[Monitoring] Partition UTask outcomes correctly into success and error (

#4516)

### Motivation

#4458 implemented an error rate for utasks, only considering exceptions.
In #4499 , outcomes were split between success, failure and maybe_retry
conditions. There we learned that the volume of retryable outcomes is
negligible, so it makes sense to count them as failures.

Listing out all the success conditions under _MetricRecorder is not
desirable. However, we are consciously taking this technical debt so we
can deliver #4271 .

A refactor of uworker main will be later performed, so we can split the
success and failure conditions, both of which are mixed in
uworker_output.ErrorType.

Reference for tech debt acknowledgement: #4517

src/clusterfuzz/_internal/bot/tasks/utasks/__init__.py

@@ -84,6 +84,17 @@ def __init__(self, subtask: _Subtask):
     self._subtask = subtask
     self._labels = None
     self.utask_main_failure = None
+    self._utask_success_conditions = [
+        None,  # This can be a successful return value in, ie, fuzz task
+        uworker_msg_pb2.ErrorType.NO_ERROR,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.ANALYZE_NO_CRASH,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.PROGRESSION_BAD_STATE_MIN_MAX,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.REGRESSION_NO_CRASH,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.REGRESSION_LOW_CONFIDENCE_IN_REGRESSION_RANGE,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.MINIMIZE_CRASH_TOO_FLAKY,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.LIBFUZZER_MINIMIZATION_UNREPRODUCIBLE,  # pylint: disable=no-member
+        uworker_msg_pb2.ErrorType.ANALYZE_CLOSE_INVALID_UPLOADED,  # pylint: disable=no-member
+    ]
 
     if subtask == _Subtask.PREPROCESS:
       self._preprocess_start_time_ns = self.start_time_ns
@@ -125,6 +136,12 @@ def set_task_details(self,
       # Ensure we always have a value after this method returns.
       assert self._preprocess_start_time_ns is not None
 
+  def _infer_uworker_main_outcome(self, exc_type, uworker_error) -> bool:
+    """Returns True if task succeeded, False otherwise."""
+    if exc_type or uworker_error not in self._utask_success_conditions:
+      return False
+    return True
+
   def __exit__(self, _exc_type, _exc_value, _traceback):
     # Ignore exception details, let Python continue unwinding the stack.
 
@@ -145,11 +162,12 @@ def __exit__(self, _exc_type, _exc_value, _traceback):
     # The only case where a task might fail without throwing, is in
     # utask_main, by returning an ErrorType proto which indicates
     # failure.
-    outcome = 'error' if _exc_type or self.utask_main_failure else 'success'
+    task_succeeded = self._infer_uworker_main_outcome(_exc_type,
+                                                      self.utask_main_failure)
     monitoring_metrics.TASK_OUTCOME_COUNT.increment({
-        **self._labels, 'outcome': outcome
+        **self._labels, 'task_succeeded': task_succeeded
     })
-    if outcome == "success":
+    if task_succeeded:
       error_condition = 'N/A'
     elif _exc_type:
       error_condition = 'UNHANDLED_EXCEPTION'
@@ -161,16 +179,11 @@ def __exit__(self, _exc_type, _exc_value, _traceback):
     # labels limit recommended by gcp.
     trimmed_labels = self._labels
     del trimmed_labels['job']
-    trimmed_labels['outcome'] = outcome
+    trimmed_labels['task_succeeded'] = task_succeeded
     trimmed_labels['error_condition'] = error_condition
     monitoring_metrics.TASK_OUTCOME_COUNT_BY_ERROR_TYPE.increment(
         trimmed_labels)
 
-    if error_condition != 'UNHANDLED_EXCEPTION':
-      task = self._labels['task']
-      subtask = self._labels['subtask']
-      logs.info(f'Task {task}, at subtask {subtask}, finished successfully.')
-
 
 def ensure_uworker_env_type_safety(uworker_env):
   """Converts all values in |uworker_env| to str types.

src/clusterfuzz/_internal/metrics/monitoring_metrics.py

@@ -263,7 +263,7 @@
         monitor.StringField('subtask'),
         monitor.StringField('mode'),
         monitor.StringField('platform'),
-        monitor.StringField('outcome'),
+        monitor.BooleanField('task_succeeded'),
     ])
 
 TASK_OUTCOME_COUNT_BY_ERROR_TYPE = monitor.CounterMetric(
@@ -274,7 +274,7 @@
         monitor.StringField('subtask'),
         monitor.StringField('mode'),
         monitor.StringField('platform'),
-        monitor.StringField('outcome'),
+        monitor.BooleanField('task_succeeded'),
         monitor.StringField('error_condition'),
     ])