Script to generate buildbot json for both the perf waterfall and fyi

tools/perf/generate_perf_json.py

+#!/usr/bin/env python
+# Copyright 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""Script to generate chromium.perf.json and chromium.perf.fyi.json in
+the src/testing/buildbot directory. Maintaining these files by hand is
+too unwieldy.
+"""
+
+import json
+import os
+import sys
+
+from chrome_telemetry_build import chromium_config
+
+sys.path.append(chromium_config.GetTelemetryDir())
+from telemetry import benchmark as benchmark_module
+from telemetry.core import discover
+

[dtu, 2016/08/25 17:39:05]

style guide: two spaces between all top level blocks

[eyaich1, 2016/08/29 18:38:09]

Done.

+SCRIPT_TESTS = [
+  {
+    'args': [
+      'gpu_perftests'
+    ],
+    'name': 'gpu_perftests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Android Galaxy S5 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Android Nexus5 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Android Nexus7v2 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Android Nexus9 Perf',
+        'shards': [2]
+      },
+    ]
+  },
+  {
+    'args': [
+      'cc_perftests'
+    ],
+    'name': 'cc_perftests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Android Galaxy S5 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Android Nexus5 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Android Nexus6 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Android Nexus7v2 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Android Nexus9 Perf',
+        'shards': [2]
+      },
+    ]
+  },
+  {
+    'args': [
+      'cc_perftests',
+      '--test-launcher-print-test-stdio=always'
+    ],
+    'name': 'cc_perftests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Linux Perf',
+        'shards': [3]
+      },
+    ]
+  },
+  {
+    'args': [
+      'load_library_perf_tests',
+      '--test-launcher-print-test-stdio=always'
+    ],
+    'name': 'load_library_perf_tests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Linux Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Win 7 ATI GPU Perf',
+        'shards': [2]

[Ken Russell (switch to Gerrit), 2016/08/25 01:26:17]

Out of curiosity, why is only shard 2 for this bot covered by
chromium.perf.json? Where do the other 4 shards of this bot pick up their test
descriptions? Same basic question for some of the others.

[eyaich1, 2016/08/25 13:09:46]

Like Dave noted I think this is somewhat random and should be standardized.  For
this CL I am just trying to generate what is there, I think the standardization
should be done in a follow on CL.

+      },
+      {
+        'name': 'Win 7 Intel GPU Perf',
+        'shards': [1, 2, 3, 4, 5]
+      },
+      {
+        'name': 'Win 7 Nvidia GPU Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 7 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Win 7 x64 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 8 Perf',
+        'shards': [2]
+      },
+    ]
+  },
+  {
+    'args': [
+      'performance_browser_tests',
+      '--test-launcher-print-test-stdio=always',
+      '--gtest_filter=TabCapturePerformanceTest.*:CastV2PerformanceTest.*',
+      '--test-launcher-jobs=1',
+      '--enable-gpu'
+    ],
+    'name': 'performance_browser_tests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Mac 10.8 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Mac 10.9 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Win 7 ATI GPU Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 7 Intel GPU Perf',
+        'shards': [1, 2, 3, 4, 5]
+      },
+      {
+        'name': 'Win 7 Nvidia GPU Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 7 Perf',
+        'shards': [3]
+      },
+      {
+        'name': 'Win 7 x64 Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 8 Perf',
+        'shards': [2]
+      },
+    ]
+  },
+  {
+    'args': [
+      'angle_perftests',
+      '--test-launcher-print-test-stdio=always',
+      '--test-launcher-jobs=1'
+    ],
+    'name': 'angle_perftests',
+    'script': 'gtest_perf_test.py',
+    'testers': [
+      {
+        'name': 'Win 7 ATI GPU Perf',
+        'shards': [2]
+      },
+      {
+        'name': 'Win 7 Intel GPU Perf',
+        'shards': [1, 2, 3, 4, 5]
+      },
+      {
+        'name': 'Win 7 Nvidia GPU Perf',
+        'shards': [2]
+      },
+    ]
+  },
+]
+
+def AddTester(waterfall, name, perf_id, platform, target_bits=64,
+  num_host_shards=1, num_device_shards=1, swarming=None):

[dtu, 2016/08/25 17:39:05]

style guide: align indent with parameters on the first line

[eyaich1, 2016/08/29 18:38:09]

Done.

+  del perf_id # this will be needed
+  waterfall['testers'][name] = {
+   'platform': platform,

[dtu, 2016/08/25 17:39:05]

style guide: 4-space indent

[eyaich1, 2016/08/29 18:38:09]

Done.

+   'num_device_shards': num_device_shards,
+   'num_host_shards': num_host_shards,
+   'target_bits': target_bits
+  }
+  if swarming is not None:

[dtu, 2016/08/25 17:39:05]

style guide: implicit false. if swarming:

[eyaich1, 2016/08/29 18:38:09]

Done.

+    waterfall['testers'][name]['swarming_dimensions'] = {
+      'gpu': swarming['gpu'],
+      'os': swarming['os']
+    }
+    waterfall['testers'][name]['swarming'] = True
+
+  return waterfall
+
+def generateFyiWaterfall():

[dtu, 2016/08/25 17:39:05]

style guide: lower_with_under() or CapWords()

[eyaich1, 2016/08/29 18:38:10]

Done.

+  waterfall = {'builders':[], 'testers': {}}
+  waterfall = AddTester(waterfall, 'Win 10 Low-End 2 Core Perf',
+    'win-low-end-2-core', 'win',

[dtu, 2016/08/25 17:39:05]

style guide: align indent with parameters on first line

[eyaich1, 2016/08/29 18:38:10]

Done.

+    swarming={'gpu': '8086:22b1', 'os': 'Windows-10-10240'})
+  waterfall = AddTester(waterfall, 'Win 10 Low-End 4 Core Perf',
+    'win-low-end-4-core', 'win',
+    swarming={'gpu': '1002:9830', 'os': 'Windows-10-10586'})
+  return waterfall
+
+def generateWaterfall():
+  waterfall = {'builders':[], 'testers': {}}
+  # These configurations are taken from chromium_perf.py in
+  # build/scripts/slave/recipe_modules/chromium_tests and must be kept in sync
+  # to generate the correct json for each tester
+  waterfall = AddTester(waterfall, 'Android Galaxy S5 Perf',
+    'android-galaxy-s5', 'android', target_bits=32,
+    num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android Nexus5 Perf', 'android-nexus5',
+    'android', target_bits=32, num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android Nexus5X Perf', 'android-nexus5X',
+    'android', target_bits=32, num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android Nexus6 Perf', 'android-nexus6',
+    'android', target_bits=32, num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android Nexus7v2 Perf', 'android-nexus7v2',
+   'android', target_bits=32, num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android Nexus9 Perf', 'android-nexus9',
+    'android', num_device_shards=7, num_host_shards=3)
+  waterfall = AddTester(waterfall, 'Android One Perf', 'android-one', 'android',
+    target_bits=32, num_device_shards=7, num_host_shards=3)
+
+  waterfall = AddTester(waterfall, 'Win Zenbook Perf', 'win-zenbook', 'win',
+    num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 10 Perf', 'chromium-rel-win10', 'win',
+    num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 8 Perf', 'chromium-rel-win8-dual',
+    'win', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 7 Perf', 'chromium-rel-win7-dual',
+    'win', target_bits=32, num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 7 x64 Perf',
+    'chromium-rel-win7-x64-dual', 'win', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 7 ATI GPU Perf',
+    'chromium-rel-win7-gpu-ati', 'win', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 7 Intel GPU Perf',
+    'chromium-rel-win7-gpu-intel', 'win', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Win 7 Nvidia GPU Perf',
+    'chromium-rel-win7-gpu-nvidia', 'win', num_host_shards=5)
+
+  waterfall = AddTester(waterfall, 'Mac 10.11 Perf', 'chromium-rel-mac11',
+    'mac', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Mac 10.10 Perf', 'chromium-rel-mac10',
+    'mac', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Mac Retina Perf', 'chromium-rel-mac-retina',
+   'mac', num_host_shards=5)
+  waterfall = AddTester(waterfall, 'Mac HDD Perf', 'chromium-rel-mac-hdd',
+    'mac', num_host_shards=5)
+
+  waterfall = AddTester(waterfall, 'Linux Perf', 'linux-release', 'linux',
+               num_host_shards=5)
+
+  return waterfall
+
+def generate_telemetry_test(tester_config, benchmark_name, browser):
+  # The step name must end in 'test' or 'tests' in order for the
+  # results to automatically show up on the flakiness dashboard.
+  # (At least, this was true some time ago.) Continue to use this
+  # naming convention for the time being to minimize changes.
+
+  test_args = [
+    benchmark_name,
+    '-v',
+    '--upload_results',
+    '--output-format=chartjson',
+    '--browser=%s' % browser
+  ]

[dtu, 2016/08/25 17:39:05]

We need --device=android on android.

[eyaich1, 2016/08/29 18:38:09]

We don't need it yet, I added a comment.

+
+  step_name = benchmark_name
+  if browser == 'reference':
+    test_args.append('--output-trace-tag=_ref')
+    step_name += '.reference'
+  swarming = {
+    # Always say this is true regardless of whether the tester
+    # supports swarming. It doesn't hurt.
+    'can_use_on_swarming_builders': True,
+    'dimension_sets': [
+      tester_config['swarming_dimensions']
+    ]
+  }
+
+  result = {
+    'args': test_args,
+    'isolate_name': 'telemetry_perf_tests',
+    'name': step_name,
+    'override_compile_targets': 'telemetry_perf_tests',
+    'swarming': swarming,
+  }
+
+  return result
+
+def script_test_enabled_on_tester(test, tester_name, shard):
+  for enabled_tester in test['testers']:
+    if enabled_tester['name'] == tester_name:
+      for enabled_shard in enabled_tester['shards']:

[dtu, 2016/08/25 17:39:05]

if shard in enabled_tester['shards']:
  return True

[eyaich1, 2016/08/29 18:38:09]

Done.

+        if enabled_shard == shard:
+          return True
+  return False
+
+def generate_script_tests(tester_name, shard):
+  script_tests = []
+  for test in SCRIPT_TESTS:
+    if script_test_enabled_on_tester(test, tester_name, shard):
+      script = {
+        'args': test['args'],
+        'name': test['name'],
+        'script': test['script']
+      }
+      script_tests.append(script)
+  return script_tests
+
+def generate_telemetry_tests(tester_config, benchmarks):
+  isolated_scripts = []
+  # First determine the browser that you need based on the tester
+  browser_name = ''
+  if tester_config['platform'] == 'android':
+    browser_name = 'android-chromium'
+  elif tester_config['platform'] == 'win' \

[dtu, 2016/08/25 17:39:05]

style guide: don't use \. wrap in parentheses.

[eyaich1, 2016/08/29 18:38:09]

Done.

+    and tester_config['target_bits'] == 64:
+    browser_name = 'release_x64'
+  else:
+    browser_name ='release'
+  for benchmark in benchmarks:
+    test = generate_telemetry_test(
+      tester_config, benchmark.Name(), browser_name)
+    isolated_scripts.append(test)
+    # Now create another executable for this benchmark on the reference browser
+    reference_test = generate_telemetry_test(
+      tester_config, benchmark.Name(),'reference')
+    isolated_scripts.append(reference_test)
+  return isolated_scripts
+
+def current_benchmarks():
+  current_dir = os.path.dirname(__file__)
+  benchmarks_dir = os.path.join(current_dir, 'benchmarks')
+  top_level_dir = os.path.dirname(benchmarks_dir)
+
+  return discover.DiscoverClasses(
+      benchmarks_dir, top_level_dir, benchmark_module.Benchmark,
+      index_by_class_name=True).values()
+
+def generate_all_tests(waterfall, is_fyi):
+  tests = {}
+  for builder in waterfall['builders']:
+    tests[builder] = {}
+  for name, config in waterfall['testers'].iteritems():
+    if is_fyi:
+      # Right now we are only generating benchmarks for the fyi waterfall
+      all_benchmarks = current_benchmarks()
+      isolated_scripts = generate_telemetry_tests(config, all_benchmarks)
+      tests[name] = {
+        'isolated_scripts': sorted(isolated_scripts, key=lambda x: x['name'])
+      }
+    else:
+      # scripts are only currently run in addition to the main waterfall.  They
+      # are currently the only thing generated in the perf json file.
+      # TODO eyaich: will need to handle the sharding differently when we have
+      # swarmed bots on the main waterfall.
+      for shard in range(0, config['num_host_shards']):
+        tester_name = '%s (%d)' % (name, shard + 1)
+        scripts = generate_script_tests(name, shard + 1)
+        if len(scripts) > 0:

[dtu, 2016/08/25 17:39:05]

style guide: implicit false. if scripts:

[eyaich1, 2016/08/29 18:38:09]

Done.

+          tests[tester_name] = {
+            'scripts': sorted(scripts, key=lambda x: x['name'])
+          }
+
+  tests['AAAAA1 AUTOGENERATED FILE DO NOT EDIT'] = {}
+  tests['AAAAA2 See generate_perf_json.py to make changes'] = {}
+  filename = 'chromium.perf.fyi.json' if is_fyi else 'chromium.perf.json'
+
+  current_dir = os.path.dirname(os.path.abspath(__file__))
+  src_dir = os.path.dirname(os.path.dirname(current_dir))
+
+  with open(os.path.join(src_dir, 'testing', 'buildbot', filename), 'w') as fp:
+    json.dump(tests, fp, indent=2, separators=(',', ': '), sort_keys=True)
+    fp.write('\n')
+
+def main():
+  waterfall = generateWaterfall()

[dtu, 2016/08/25 17:39:05]

It's inconsistent that generateWaterfall and generate_all_tests have similar
names, but one returns its result and the other writes its result to a file.

[eyaich1, 2016/08/29 18:38:10]

Done.

+  fyi_waterfall = generateFyiWaterfall()
+  generate_all_tests(fyi_waterfall, True)
+  generate_all_tests(waterfall, False)
+  return 0
+
+if __name__ == "__main__":
+  sys.exit(main())