Reduce bug filing threshold for libfuzzer-based Android crashes

Updated triage.py to file bugs for fuzzer crashes by lowering the
threshold limit. These fuzzers run on android platform and on
libfuzzer engine.

src/clusterfuzz/_internal/cron/triage.py

@@ -98,6 +98,60 @@ def _is_bug_filed(testcase):
   return False
 
 
+def _is_blocking_progress(testcase):
+  """Checks the crash frequency if it is reported on libfuzzer"""
+  if testcase.job_type.startswith('libfuzzer'):
+    # Get crash statistics data on this unreproducible crash for last X days.
+    last_hour = crash_stats.get_last_successful_hour()
+    if not last_hour:
+      # No crash stats available, skip.
+      return False
+
+    _, rows = crash_stats.get(
+        end=last_hour,
+        block='day',
+        days=data_types.FILE_CONSISTENT_UNREPRODUCIBLE_TESTCASE_DEADLINE,
+        group_by='reproducible_flag',
+        where_clause=(
+            'crash_type = %s AND crash_state = %s AND security_flag = %s' %
+            (json.dumps(testcase.crash_type), json.dumps(testcase.crash_state),
+             json.dumps(testcase.security_flag))),
+        group_having_clause='',
+        sort_by='total_count',
+        offset=0,
+        limit=1)
+
+    # Calculate total crash count and crash days count.
+    crash_days_indices = set()
+    total_crash_count = 0
+    for row in rows:
+      if 'groups' not in row:
+        continue
+
+      total_crash_count += row['totalCount']
+      for group in row['groups']:
+        for index in group['indices']:
+          crash_days_indices.add(index['hour'])
+
+    crash_days_count = len(crash_days_indices)
+    # Considers an unreproducible testcase as important if the crash
+    # occurred atleast once everyday for the last 14 days and total
+    # crash count exceeded 14.
+    return (crash_days_count ==
+            data_types.FILE_CONSISTENT_UNREPRODUCIBLE_TESTCASE_DEADLINE and
+            total_crash_count >=
+            data_types.FILE_UNREPRODUCIBLE_TESTCASE_MIN_STARTUP_CRASH_THRESHOLD)
+
+  return False
+
+
+def is_crash_important_android(testcase):
+  """"Indicate if the android crash is important to file."""
+    if _is_blocking_progress():
+      return True
+    return False
+
+
 def _is_crash_important(testcase):
   """Indicate if the crash is important to file."""
   if not testcase.one_time_crasher_flag:
@@ -388,9 +442,12 @@ def main():
     # Check if the crash is important, i.e. it is either a reproducible crash
     # or an unreproducible crash happening frequently.
     if not _is_crash_important(testcase):
-      logs.info(
-          f'Skipping testcase {testcase_id}, since the crash is not important.')
-      continue
+      # Check if the crash is a startup crash, i.e. it is causing the fuzzer
+      # to crash on startup and not allowing the fuzzer to run longer
+      if if testcase.platform == "android" and not is_crash_important_android(testcase):
+        logs.info(
+            f'Skipping testcase {testcase_id}, since the crash is not important.')
+        continue
 
     # Require that all tasks like minimizaton, regression testing, etc have
     # finished.

src/clusterfuzz/_internal/datastore/data_types.py

@@ -58,6 +58,9 @@
 # Minimum number of unreproducible crashes to see before filing it.
 FILE_UNREPRODUCIBLE_TESTCASE_MIN_CRASH_THRESHOLD = 100
 
+# Minimum number of unreproducible crashes to see before filing it for android.
+FILE_UNREPRODUCIBLE_TESTCASE_MIN_STARTUP_CRASH_THRESHOLD = 14
+
 # Heartbeat wait interval.
 HEARTBEAT_WAIT_INTERVAL = 10 * 60
 

src/clusterfuzz/_internal/tests/appengine/handlers/cron/triage_test.py

@@ -25,6 +25,67 @@
 from clusterfuzz._internal.tests.test_libs import test_utils
 
 
+@test_utils.with_cloud_emulators('datastore')
+class AndroidCrashImportantTest(unittest.TestCase):
+  """Tests for is_crash_important_android."""
+
+  def setUp(self):
+    helpers.patch(self, [
+        'clusterfuzz._internal.metrics.crash_stats.get_last_successful_hour',
+        'clusterfuzz._internal.metrics.crash_stats.get',
+        'clusterfuzz._internal.base.utils.utcnow',
+    ])
+    self.mock.utcnow.return_value = test_utils.CURRENT_TIME
+
+  def test_is_crash_important_android_1(self):
+    """If this unreproducible testcase (libfuzzer) is crashing frequently,
+    then it is an important crash."""
+    self.mock.get_last_successful_hour.return_value = 417325
+    indices = [{
+        'count': 1,
+        'hour': day_index
+    } for day_index in range(417325, 416989, -24)]
+
+    self.mock.get.return_value = (1, [{
+        'totalCount': 14,
+        'groups': [{
+            'indices': indices,
+            'name': 'false',
+        },]
+    }])
+    testcase = test_utils.create_generic_testcase()
+    testcase.job_type = 'libfuzzer_test'
+    testcase.platform = 'android'
+    testcase.one_time_crasher_flag = True
+    testcase.put()
+
+    self.assertTrue(triage.is_crash_important_android(testcase))
+
+  def test_is_crash_important_android_2(self):
+    """If this unreproducible testcase (libfuzzer) is less than the
+    total crash threshold, then it is not important."""
+    self.mock.get_last_successful_hour.return_value = 417325
+    indices = [{
+        'count': day_index % 5 == 0,
+        'hour': day_index
+    } for day_index in range(417325, 416989, -24)]
+
+    self.mock.get.return_value = (1, [{
+        'totalCount': 3,
+        'groups': [{
+            'indices': indices,
+            'name': 'false',
+        },]
+    }])
+    testcase = test_utils.create_generic_testcase()
+    testcase.job_type = 'libfuzzer_test'
+    testcase.platform = 'android'
+    testcase.one_time_crasher_flag = True
+    testcase.put()
+
+    self.assertFalse(triage.is_crash_important_android(testcase))
+
+
 @test_utils.with_cloud_emulators('datastore')
 class CrashImportantTest(unittest.TestCase):
   """Tests for _is_crash_important."""