[tools/perf] Mark test failures as unexpected

Failures during the processing of test results should be marked
unexpected.

Bug: 1019137
Change-Id: Ib3b880a462a038ea98fc8c56e3ec815ba613ec93
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1886894
Commit-Queue: Mikhail Khokhlov <khokhlov@google.com>
Reviewed-by: Juan Antonio Navarro Pérez <perezju@chromium.org>
Cr-Commit-Position: refs/heads/master@{#710330}

tools/perf/core/results_processor/compute_metrics.py

@@ -7,6 +7,8 @@ import logging
 import os
 import time
 
+from core.results_processor import util
+
 from tracing.metrics import metric_runner
 
 
@@ -46,7 +48,7 @@ def _RunMetric(test_result):
       test_result['testPath'], time.time() - start))
 
   if mre_result.failures:
-    test_result['status'] = 'FAIL'
+    util.SetUnexpectedFailure(test_result)
     for f in mre_result.failures:
       logging.error('Failure recorded for test %s: %s',
                     test_result['testPath'], f)
@@ -85,7 +87,7 @@ def ComputeTBMv2Metrics(test_result):
   # details.
   # TODO(crbug.com/1010041): Return a non-zero exit code in this case.
   if trace_size_in_mib > 400:
-    test_result['status'] = 'FAIL'
+    util.SetUnexpectedFailure(test_result)
     logging.error('%s: Trace size is too big: %s MiB',
                   test_result['testPath'], trace_size_in_mib)
     return

tools/perf/core/results_processor/compute_metrics_unittest.py

@@ -64,6 +64,7 @@ class ComputeMetricsTest(unittest.TestCase):
     histogram_dicts = test_result['_histograms'].AsDicts()
     self.assertEqual(histogram_dicts, [])
     self.assertEqual(test_result['status'], 'FAIL')
+    self.assertFalse(test_result['expected'])
 
   def testComputeTBMv2MetricsFailure(self):
     test_result = testing.TestResult(
@@ -87,6 +88,7 @@ class ComputeMetricsTest(unittest.TestCase):
     histogram_dicts = test_result['_histograms'].AsDicts()
     self.assertEqual(histogram_dicts, [])
     self.assertEqual(test_result['status'], 'FAIL')
+    self.assertFalse(test_result['expected'])
 
   def testComputeTBMv2MetricsSkipped(self):
     test_result = testing.TestResult(

tools/perf/core/results_processor/processor.py

@@ -77,7 +77,7 @@ def ProcessResults(options):
           result, upload_bucket, results_label, run_identifier,
           test_suite_start, should_compute_metrics, max_num_values),
       test_results,
-      on_failure=lambda result: result.update(status='FAIL'),
+      on_failure=util.SetUnexpectedFailure,
   )
 
   if should_compute_metrics:
@@ -110,7 +110,7 @@ def ProcessTestResult(test_result, upload_bucket, results_label,
     if max_num_values is not None and num_values > max_num_values:
       logging.error('%s produced %d values, but only %d are allowed.',
                     test_result['testPath'], num_values, max_num_values)
-      test_result['status'] = 'FAIL'
+      util.SetUnexpectedFailure(test_result)
       del test_result['_histograms']
     else:
       AddDiagnosticsToHistograms(test_result, test_suite_start, results_label)

tools/perf/core/results_processor/processor_test.py

@@ -162,6 +162,7 @@ class ResultsProcessorIntegrationTests(unittest.TestCase):
 
     self.assertEqual(results['tests']['benchmark']['story1']['actual'], 'PASS')
     self.assertEqual(results['tests']['benchmark']['story2']['actual'], 'FAIL')
+    self.assertTrue(results['tests']['benchmark']['story2']['is_unexpected'])
 
   def testHistogramsOutput(self):
     self.SerializeIntermediateResults(

tools/perf/core/results_processor/util.py

@@ -56,3 +56,9 @@ def IsoTimestampToEpoch(timestamp):
   except ValueError:
     dt = datetime.datetime.strptime(timestamp, '%Y-%m-%dT%H:%M:%SZ')
   return calendar.timegm(dt.timetuple()) + dt.microsecond / 1e6
+
+
+def SetUnexpectedFailure(test_result):
+  """Update fields of a test result in a case of processing failure."""
+  test_result['status'] = 'FAIL'
+  test_result['expected'] = False