Skip to content

Automate browser based workflows with AI

Remove the biased workflow_status log field in favour of workflow_run_status (#8355)

Aaron Perezd9dbcba2a769
diff --git a/skyvern/forge/sdk/workflow/service.py b/skyvern/forge/sdk/workflow/service.py
index 5820e57e8..277ec6996 100644
--- a/skyvern/forge/sdk/workflow/service.py
+++ b/skyvern/forge/sdk/workflow/service.py
@@ -7904,7 +7904,7 @@ class WorkflowService:
block_type=block.block_type,
block_label=block.label,
block_status=block_result.status,
- workflow_status=target_status,
+ resolved_terminal_status=target_status,
deferred=defer_status_write,
)
if defer_status_write:
@@ -9498,7 +9498,6 @@ class WorkflowService:
LOG.info(
f"Marking workflow run {workflow_run_id} as completed",
workflow_run_id=workflow_run_id,
- workflow_status="completed",
)
# Add workflow completion tag to trace
@@ -9695,7 +9694,6 @@ class WorkflowService:
LOG.info(
f"Marking workflow run {workflow_run_id} as failed",
workflow_run_id=workflow_run_id,
- workflow_status="failed",
failure_reason=failure_reason,
)
@@ -9713,7 +9711,6 @@ class WorkflowService:
LOG.info(
"Workflow run failure classified",
workflow_run_id=workflow_run_id,
- workflow_status="failed",
failure_category=failure_category,
primary_failure_category=failure_category[0].get("category") if failure_category else None,
failure_category_source=failure_category_source,
@@ -9767,7 +9764,6 @@ class WorkflowService:
LOG.info(
f"Marked workflow run {workflow_run_id} as failed (conditional)",
workflow_run_id=workflow_run_id,
- workflow_status="failed",
failure_category=failure_category,
)
if cascade_children:
@@ -9816,7 +9812,6 @@ class WorkflowService:
LOG.info(
f"Marked workflow run {workflow_run_id} as running",
workflow_run_id=workflow_run_id,
- workflow_status="running",
run_with=run_with,
queued_seconds=queued_seconds,
)
@@ -9832,7 +9827,6 @@ class WorkflowService:
LOG.info(
f"Marking workflow run {workflow_run_id} as terminated",
workflow_run_id=workflow_run_id,
- workflow_status="terminated",
failure_reason=failure_reason,
)
@@ -9852,7 +9846,6 @@ class WorkflowService:
LOG.info(
"Workflow run failure classified",
workflow_run_id=workflow_run_id,
- workflow_status="terminated",
failure_category=failure_category,
primary_failure_category=failure_category[0].get("category") if failure_category else None,
failure_category_source=failure_category_source,
@@ -9880,7 +9873,6 @@ class WorkflowService:
LOG.info(
f"Marking workflow run {workflow_run_id} as canceled",
workflow_run_id=workflow_run_id,
- workflow_status="canceled",
)
updated = await self.mark_workflow_run_as_canceled_if_not_final(
@@ -9920,7 +9912,6 @@ class WorkflowService:
LOG.info(
f"Marked workflow run {workflow_run_id} as canceled (conditional)",
workflow_run_id=workflow_run_id,
- workflow_status="canceled",
)
otel_trace.get_current_span().set_attribute("task.completion_status", WorkflowRunStatus.canceled)
@@ -9990,7 +9981,6 @@ class WorkflowService:
LOG.info(
f"Marking workflow run {workflow_run_id} as timed out",
workflow_run_id=workflow_run_id,
- workflow_status="timed_out",
)
failure_category = self._classify_workflow_terminal_failure(
@@ -10000,7 +9990,6 @@ class WorkflowService:
LOG.info(
"Workflow run failure classified",
workflow_run_id=workflow_run_id,
- workflow_status="timed_out",
failure_category=failure_category,
primary_failure_category=failure_category[0].get("category") if failure_category else None,
failure_category_source="code_level",
diff --git a/tests/unit/test_workflow_status_field_guard.py b/tests/unit/test_workflow_status_field_guard.py
new file mode 100644
index 000000000..d758d03f5
--- /dev/null
+++ b/tests/unit/test_workflow_status_field_guard.py
@@ -0,0 +1,71 @@
+"""AST guard against reintroducing the ``workflow_status`` log field.
+
+``workflow_status`` was a redundant duplicate of ``workflow_run_status``, emitted only by the
+``mark_workflow_run_as_*`` wrappers. The unified finalizer ``_finalize_workflow_run_status``
+bypasses those wrappers for completed/terminated runs while failures still route through
+``mark_workflow_run_as_failed``, so the field was populated on some terminal paths and not
+others -- a biased sample that reads as a ~99% failure rate (measured 7 completed vs 1,882
+failed over 3h, against a true 4,548 / 949).
+
+``workflow_run_status`` on the ``"Workflow run duration metrics"`` log is the canonical outcome
+facet: ``_after_workflow_run_status_write`` emits exactly one per final status write, and both
+status-write helpers call it.
+"""
+
+from __future__ import annotations
+
+import ast
+from pathlib import Path
+
+_REPO_ROOT = Path(__file__).resolve().parents[2]
+_SCANNED_ROOTS = ("skyvern", "cloud")
+_BANNED_KEYWORD = "workflow_status"
+
+
+def _log_call_keywords(tree: ast.AST) -> list[tuple[int, str]]:
+ """Keyword names passed to ``LOG.<method>(...)`` calls, with line numbers."""
+ found: list[tuple[int, str]] = []
+ for node in ast.walk(tree):
+ if not isinstance(node, ast.Call):
+ continue
+ func = node.func
+ if not isinstance(func, ast.Attribute):
+ continue
+ receiver = func.value
+ if not (isinstance(receiver, ast.Name) and receiver.id == "LOG"):
+ continue
+ for keyword in node.keywords:
+ if keyword.arg is not None:
+ found.append((node.lineno, keyword.arg))
+ return found
+
+
+def _python_files() -> list[Path]:
+ files: list[Path] = []
+ for root in _SCANNED_ROOTS:
+ files.extend((_REPO_ROOT / root).rglob("*.py"))
+ return files
+
+
+def test_no_production_log_emits_workflow_status() -> None:
+ offenders: list[str] = []
+ for path in _python_files():
+ try:
+ tree = ast.parse(path.read_text(encoding="utf-8"))
+ except (SyntaxError, UnicodeDecodeError): # pragma: no cover - not our source
+ continue
+ for lineno, keyword in _log_call_keywords(tree):
+ if keyword == _BANNED_KEYWORD:
+ offenders.append(f"{path.relative_to(_REPO_ROOT)}:{lineno}")
+
+ assert not offenders, (
+ "`workflow_status` is a biased half-sample of terminal outcomes -- the unified finalizer "
+ "bypasses the mark_* wrappers that emit it. Use `workflow_run_status` on the "
+ '"Workflow run duration metrics" log instead. Offenders: ' + ", ".join(sorted(offenders))
+ )
+
+
+def test_guard_detects_a_planted_violation() -> None:
+ """The guard must fail on a real violation, not merely pass on clean source."""
+ tree = ast.parse('LOG.info("x", workflow_run_id="wr_1", workflow_status="completed")\n')
+ assert (1, _BANNED_KEYWORD) in _log_call_keywords(tree)