[Monitoring] Remove granularity for stuck testcases metric (#4496)

The metric for untriaged testcae age was not considering bugs that were
being filed legitimately, so there was no metric emission at all.

Also, removes granularity in the stuck testcase count metric.

Author: vitorguidi

src/clusterfuzz/_internal/cron/triage.py

@@ -309,22 +309,11 @@ def _file_issue(testcase, issue_tracker, throttler):
   return filed
 
 
-untriaged_testcase_count = {}
-
-
-def _increment_untriaged_testcase_count(testcase: data_types.Testcase):
-  identifier = (testcase.job_type, testcase.platform)
-  if identifier not in untriaged_testcase_count:
-    untriaged_testcase_count[identifier] = 0
-  untriaged_testcase_count[identifier] += 1
-
-
 def _emit_untriaged_testcase_age_metric(testcase: data_types.Testcase):
   """Emmits a metric to track age of untriaged testcases."""
   if not testcase.timestamp:
     return
 
-  _increment_untriaged_testcase_count(testcase)
   logs.info(f'Emiting UNTRIAGED_TESTCASE_AGE for testcase {testcase.key.id()} '
             f'(age = {testcase.get_age_in_seconds()})')
   monitoring_metrics.UNTRIAGED_TESTCASE_AGE.add(
@@ -335,16 +324,6 @@ def _emit_untriaged_testcase_age_metric(testcase: data_types.Testcase):
       })
 
 
-def _emit_untriaged_testcase_count_metric():
-  for (job, platform) in untriaged_testcase_count:
-    monitoring_metrics.UNTRIAGED_TESTCASE_COUNT.set(
-        untriaged_testcase_count[(job, platform)],
-        labels={
-            'job': job,
-            'platform': platform,
-        })
-
-
 def main():
   """Files bugs."""
   try:
@@ -367,6 +346,8 @@ def main():
 
   throttler = Throttler()
 
+  untriaged_testcases = 0
+
   for testcase_id in data_handler.get_open_testcase_id_iterator():
     logs.info(f'Triaging {testcase_id}')
     try:
@@ -395,6 +376,7 @@ def main():
     if testcase.get_metadata('progression_pending'):
       logs.info(f'Skipping testcase {testcase_id}, progression pending')
       _emit_untriaged_testcase_age_metric(testcase)
+      untriaged_testcases += 1
       continue
 
     # If the testcase has a bug filed already, no triage is needed.
@@ -414,6 +396,7 @@ def main():
     # finished.
     if not critical_tasks_completed:
       _emit_untriaged_testcase_age_metric(testcase)
+      untriaged_testcases += 1
       logs.info(
           f'Skipping testcase {testcase_id}, critical tasks still pending.')
       continue
@@ -431,12 +414,14 @@ def main():
     if not testcase.group_id and not dates.time_has_expired(
         testcase.timestamp, hours=data_types.MIN_ELAPSED_TIME_SINCE_REPORT):
       _emit_untriaged_testcase_age_metric(testcase)
+      untriaged_testcases += 1
       logs.info(f'Skipping testcase {testcase_id}, pending grouping.')
       continue
 
     if not testcase.get_metadata('ran_grouper'):
       # Testcase should be considered by the grouper first before filing.
       _emit_untriaged_testcase_age_metric(testcase)
+      untriaged_testcases += 1
       logs.info(f'Skipping testcase {testcase_id}, pending grouping.')
       continue
 
@@ -463,10 +448,13 @@ def main():
     # Clean up old triage messages that would be not applicable now.
     testcase.delete_metadata(TRIAGE_MESSAGE_KEY, update_testcase=False)
 
+    # A testcase is untriaged, until immediately before a bug is opened
+    _emit_untriaged_testcase_age_metric(testcase)
+    untriaged_testcases += 1
+
     # File the bug first and then create filed bug metadata.
     if not _file_issue(testcase, issue_tracker, throttler):
       logs.info(f'Issue filing failed for testcase id {testcase_id}')
-      _emit_untriaged_testcase_age_metric(testcase)
       continue
 
     _create_filed_bug_metadata(testcase)
@@ -475,7 +463,8 @@ def main():
     logs.info('Filed new issue %s for testcase %d.' % (testcase.bug_information,
                                                        testcase_id))
 
-  _emit_untriaged_testcase_count_metric()
+    monitoring_metrics.UNTRIAGED_TESTCASE_COUNT.set(
+        untriaged_testcases, labels={})
 
   logs.info('Triage testcases succeeded.')
   return True

src/clusterfuzz/_internal/metrics/monitoring_metrics.py

@@ -370,10 +370,8 @@
     description='Number of testcases that were not yet triaged '
     '(have not yet completed analyze, regression,'
     ' minimization, impact task), in hours.',
-    field_spec=[
-        monitor.StringField('job'),
-        monitor.StringField('platform'),
-    ])
+    field_spec=[],
+)
 
 ANALYZE_TASK_REPRODUCIBILITY = monitor.CounterMetric(
     'task/analyze/reproducibility',