fix(tracing): capture span body exceptions and export SGP status=ERROR

Spans exported to SGP always showed status=SUCCESS even when the operation
they represent failed. The SGP span defaults status to "SUCCESS" and only
flips to "ERROR" inside its own __exit__ context manager, but the agentex
processor builds SGP spans via create_span(...) and flushes them directly,
so __exit__ never runs. The Trace.span()/AsyncTrace.span() context managers
also ended spans in a bare finally, so a body exception was never recorded.

Capture the exception in both context managers and carry it on the span so
the SGP processor can map it:

- add span_error.py with set_span_error/get_span_error, storing the failure
  under the reserved span.data["__error__"] key (the Span model is generated
  from the OpenAPI spec and has no status/error field; data is a real field
  that survives model_copy(deep=True) and round-trips to both stores)
- Trace.span()/AsyncTrace.span(): except -> set_span_error(span, exc); raise
- _build_sgp_span(): when an error is present, set sgp_span.status = "ERROR"
  plus error/error.type/error.message metadata (matching SGP's native
  __exit__ shape)

Exceptions still propagate; asyncio.CancelledError/KeyboardInterrupt are not
flagged (control flow, not failures).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: NiteshDhanpal

src/agentex/lib/core/tracing/processors/sgp_tracing_processor.py

@@ -15,6 +15,7 @@
 from agentex.lib.utils.logging import make_logger
 from agentex.lib.core.observability import tracing_metrics_recording as _metrics
 from agentex.lib.environment_variables import EnvironmentVariables
+from agentex.lib.core.tracing.span_error import get_span_error
 from agentex.lib.core.tracing.processors.tracing_processor_interface import (
     SyncTracingProcessor,
     AsyncTracingProcessor,
@@ -83,6 +84,12 @@ def _build_sgp_span(span: Span, env_vars: EnvironmentVariables) -> SGPSpan:
         ),
     )
     sgp_span.start_time = span.start_time.isoformat()  # type: ignore[union-attr]
+    error = get_span_error(span)
+    if error is not None:
+        sgp_span.status = "ERROR"
+        sgp_span.metadata["error"] = True
+        sgp_span.metadata["error.type"] = error["type"]
+        sgp_span.metadata["error.message"] = error["message"]
     return sgp_span
 
 

src/agentex/lib/core/tracing/span_error.py

@@ -0,0 +1,36 @@
+from __future__ import annotations
+
+from typing import Any
+
+from agentex.types.span import Span
+
+# Reserved key under ``Span.data`` carrying failure info for a span whose
+# context-manager body raised. Mirrors the existing ``__span_type__`` /
+# ``__source__`` reserved-key convention already read/written by the SGP
+# processor. Stored in ``data`` because the Span model is generated from the
+# OpenAPI spec and has no first-class status/error field; ``data`` is a real
+# field, so it survives ``model_copy(deep=True)`` and round-trips to both the
+# SGP and agentex-native span stores.
+SPAN_ERROR_KEY = "__error__"
+
+
+def set_span_error(span: Span, exc: BaseException) -> None:
+    """Record an exception on ``span`` under ``data[SPAN_ERROR_KEY]``.
+
+    No-op when ``span.data`` is a list (matching ``_add_source_to_span``, which
+    only attaches metadata to dict-shaped data).
+    """
+    error = {"type": type(exc).__name__, "message": str(exc)}
+    if span.data is None:
+        span.data = {}
+    if isinstance(span.data, dict):
+        span.data[SPAN_ERROR_KEY] = error
+
+
+def get_span_error(span: Span) -> dict[str, Any] | None:
+    """Return the error recorded by :func:`set_span_error`, or ``None``."""
+    if isinstance(span.data, dict):
+        value = span.data.get(SPAN_ERROR_KEY)
+        if isinstance(value, dict):
+            return value
+    return None

src/agentex/lib/core/tracing/trace.py

@@ -11,6 +11,7 @@
 from agentex.types.span import Span
 from agentex.lib.utils.logging import make_logger
 from agentex.lib.utils.model_utils import recursive_model_dump
+from agentex.lib.core.tracing.span_error import set_span_error
 from agentex.lib.core.tracing.span_queue import (
     SpanEventType,
     AsyncSpanQueue,
@@ -165,6 +166,9 @@ def span(
         span = self.start_span(name, parent_id, input, data, task_id=task_id)
         try:
             yield span
+        except Exception as exc:
+            set_span_error(span, exc)
+            raise
         finally:
             self.end_span(span)
 
@@ -321,5 +325,8 @@ async def span(
         span = await self.start_span(name, parent_id, input, data, task_id=task_id)
         try:
             yield span
+        except Exception as exc:
+            set_span_error(span, exc)
+            raise
         finally:
             await self.end_span(span)

tests/lib/core/tracing/test_span_error.py

@@ -0,0 +1,135 @@
+from __future__ import annotations
+
+import uuid
+from datetime import UTC, datetime
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from agentex.types.span import Span
+from agentex.lib.core.tracing.trace import Trace, AsyncTrace
+from agentex.lib.core.tracing.span_error import (
+    SPAN_ERROR_KEY,
+    get_span_error,
+    set_span_error,
+)
+
+PROCESSOR_MODULE = "agentex.lib.core.tracing.processors.sgp_tracing_processor"
+
+
+def _make_span(data=None) -> Span:
+    return Span(
+        id=str(uuid.uuid4()),
+        name="test-span",
+        start_time=datetime.now(UTC),
+        trace_id="trace-1",
+        data=data,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Helpers: set_span_error / get_span_error
+# ---------------------------------------------------------------------------
+
+
+class TestSpanErrorHelpers:
+    def test_set_then_get_on_none_data(self):
+        span = _make_span(data=None)
+        set_span_error(span, ValueError("boom"))
+        assert get_span_error(span) == {"type": "ValueError", "message": "boom"}
+        assert span.data[SPAN_ERROR_KEY] == {"type": "ValueError", "message": "boom"}
+
+    def test_set_preserves_existing_dict_keys(self):
+        span = _make_span(data={"__span_type__": "LLM"})
+        set_span_error(span, RuntimeError("nope"))
+        assert span.data["__span_type__"] == "LLM"
+        assert get_span_error(span)["type"] == "RuntimeError"
+
+    def test_get_returns_none_when_no_error(self):
+        assert get_span_error(_make_span(data={"foo": "bar"})) is None
+        assert get_span_error(_make_span(data=None)) is None
+
+    def test_set_is_noop_on_list_data(self):
+        span = _make_span(data=[{"a": 1}])
+        set_span_error(span, ValueError("boom"))
+        # list-shaped data is left untouched (mirrors _add_source_to_span)
+        assert span.data == [{"a": 1}]
+        assert get_span_error(span) is None
+
+
+# ---------------------------------------------------------------------------
+# Capture: the context managers record body exceptions onto the span
+# ---------------------------------------------------------------------------
+
+
+class TestContextManagerCapture:
+    def test_sync_span_records_error_and_reraises(self):
+        trace = Trace(processors=[], client=MagicMock(), trace_id="t1")
+        captured = {}
+        with pytest.raises(ValueError, match="boom"):
+            with trace.span("op") as span:
+                captured["span"] = span
+                raise ValueError("boom")
+        err = get_span_error(captured["span"])
+        assert err == {"type": "ValueError", "message": "boom"}
+
+    def test_sync_span_success_has_no_error(self):
+        trace = Trace(processors=[], client=MagicMock(), trace_id="t1")
+        with trace.span("op") as span:
+            pass
+        assert get_span_error(span) is None
+
+    @pytest.mark.asyncio
+    async def test_async_span_records_error_and_reraises(self):
+        trace = AsyncTrace(processors=[], client=MagicMock(), trace_id="t1")
+        captured = {}
+        with pytest.raises(RuntimeError, match="kaboom"):
+            async with trace.span("op") as span:
+                captured["span"] = span
+                raise RuntimeError("kaboom")
+        err = get_span_error(captured["span"])
+        assert err == {"type": "RuntimeError", "message": "kaboom"}
+
+
+# ---------------------------------------------------------------------------
+# Map: _build_sgp_span translates the recorded error into SGP status=ERROR
+# ---------------------------------------------------------------------------
+
+
+class _FakeSGPSpan:
+    def __init__(self, metadata):
+        self.status = "SUCCESS"
+        self.metadata = metadata if metadata is not None else {}
+        self.start_time = None
+
+
+def _fake_create_span(**kwargs):
+    return _FakeSGPSpan(kwargs.get("metadata"))
+
+
+class TestBuildSGPSpanMapping:
+    @staticmethod
+    def _env():
+        return MagicMock(ACP_TYPE=None, AGENT_NAME=None, AGENT_ID=None)
+
+    def test_error_maps_to_status_error(self):
+        from agentex.lib.core.tracing.processors.sgp_tracing_processor import _build_sgp_span
+
+        span = _make_span(data={SPAN_ERROR_KEY: {"type": "ValueError", "message": "boom"}})
+        with patch(f"{PROCESSOR_MODULE}.create_span", side_effect=_fake_create_span):
+            sgp_span = _build_sgp_span(span, self._env())
+
+        assert sgp_span.status == "ERROR"
+        assert sgp_span.metadata["error"] is True
+        assert sgp_span.metadata["error.type"] == "ValueError"
+        assert sgp_span.metadata["error.message"] == "boom"
+
+    def test_no_error_leaves_status_success(self):
+        from agentex.lib.core.tracing.processors.sgp_tracing_processor import _build_sgp_span
+
+        span = _make_span(data={"__span_type__": "LLM"})
+        with patch(f"{PROCESSOR_MODULE}.create_span", side_effect=_fake_create_span):
+            sgp_span = _build_sgp_span(span, self._env())
+
+        assert sgp_span.status == "SUCCESS"
+        assert "error" not in sgp_span.metadata