Differentiate between timeouts and crashes

3 files changed, 23 insertions, 14 deletions

diff --git a/tools/profiling/microbenchmarks/bm_diff/bm_diff.py b/tools/profiling/microbenchmarks/bm_diff/bm_diff.py
index b049f41ca0..72a8d11eea 100755
--- a/tools/profiling/microbenchmarks/bm_diff/bm_diff.py
+++ b/tools/profiling/microbenchmarks/bm_diff/bm_diff.py
@@ -51,7 +51,7 @@ def _median(ary):
     ary = sorted(ary)
     n = len(ary)
     if n % 2 == 0:
-        return (ary[n / 2] + ary[n / 2 + 1]) / 2.0
+        return (ary[(n - 1) / 2] + ary[(n - 1) / 2 + 1]) / 2.0
     else:
         return ary[n / 2]
 
@@ -130,23 +130,30 @@ class Benchmark:
         return [self.final[f] if f in self.final else '' for f in flds]
 
 
-def _read_json(filename, badfiles):
+def _read_json(filename, badjson_files, nonexistant_files):
     stripped = ".".join(filename.split(".")[:-2])
     try:
         with open(filename) as f:
             return json.loads(f.read())
+    except IOError, e:
+        if stripped in nonexistant_files:
+            nonexistant_files[stripped] += 1
+        else:
+            nonexistant_files[stripped] = 1
+        return None        
     except ValueError, e:
-        if stripped in badfiles:
-            badfiles[stripped] += 1
+        if stripped in badjson_files:
+            badjson_files[stripped] += 1
         else:
-            badfiles[stripped] = 1
+            badjson_files[stripped] = 1
         return None
 
 
 def diff(bms, loops, track, old, new):
     benchmarks = collections.defaultdict(Benchmark)
 
-    badfiles = {}
+    badjson_files = {}
+    nonexistant_files = {}
     for bm in bms:
         for loop in range(0, loops):
             for line in subprocess.check_output(
@@ -156,16 +163,16 @@ def diff(bms, loops, track, old, new):
                     "<", "_").replace(">", "_").replace(", ", "_")
                 js_new_ctr = _read_json('%s.%s.counters.%s.%d.json' %
                                         (bm, stripped_line, new, loop),
-                                        badfiles)
+                                        badjson_files, nonexistant_files)
                 js_new_opt = _read_json('%s.%s.opt.%s.%d.json' %
                                         (bm, stripped_line, new, loop),
-                                        badfiles)
+                                        badjson_files, nonexistant_files)
                 js_old_ctr = _read_json('%s.%s.counters.%s.%d.json' %
                                         (bm, stripped_line, old, loop),
-                                        badfiles)
+                                        badjson_files, nonexistant_files)
                 js_old_opt = _read_json('%s.%s.opt.%s.%d.json' %
                                         (bm, stripped_line, old, loop),
-                                        badfiles)
+                                        badjson_files, nonexistant_files)
 
                 if js_new_ctr:
                     for row in bm_json.expand_json(js_new_ctr, js_new_opt):
@@ -191,7 +198,8 @@ def diff(bms, loops, track, old, new):
     for name in sorted(benchmarks.keys()):
         if benchmarks[name].skip(): continue
         rows.append([name] + benchmarks[name].row(fields))
-    note = 'flakiness data = %s' % str(badfiles)
+    note = 'Corrupt JSON data (indicates timeout or crash) = %s' % str(badjson_files)
+    note += '\n\nMissing files (new benchmark) = %s' % str(nonexistant_files)
     if rows:
         return tabulate.tabulate(rows, headers=headers, floatfmt='+.2f'), note
     else:
@@ -204,4 +212,4 @@ if __name__ == '__main__':
                       args.new)
     print note
     print ""
-    print diff
+    print diff if diff else "No performance differences"
diff --git a/tools/profiling/microbenchmarks/bm_diff/bm_main.py b/tools/profiling/microbenchmarks/bm_diff/bm_main.py
index 4c6eb8b48c..47381f4ec8 100755
--- a/tools/profiling/microbenchmarks/bm_diff/bm_main.py
+++ b/tools/profiling/microbenchmarks/bm_diff/bm_main.py
@@ -139,6 +139,8 @@ def main(args):
         text = 'Performance differences noted:\n' + diff
     else:
         text = 'No significant performance differences'
+    print note
+    print ""
     print text
     comment_on_pr.comment_on_pr('```\n%s\n\n%s\n```' % (note, text))
 
diff --git a/tools/profiling/microbenchmarks/bm_diff/bm_run.py b/tools/profiling/microbenchmarks/bm_diff/bm_run.py
index e281e9e61c..6ad9f1a3b7 100755
--- a/tools/profiling/microbenchmarks/bm_diff/bm_run.py
+++ b/tools/profiling/microbenchmarks/bm_diff/bm_run.py
@@ -107,8 +107,7 @@ def _collect_bm_data(bm, cfg, name, reps, idx, loops):
                 shortname='%s %s %s %s %d/%d' % (bm, line, cfg, name, idx + 1,
                                                  loops),
                 verbose_success=True,
-                timeout_seconds=60 * 10,
-                timeout_retries=3))
+                timeout_seconds=60 * 2))
     return jobs_list