fix(telemetry): set error.type on spans that end in an error

When a model call or tool failed, OpenTelemetry marked every span the
exception passed through as ERROR, but only execute_tool said which
error it was. The invocation, invoke_workflow, invoke_agent, call_llm
and generate_content spans had no error.type, even though the metrics
for the same operations record it.

Open these spans through a small helper that sets error.type from
resolve_error_type() when an exception escapes, and set it where a
workflow node hands its failure back as data. The functional goldens
for the four error cases are re-recorded, and the generate_content
divergence from the OTel instrumentor is now a value difference (429
vs ClientError), the same one already accepted for the duration metric.

Fixes #7494

Author: ankit2235

src/google/adk/flows/llm_flows/core/_model_call.py

@@ -33,8 +33,8 @@
 from ....models.llm_request import LlmRequest
 from ....models.llm_response import LlmResponse
 from ....telemetry import _instrumentation
+from ....telemetry.tracing import start_as_current_span
 from ....telemetry.tracing import trace_call_llm
-from ....telemetry.tracing import tracer
 from ....utils._runner_utils import _with_caller_context
 from ....utils.context_utils import Aclosing
 from ._finalizer import has_meaningful_content
@@ -184,7 +184,7 @@ async def call_llm_async(
   caller_context = otel_context.get_current()
 
   async def _call_llm_with_tracing() -> AsyncGenerator[LlmResponse, None]:
-    with tracer.start_as_current_span('call_llm') as span:
+    with start_as_current_span('call_llm') as span:
       # Runs before_model_callback inside the call_llm span so
       # plugins observe the same span as after/error callbacks.
       if response := await flow._handle_before_model_callback(

src/google/adk/telemetry/_instrumentation.py

@@ -153,7 +153,7 @@ def record_invocation(
     Nothing; the span (if any) is active for the duration of the block.
   """
   if resolve_schema_version() < SCHEMA_VERSION_SEMCONV_ALIGNED:
-    with tracing.tracer.start_as_current_span("invocation"):
+    with tracing.start_as_current_span("invocation"):
       yield
     return
 
@@ -582,7 +582,7 @@ async def record_agent_invocation(
       context_api.set_value(_AGENT_INVOCATION_SCOPE_KEY, scope)
   )
   try:
-    with tracing.tracer.start_as_current_span(span_name) as s:
+    with tracing.start_as_current_span(span_name) as s:
       span = s
       tracing.trace_agent_invocation(span, agent, ctx)
       yield scope

src/google/adk/telemetry/node_tracing.py

@@ -28,6 +28,7 @@
 from opentelemetry import context as context_api
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import GEN_AI_CONVERSATION_ID
 from opentelemetry.semconv._incubating.attributes.gen_ai_attributes import GEN_AI_OPERATION_NAME
+from opentelemetry.semconv.attributes.error_attributes import ERROR_TYPE
 from opentelemetry.trace import Span
 from opentelemetry.trace import Status
 from opentelemetry.trace import StatusCode
@@ -37,7 +38,8 @@
 from ..workflow._base_node import BaseNode
 from .context import TelemetryConfig
 from .tracing import _telemetry_config_from_invocation_context
-from .tracing import tracer
+from .tracing import resolve_error_type
+from .tracing import start_as_current_span
 
 if TYPE_CHECKING:
   from opentelemetry.util.types import AttributeValue
@@ -153,7 +155,7 @@ def _invoke_node_span(
     context: Context, node: BaseNode
 ) -> Iterator[TelemetryContext]:
   """Opens an `invoke_node` span for a plain node."""
-  with tracer.start_as_current_span(
+  with start_as_current_span(
       f"invoke_node {node.name}",
       attributes={
           GEN_AI_OPERATION_NAME: "invoke_node",
@@ -247,7 +249,7 @@ def _use_invoke_workflow_span(
   recorded_error: BaseException | None = None
   try:
     with (
-        tracer.start_as_current_span(
+        start_as_current_span(
             name=span_name,
             attributes=attributes,
             context=otel_context,
@@ -272,6 +274,7 @@ def _use_invoke_workflow_span(
           recorded_error = get_recorded_error()
           if recorded_error is not None:
             span.record_exception(recorded_error)
+            span.set_attribute(ERROR_TYPE, resolve_error_type(recorded_error))
             span.set_status(Status(StatusCode.ERROR, str(recorded_error)))
   finally:
     _metrics.record_workflow_invocation_duration(

src/google/adk/telemetry/tracing.py

@@ -32,6 +32,7 @@
 import logging
 import os
 import re
+from typing import Any
 from typing import Final
 from typing import TYPE_CHECKING
 
@@ -211,6 +212,21 @@ def resolve_error_type(error: BaseException) -> str:
   return type(error).__name__
 
 
+@contextmanager
+def start_as_current_span(name: str, **kwargs: Any) -> Iterator[Span]:
+  """Opens a span with ``tracer`` that names the error escaping it, if any.
+
+  OpenTelemetry marks such a span ERROR on its own, but only ``error.type``
+  says which error it was, and the metric for the same operation records it.
+  """
+  with tracer.start_as_current_span(name, **kwargs) as span:
+    try:
+      yield span
+    except Exception as e:
+      span.set_attribute(ERROR_TYPE, resolve_error_type(e))
+      raise
+
+
 def trace_agent_invocation(
     span: trace.Span, agent: BaseAgent, ctx: InvocationContext
 ) -> None:
@@ -1120,7 +1136,7 @@ def _use_native_generate_content_span_stable_semconv(
 ) -> Iterator[GenerateContentSpan]:
   telemetry_config = telemetry_config or TelemetryConfig()
   system_name = _resolve_gen_ai_system_name(llm_request.model)
-  with tracer.start_as_current_span(
+  with start_as_current_span(
       f"generate_content {llm_request.model or ''}".strip()
   ) as span:
     span.set_attribute(GEN_AI_SYSTEM, system_name)
@@ -1174,7 +1190,7 @@ def _use_native_generate_content_span(
       yield gc_span
     return
 
-  with tracer.start_as_current_span(
+  with start_as_current_span(
       f"generate_content {llm_request.model or ''}".strip()
   ) as span:
     _set_common_generate_content_attributes(

tests/unittests/telemetry/functional/scenarios/telemetry_setup.py

@@ -24,7 +24,6 @@
 from typing import NamedTuple
 
 from google.adk.telemetry import _metrics
-from google.adk.telemetry import node_tracing
 from google.adk.telemetry import tracing
 from opentelemetry.sdk._logs import LoggerProvider
 from opentelemetry.sdk._logs.export import InMemoryLogRecordExporter
@@ -235,13 +234,12 @@ def install_telemetry(
   tracer_provider.add_span_processor(SimpleSpanProcessor(span_exporter))
   real_tracer = tracer_provider.get_tracer(__name__)
 
-  for module in (tracing, node_tracing):
-    monkeypatch.setattr(
-        module.tracer,
-        "start_as_current_span",
-        real_tracer.start_as_current_span,
-    )
-    monkeypatch.setattr(module.tracer, "start_span", real_tracer.start_span)
+  monkeypatch.setattr(
+      tracing.tracer,
+      "start_as_current_span",
+      real_tracer.start_as_current_span,
+  )
+  monkeypatch.setattr(tracing.tracer, "start_span", real_tracer.start_span)
 
   logger_provider = LoggerProvider()
   logger_provider.add_log_record_processor(

tests/unittests/telemetry/functional_divergences.json

@@ -75,6 +75,16 @@
         "kind": "otel_bug",
         "reason": "The instrumentor reads `FunctionDeclaration.parameters`, which ADK leaves unset in favour of `parameters_json_schema`, so it advertises the tools by name and description but without their schema."
       },
+      {
+        "path": [
+          "attributes",
+          "error.type"
+        ],
+        "example_native_instrumentation_value": "429",
+        "example_otel_instrumentation_value": "ClientError",
+        "kind": "desired_behavior",
+        "reason": "ADK reports the provider's HTTP status code (`429`), the same value as on its duration metric; the instrumentor reports the exception class (`ClientError`), which `google.genai` collapses every 4xx into."
+      },
       {
         "path": [
           "attributes",
@@ -99,18 +109,6 @@
         "kind": "otel_bug",
         "reason": "On a call that failed the instrumentor records an empty finish-reason list for a response it never got."
       },
-      {
-        "path": [
-          "attributes",
-          "error.type"
-        ],
-        "example_native_instrumentation_value": {
-          "missing_on_this_side": true
-        },
-        "example_otel_instrumentation_value": "ClientError",
-        "kind": "adk_bug",
-        "reason": "ADK reports a failed call on the enclosing spans and on the duration metric, but leaves `error.type` off the inference span itself."
-      },
       {
         "path": [
           "attributes",

tests/unittests/telemetry/functional_goldens/agent/inference-error-resource-exhausted-schema-v1.json

@@ -1,7 +1,9 @@
 {
   "root_span": {
     "name": "invocation",
-    "attributes": {},
+    "attributes": {
+      "error.type": "429"
+    },
     "status": "ERROR",
     "children": [
       {
@@ -10,13 +12,16 @@
           "gen_ai.operation.name": "invoke_agent",
           "gen_ai.agent.description": "A sample root agent.",
           "gen_ai.agent.name": "some_root_agent",
-          "gen_ai.conversation.id": "PRESENT"
+          "gen_ai.conversation.id": "PRESENT",
+          "error.type": "429"
         },
         "status": "ERROR",
         "children": [
           {
             "name": "call_llm",
-            "attributes": {},
+            "attributes": {
+              "error.type": "429"
+            },
             "status": "ERROR",
             "children": [
               {
@@ -28,7 +33,8 @@
                   "gen_ai.agent.name": "some_root_agent",
                   "gen_ai.conversation.id": "PRESENT",
                   "gcp.vertex.agent.event_id": "PRESENT",
-                  "gcp.vertex.agent.invocation_id": "PRESENT"
+                  "gcp.vertex.agent.invocation_id": "PRESENT",
+                  "error.type": "429"
                 },
                 "status": "ERROR",
                 "children": [],

tests/unittests/telemetry/functional_goldens/agent/inference-error-resource-exhausted-schema-v2.json

@@ -4,7 +4,8 @@
     "attributes": {
       "gen_ai.operation.name": "invoke_workflow",
       "gen_ai.conversation.id": "PRESENT",
-      "gen_ai.workflow.name": "some_root_agent"
+      "gen_ai.workflow.name": "some_root_agent",
+      "error.type": "429"
     },
     "status": "ERROR",
     "children": [
@@ -14,13 +15,16 @@
           "gen_ai.operation.name": "invoke_agent",
           "gen_ai.agent.description": "A sample root agent.",
           "gen_ai.agent.name": "some_root_agent",
-          "gen_ai.conversation.id": "PRESENT"
+          "gen_ai.conversation.id": "PRESENT",
+          "error.type": "429"
         },
         "status": "ERROR",
         "children": [
           {
             "name": "call_llm",
-            "attributes": {},
+            "attributes": {
+              "error.type": "429"
+            },
             "status": "ERROR",
             "children": [
               {
@@ -32,7 +36,8 @@
                   "gen_ai.agent.name": "some_root_agent",
                   "gen_ai.conversation.id": "PRESENT",
                   "gcp.vertex.agent.event_id": "PRESENT",
-                  "gcp.vertex.agent.invocation_id": "PRESENT"
+                  "gcp.vertex.agent.invocation_id": "PRESENT",
+                  "error.type": "429"
                 },
                 "status": "ERROR",
                 "children": [],

tests/unittests/telemetry/functional_goldens/agent/inference-error-valueerror-schema-v2.json

@@ -4,7 +4,8 @@
     "attributes": {
       "gen_ai.operation.name": "invoke_workflow",
       "gen_ai.conversation.id": "PRESENT",
-      "gen_ai.workflow.name": "some_root_agent"
+      "gen_ai.workflow.name": "some_root_agent",
+      "error.type": "ValueError"
     },
     "status": "ERROR",
     "children": [
@@ -14,13 +15,16 @@
           "gen_ai.operation.name": "invoke_agent",
           "gen_ai.agent.description": "A sample root agent.",
           "gen_ai.agent.name": "some_root_agent",
-          "gen_ai.conversation.id": "PRESENT"
+          "gen_ai.conversation.id": "PRESENT",
+          "error.type": "ValueError"
         },
         "status": "ERROR",
         "children": [
           {
             "name": "call_llm",
-            "attributes": {},
+            "attributes": {
+              "error.type": "ValueError"
+            },
             "status": "ERROR",
             "children": [
               {
@@ -32,7 +36,8 @@
                   "gen_ai.agent.name": "some_root_agent",
                   "gen_ai.conversation.id": "PRESENT",
                   "gcp.vertex.agent.event_id": "PRESENT",
-                  "gcp.vertex.agent.invocation_id": "PRESENT"
+                  "gcp.vertex.agent.invocation_id": "PRESENT",
+                  "error.type": "ValueError"
                 },
                 "status": "ERROR",
                 "children": [],

tests/unittests/telemetry/functional_goldens/agent/tool-error-valueerror-schema-v2.json

@@ -4,7 +4,8 @@
     "attributes": {
       "gen_ai.operation.name": "invoke_workflow",
       "gen_ai.conversation.id": "PRESENT",
-      "gen_ai.workflow.name": "some_root_agent"
+      "gen_ai.workflow.name": "some_root_agent",
+      "error.type": "ValueError"
     },
     "status": "ERROR",
     "children": [
@@ -14,7 +15,8 @@
           "gen_ai.operation.name": "invoke_agent",
           "gen_ai.agent.description": "A sample root agent.",
           "gen_ai.agent.name": "some_root_agent",
-          "gen_ai.conversation.id": "PRESENT"
+          "gen_ai.conversation.id": "PRESENT",
+          "error.type": "ValueError"
         },
         "status": "ERROR",
         "children": [