_name and event.content:
-        response_content = event.content
-        break
-
-    attrs = {
-        gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
-        GEN_AI_INPUT_TYPE: _get_modality_from_content(user_content),
-        gen_ai_attributes.GEN_AI_OUTPUT_TYPE: _get_modality_from_content(
-            response_content
-        ),
-    }
-    if error is not None:
-      attrs[error_attributes.ERROR_TYPE] = type(error).__name__
-    _agent_invocation_duration.record(elapsed_ms, attributes=attrs)
-  except Exception:  # pylint: disable=broad-exception-caught
-    logger.exception(
-        "Failed to record agent invocation duration for agent %s", agent_name
-    )
+  response_content: types.Content | None = None
+  for event in reversed(events):
+    if event.author == agent_name and event.content:
+      response_content = event.content
+      break
+
+  attrs = {
+      gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
+      GEN_AI_INPUT_TYPE: _get_modality_from_content(user_content),
+      gen_ai_attributes.GEN_AI_OUTPUT_TYPE: _get_modality_from_content(
+          response_content
+      ),
+  }
+  if error is not None:
+    attrs[error_attributes.ERROR_TYPE] = type(error).__name__
+  _agent_invocation_duration.record(elapsed_ms, attributes=attrs)
 
 
 def record_agent_response_size(agent_name: str, events: list[Event]):
   """Records the size of the agent response by extracting content from events."""
-  try:
-    response_content: types.Content | None = None
-    for event in reversed(events):
-      # Need to look for author matching agent_name and having content
-      if event.author == agent_name and event.content:
-        response_content = event.content
-        break
-
-    size = _get_content_size(response_content)
-    attrs = {
-        gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
-        gen_ai_attributes.GEN_AI_OUTPUT_TYPE: _get_modality_from_content(
-            response_content
-        ),
-    }
-    _agent_response_size.record(size, attributes=attrs)
-  except Exception:  # pylint: disable=broad-exception-caught
-    logger.exception(
-        "Failed to record agent response size for agent %s", agent_name
-    )
+  response_content: types.Content | None = None
+  for event in reversed(events):
+    # Need to look for author matching agent_name and having content
+    if event.author == agent_name and event.content:
+      response_content = event.content
+      break
+
+  size = _get_content_size(response_content)
+  attrs = {
+      gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
+      gen_ai_attributes.GEN_AI_OUTPUT_TYPE: _get_modality_from_content(
+          response_content
+      ),
+  }
+  _agent_response_size.record(size, attributes=attrs)
 
 
 def record_agent_workflow_steps(agent_name: str, steps_count: int):
   """Records the number of steps in the agent workflow."""
-  try:
-    attrs = {
-        gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
-    }
-    _agent_workflow_steps.record(steps_count, attributes=attrs)
-  except Exception:  # pylint: disable=broad-exception-caught
-    logger.exception(
-        "Failed to record agent workflow steps for agent %s", agent_name
-    )
+  attrs = {
+      gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
+  }
+  _agent_workflow_steps.record(steps_count, attributes=attrs)
 
 
 def record_tool_execution_duration(
@@ -159,23 +139,18 @@ def record_tool_execution_duration(
     error: Exception | None = None,
 ):
   """Records the duration of the tool execution."""
-  try:
-    attrs = {
-        gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
-        gen_ai_attributes.GEN_AI_TOOL_NAME: tool_name,
-        GEN_AI_INPUT_TYPE: _get_modality_from_content(input_content),
-    }
-    if error is not None:
-      attrs[error_attributes.ERROR_TYPE] = type(error).__name__
-    else:
-      attrs[gen_ai_attributes.GEN_AI_OUTPUT_TYPE] = _get_modality_from_content(
-          output_content
-      )
-    _tool_execution_duration.record(elapsed_ms, attributes=attrs)
-  except Exception:  # pylint: disable=broad-exception-caught
-    logger.exception(
-        "Failed to record tool execution duration for tool %s", tool_name
+  attrs = {
+      gen_ai_attributes.GEN_AI_AGENT_NAME: agent_name,
+      gen_ai_attributes.GEN_AI_TOOL_NAME: tool_name,
+      GEN_AI_INPUT_TYPE: _get_modality_from_content(input_content),
+  }
+  if error is not None:
+    attrs[error_attributes.ERROR_TYPE] = type(error).__name__
+  else:
+    attrs[gen_ai_attributes.GEN_AI_OUTPUT_TYPE] = _get_modality_from_content(
+        output_content
     )
+  _tool_execution_duration.record(elapsed_ms, attributes=attrs)
 
 
 # Helper functions copied from metrics_plugin.py