OLD | NEW |
| (Empty) |
1 #!/usr/bin/python | |
2 | |
3 # Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file | |
4 # for details. All rights reserved. Use of this source code is governed by a | |
5 # BSD-style license that can be found in the LICENSE file. | |
6 | |
7 import datetime | |
8 import math | |
9 import optparse | |
10 import os | |
11 from os.path import dirname, abspath | |
12 import pickle | |
13 import platform | |
14 import random | |
15 import re | |
16 import shutil | |
17 import stat | |
18 import subprocess | |
19 import sys | |
20 import time | |
21 | |
22 TOOLS_PATH = os.path.join(dirname(dirname(dirname(abspath(__file__))))) | |
23 TOP_LEVEL_DIR = abspath(os.path.join(dirname(abspath(__file__)), '..', '..', | |
24 '..')) | |
25 DART_REPO_LOC = abspath(os.path.join(dirname(abspath(__file__)), '..', '..', | |
26 '..', '..', '..', | |
27 'dart_checkout_for_perf_testing', | |
28 'dart')) | |
29 # How far back in time we want to test. | |
30 EARLIEST_REVISION = 33076 | |
31 sys.path.append(TOOLS_PATH) | |
32 sys.path.append(os.path.join(TOP_LEVEL_DIR, 'internal', 'tests')) | |
33 import post_results | |
34 import utils | |
35 | |
36 """This script runs to track performance and size progress of | |
37 different svn revisions. It tests to see if there a newer version of the code on | |
38 the server, and will sync and run the performance tests if so.""" | |
39 class TestRunner(object): | |
40 | |
41 def __init__(self): | |
42 self.verbose = False | |
43 self.has_shell = False | |
44 if platform.system() == 'Windows': | |
45 # On Windows, shell must be true to get the correct environment variables. | |
46 self.has_shell = True | |
47 self.current_revision_num = None | |
48 | |
49 def RunCmd(self, cmd_list, outfile=None, append=False, std_in=''): | |
50 """Run the specified command and print out any output to stdout. | |
51 | |
52 Args: | |
53 cmd_list: a list of strings that make up the command to run | |
54 outfile: a string indicating the name of the file that we should write | |
55 stdout to | |
56 append: True if we want to append to the file instead of overwriting it | |
57 std_in: a string that should be written to the process executing to | |
58 interact with it (if needed)""" | |
59 if self.verbose: | |
60 print ' '.join(cmd_list) | |
61 out = subprocess.PIPE | |
62 if outfile: | |
63 mode = 'w' | |
64 if append: | |
65 mode = 'a+' | |
66 out = open(outfile, mode) | |
67 if append: | |
68 # Annoying Windows "feature" -- append doesn't actually append unless | |
69 # you explicitly go to the end of the file. | |
70 # http://mail.python.org/pipermail/python-list/2009-October/1221859.html | |
71 out.seek(0, os.SEEK_END) | |
72 p = subprocess.Popen(cmd_list, stdout = out, stderr=subprocess.PIPE, | |
73 stdin=subprocess.PIPE, shell=self.has_shell) | |
74 output, stderr = p.communicate(std_in) | |
75 if output: | |
76 print output | |
77 if stderr: | |
78 print stderr | |
79 return output, stderr | |
80 | |
81 def RunBrowserPerfRunnerCmd(self, browser, url_path, file_path_to_test_code, | |
82 trace_file, code_root=''): | |
83 command_list = [os.path.join(DART_REPO_LOC, utils.GetBuildRoot( | |
84 utils.GuessOS(), 'release', 'ia32'), 'dart-sdk', 'bin', 'dart'), | |
85 '--package-root=%s' % os.path.join(file_path_to_test_code, 'packages'), | |
86 os.path.join(file_path_to_test_code, 'packages', | |
87 'browser_controller', 'browser_perf_testing.dart'), '--browser', | |
88 browser, '--test_path=%s' % url_path] | |
89 if code_root != '': | |
90 command_list += ['--code_root=%s' % code_root] | |
91 | |
92 if browser == 'dartium': | |
93 dartium_path = os.path.join(DART_REPO_LOC, 'client', 'tests', 'dartium') | |
94 if platform.system() == 'Windows': | |
95 dartium_path = os.path.join(dartium_path, 'chrome.exe'); | |
96 elif platform.system() == 'Darwin': | |
97 dartium_path = os.path.join(dartium_path, 'Chromium.app', 'Contents', | |
98 'MacOS', 'Chromium') | |
99 else: | |
100 dartium_path = os.path.join(dartium_path, 'chrome') | |
101 command_list += ['--executable=%s' % dartium_path] | |
102 | |
103 self.RunCmd(command_list, trace_file, append=True) | |
104 | |
105 def TimeCmd(self, cmd): | |
106 """Determine the amount of (real) time it takes to execute a given | |
107 command.""" | |
108 start = time.time() | |
109 self.RunCmd(cmd) | |
110 return time.time() - start | |
111 | |
112 def ClearOutUnversionedFiles(self): | |
113 """Remove all files that are unversioned by svn.""" | |
114 if os.path.exists(DART_REPO_LOC): | |
115 os.chdir(DART_REPO_LOC) | |
116 results, _ = self.RunCmd(['svn', 'st']) | |
117 for line in results.split('\n'): | |
118 if line.startswith('?'): | |
119 to_remove = line.split()[1] | |
120 if os.path.isdir(to_remove): | |
121 shutil.rmtree(to_remove, onerror=TestRunner._OnRmError) | |
122 else: | |
123 os.remove(to_remove) | |
124 elif any(line.startswith(status) for status in ['A', 'M', 'C', 'D']): | |
125 self.RunCmd(['svn', 'revert', line.split()[1]]) | |
126 | |
127 def GetArchive(self, archive_name): | |
128 """Wrapper around the pulling down a specific archive from Google Storage. | |
129 Adds a specific revision argument as needed. | |
130 Returns: A tuple of a boolean (True if we successfully downloaded the | |
131 binary), and the stdout and stderr from running this command.""" | |
132 num_fails = 0 | |
133 while True: | |
134 cmd = ['python', os.path.join(DART_REPO_LOC, 'tools', 'get_archive.py'), | |
135 archive_name] | |
136 if int(self.current_revision_num) != -1: | |
137 cmd += ['-r', str(self.current_revision_num)] | |
138 stdout, stderr = self.RunCmd(cmd) | |
139 if 'Please try again later' in stdout and num_fails < 20: | |
140 time.sleep(100) | |
141 num_fails += 1 | |
142 else: | |
143 break | |
144 return (num_fails < 20, stdout, stderr) | |
145 | |
146 def _Sync(self, revision_num=None): | |
147 """Update the repository to the latest or specified revision.""" | |
148 os.chdir(dirname(DART_REPO_LOC)) | |
149 self.ClearOutUnversionedFiles() | |
150 if not revision_num: | |
151 self.RunCmd(['gclient', 'sync']) | |
152 else: | |
153 self.RunCmd(['gclient', 'sync', '-r', str(revision_num), '-t']) | |
154 | |
155 shutil.copytree(os.path.join(TOP_LEVEL_DIR, 'internal'), | |
156 os.path.join(DART_REPO_LOC, 'internal')) | |
157 shutil.rmtree(os.path.join(DART_REPO_LOC, 'third_party', 'gsutil'), | |
158 onerror=TestRunner._OnRmError) | |
159 shutil.copytree(os.path.join(TOP_LEVEL_DIR, 'third_party', 'gsutil'), | |
160 os.path.join(DART_REPO_LOC, 'third_party', 'gsutil')) | |
161 shutil.copy(os.path.join(TOP_LEVEL_DIR, 'tools', 'get_archive.py'), | |
162 os.path.join(DART_REPO_LOC, 'tools', 'get_archive.py')) | |
163 shutil.copy( | |
164 os.path.join(TOP_LEVEL_DIR, 'tools', 'testing', 'run_selenium.py'), | |
165 os.path.join(DART_REPO_LOC, 'tools', 'testing', 'run_selenium.py')) | |
166 | |
167 @staticmethod | |
168 def _OnRmError(func, path, exc_info): | |
169 """On Windows, the output directory is marked as "Read Only," which causes | |
170 an error to be thrown when we use shutil.rmtree. This helper function | |
171 changes the permissions so we can still delete the directory.""" | |
172 if os.path.exists(path): | |
173 os.chmod(path, stat.S_IWRITE) | |
174 os.unlink(path) | |
175 | |
176 def SyncAndBuild(self, suites, revision_num=None): | |
177 """Make sure we have the latest version of of the repo, and build it. We | |
178 begin and end standing in DART_REPO_LOC. | |
179 | |
180 Args: | |
181 suites: The set of suites that we wish to build. | |
182 | |
183 Returns: | |
184 err_code = 1 if there was a problem building.""" | |
185 self._Sync(revision_num) | |
186 if not revision_num: | |
187 revision_num = SearchForRevision() | |
188 | |
189 self.current_revision_num = revision_num | |
190 success, stdout, stderr = self.GetArchive('sdk') | |
191 if (not os.path.exists(os.path.join( | |
192 DART_REPO_LOC, 'tools', 'get_archive.py')) or not success | |
193 or 'InvalidUriError' in stderr or "Couldn't download" in stdout or | |
194 'Unable to download' in stdout): | |
195 # Couldn't find the SDK on Google Storage. Build it locally. | |
196 | |
197 # TODO(efortuna): Currently always building ia32 architecture because we | |
198 # don't have test statistics for what's passing on x64. Eliminate arch | |
199 # specification when we have tests running on x64, too. | |
200 shutil.rmtree(os.path.join(os.getcwd(), | |
201 utils.GetBuildRoot(utils.GuessOS())), | |
202 onerror=TestRunner._OnRmError) | |
203 lines = self.RunCmd(['python', os.path.join('tools', 'build.py'), '-m', | |
204 'release', '--arch=ia32', 'create_sdk']) | |
205 | |
206 for line in lines: | |
207 if 'BUILD FAILED' in line: | |
208 # Someone checked in a broken build! Stop trying to make it work | |
209 # and wait to try again. | |
210 print 'Broken Build' | |
211 return 1 | |
212 return 0 | |
213 | |
214 def EnsureOutputDirectory(self, dir_name): | |
215 """Test that the listed directory name exists, and if not, create one for | |
216 our output to be placed. | |
217 | |
218 Args: | |
219 dir_name: the directory we will create if it does not exist.""" | |
220 dir_path = os.path.join(TOP_LEVEL_DIR, 'tools', | |
221 'testing', 'perf_testing', dir_name) | |
222 if not os.path.exists(dir_path): | |
223 os.makedirs(dir_path) | |
224 print 'Creating output directory ', dir_path | |
225 | |
226 def HasInterestingCode(self, revision_num=None): | |
227 """Tests if there are any versions of files that might change performance | |
228 results on the server. | |
229 | |
230 Returns: | |
231 (False, None): There is no interesting code to run. | |
232 (True, revisionNumber): There is interesting code to run at revision | |
233 revisionNumber. | |
234 (True, None): There is interesting code to run by syncing to the | |
235 tip-of-tree.""" | |
236 if not os.path.exists(DART_REPO_LOC): | |
237 self._Sync() | |
238 os.chdir(DART_REPO_LOC) | |
239 no_effect = ['dart/client', 'dart/compiler', 'dart/editor', | |
240 'dart/lib/html/doc', 'dart/pkg', 'dart/tests', 'dart/samples', | |
241 'dart/lib/dartdoc', 'dart/lib/i18n', 'dart/lib/unittest', | |
242 'dart/tools/dartc', 'dart/tools/get_archive.py', | |
243 'dart/tools/test.py', 'dart/tools/testing', | |
244 'dart/tools/utils', 'dart/third_party', 'dart/utils'] | |
245 definitely_yes = ['dart/samples/third_party/dromaeo', | |
246 'dart/lib/html/dart2js', 'dart/lib/html/dartium', | |
247 'dart/lib/scripts', 'dart/lib/src', | |
248 'dart/third_party/WebCore'] | |
249 def GetFileList(revision): | |
250 """Determine the set of files that were changed for a particular | |
251 revision.""" | |
252 # TODO(efortuna): This assumes you're using svn. Have a git fallback as | |
253 # well. Pass 'p' in if we have a new certificate for the svn server, we | |
254 # want to (p)ermanently accept it. | |
255 results, _ = self.RunCmd([ | |
256 'svn', 'log', 'http://dart.googlecode.com/svn/branches/bleeding_edge', | |
257 '-v', '-r', str(revision)], std_in='p\r\n') | |
258 results = results.split('\n') | |
259 if len(results) <= 3: | |
260 return [] | |
261 else: | |
262 # Trim off the details about revision number and commit message. We're | |
263 # only interested in the files that are changed. | |
264 results = results[3:] | |
265 changed_files = [] | |
266 for result in results: | |
267 if len(result) <= 1: | |
268 break | |
269 tokens = result.split() | |
270 if len(tokens) > 1: | |
271 changed_files += [tokens[1].replace('/branches/bleeding_edge/', '')] | |
272 return changed_files | |
273 | |
274 def HasPerfAffectingResults(files_list): | |
275 """Determine if this set of changed files might effect performance | |
276 tests.""" | |
277 def IsSafeFile(f): | |
278 if not any(f.startswith(prefix) for prefix in definitely_yes): | |
279 return any(f.startswith(prefix) for prefix in no_effect) | |
280 return False | |
281 return not all(IsSafeFile(f) for f in files_list) | |
282 | |
283 if revision_num: | |
284 return (HasPerfAffectingResults(GetFileList( | |
285 revision_num)), revision_num) | |
286 else: | |
287 latest_interesting_server_rev = None | |
288 while not latest_interesting_server_rev: | |
289 results, _ = self.RunCmd(['svn', 'st', '-u'], std_in='p\r\n') | |
290 if len(results.split('\n')) >= 2: | |
291 latest_interesting_server_rev = int( | |
292 results.split('\n')[-2].split()[-1]) | |
293 if self.backfill: | |
294 done_cls = list(UpdateSetOfDoneCls()) | |
295 done_cls.sort() | |
296 if done_cls: | |
297 last_done_cl = int(done_cls[-1]) | |
298 else: | |
299 last_done_cl = EARLIEST_REVISION | |
300 while latest_interesting_server_rev >= last_done_cl: | |
301 file_list = GetFileList(latest_interesting_server_rev) | |
302 if HasPerfAffectingResults(file_list): | |
303 return (True, latest_interesting_server_rev) | |
304 else: | |
305 UpdateSetOfDoneCls(latest_interesting_server_rev) | |
306 latest_interesting_server_rev -= 1 | |
307 else: | |
308 last_done_cl = int(SearchForRevision(DART_REPO_LOC)) + 1 | |
309 while last_done_cl <= latest_interesting_server_rev: | |
310 file_list = GetFileList(last_done_cl) | |
311 if HasPerfAffectingResults(file_list): | |
312 return (True, last_done_cl) | |
313 else: | |
314 UpdateSetOfDoneCls(last_done_cl) | |
315 last_done_cl += 1 | |
316 return (False, None) | |
317 | |
318 def GetOsDirectory(self): | |
319 """Specifies the name of the directory for the testing build of dart, which | |
320 has yet a different naming convention from utils.getBuildRoot(...).""" | |
321 if platform.system() == 'Windows': | |
322 return 'windows' | |
323 elif platform.system() == 'Darwin': | |
324 return 'macos' | |
325 else: | |
326 return 'linux' | |
327 | |
328 def ParseArgs(self): | |
329 parser = optparse.OptionParser() | |
330 parser.add_option('--suites', '-s', dest='suites', help='Run the specified ' | |
331 'comma-separated test suites from set: %s' % \ | |
332 ','.join(TestBuilder.AvailableSuiteNames()), | |
333 action='store', default=None) | |
334 parser.add_option('--forever', '-f', dest='continuous', help='Run this scri' | |
335 'pt forever, always checking for the next svn checkin', | |
336 action='store_true', default=False) | |
337 parser.add_option('--nobuild', '-n', dest='no_build', action='store_true', | |
338 help='Do not sync with the repository and do not ' | |
339 'rebuild.', default=False) | |
340 parser.add_option('--noupload', '-u', dest='no_upload', action='store_true', | |
341 help='Do not post the results of the run.', default=False) | |
342 parser.add_option('--notest', '-t', dest='no_test', action='store_true', | |
343 help='Do not run the tests.', default=False) | |
344 parser.add_option('--verbose', '-v', dest='verbose', | |
345 help='Print extra debug output', action='store_true', | |
346 default=False) | |
347 parser.add_option('--backfill', '-b', dest='backfill', | |
348 help='Backfill earlier CLs with additional results when ' | |
349 'there is idle time.', action='store_true', | |
350 default=False) | |
351 | |
352 args, ignored = parser.parse_args() | |
353 | |
354 if not args.suites: | |
355 suites = TestBuilder.AvailableSuiteNames() | |
356 else: | |
357 suites = [] | |
358 suitelist = args.suites.split(',') | |
359 for name in suitelist: | |
360 if name in TestBuilder.AvailableSuiteNames(): | |
361 suites.append(name) | |
362 else: | |
363 print ('Error: Invalid suite %s not in ' % name) + \ | |
364 '%s' % ','.join(TestBuilder.AvailableSuiteNames()) | |
365 sys.exit(1) | |
366 self.suite_names = suites | |
367 self.no_build = args.no_build | |
368 self.no_upload = args.no_upload | |
369 self.no_test = args.no_test | |
370 self.verbose = args.verbose | |
371 self.backfill = args.backfill | |
372 return args.continuous | |
373 | |
374 def RunTestSequence(self, revision_num=None, num_reruns=1): | |
375 """Run the set of commands to (possibly) build, run, and post the results | |
376 of our tests. Returns 0 on a successful run, 1 if we fail to post results or | |
377 the run failed, -1 if the build is broken. | |
378 """ | |
379 suites = [] | |
380 success = True | |
381 if not self.no_build and self.SyncAndBuild(suites, revision_num) == 1: | |
382 return -1 # The build is broken. | |
383 | |
384 if not self.current_revision_num: | |
385 self.current_revision_num = SearchForRevision(DART_REPO_LOC) | |
386 | |
387 for name in self.suite_names: | |
388 for run in range(num_reruns): | |
389 suites += [TestBuilder.MakeTest(name, self)] | |
390 | |
391 for test in suites: | |
392 success = success and test.Run() | |
393 if success: | |
394 return 0 | |
395 else: | |
396 return 1 | |
397 | |
398 | |
399 class Test(object): | |
400 """The base class to provide shared code for different tests we will run and | |
401 post. At a high level, each test has three visitors (the tester and the | |
402 file_processor) that perform operations on the test object.""" | |
403 | |
404 def __init__(self, result_folder_name, platform_list, variants, | |
405 values_list, test_runner, tester, file_processor, | |
406 extra_metrics=['Geo-Mean']): | |
407 """Args: | |
408 result_folder_name: The name of the folder where a tracefile of | |
409 performance results will be stored. | |
410 platform_list: A list containing the platform(s) that our data has been | |
411 run on. (command line, firefox, chrome, etc) | |
412 variants: A list specifying whether we hold data about Frog | |
413 generated code, plain JS code, or a combination of both, or | |
414 Dart depending on the test. | |
415 values_list: A list containing the type of data we will be graphing | |
416 (benchmarks, percentage passing, etc). | |
417 test_runner: Reference to the parent test runner object that notifies a | |
418 test when to run. | |
419 tester: The visitor that actually performs the test running mechanics. | |
420 file_processor: The visitor that processes files in the format | |
421 appropriate for this test. | |
422 extra_metrics: A list of any additional measurements we wish to keep | |
423 track of (such as the geometric mean of a set, the sum, etc).""" | |
424 self.result_folder_name = result_folder_name | |
425 # cur_time is used as a timestamp of when this performance test was run. | |
426 self.cur_time = str(time.mktime(datetime.datetime.now().timetuple())) | |
427 self.values_list = values_list | |
428 self.platform_list = platform_list | |
429 self.test_runner = test_runner | |
430 self.tester = tester | |
431 self.file_processor = file_processor | |
432 self.revision_dict = dict() | |
433 self.values_dict = dict() | |
434 self.extra_metrics = extra_metrics | |
435 # Initialize our values store. | |
436 for platform in platform_list: | |
437 self.revision_dict[platform] = dict() | |
438 self.values_dict[platform] = dict() | |
439 for f in variants: | |
440 self.revision_dict[platform][f] = dict() | |
441 self.values_dict[platform][f] = dict() | |
442 for val in values_list: | |
443 self.revision_dict[platform][f][val] = [] | |
444 self.values_dict[platform][f][val] = [] | |
445 for extra_metric in extra_metrics: | |
446 self.revision_dict[platform][f][extra_metric] = [] | |
447 self.values_dict[platform][f][extra_metric] = [] | |
448 | |
449 def IsValidCombination(self, platform, variant): | |
450 """Check whether data should be captured for this platform/variant | |
451 combination. | |
452 """ | |
453 if variant == 'dart_html' and platform != 'dartium': | |
454 return False | |
455 if platform == 'dartium' and (variant == 'js' or variant == 'dart2js_html'): | |
456 # Testing JavaScript performance on Dartium is a waste of time. Should be | |
457 # same as Chrome. | |
458 return False | |
459 if (platform == 'safari' and variant == 'dart2js' and | |
460 int(self.test_runner.current_revision_num) < 10193): | |
461 # In revision 10193 we fixed a bug that allows Safari 6 to run dart2js | |
462 # code. Since we can't change the Safari version on the machine, we're | |
463 # just not running | |
464 # for this case. | |
465 return False | |
466 return True | |
467 | |
468 def Run(self): | |
469 """Run the benchmarks/tests from the command line and plot the | |
470 results. | |
471 """ | |
472 for visitor in [self.tester, self.file_processor]: | |
473 visitor.Prepare() | |
474 | |
475 os.chdir(TOP_LEVEL_DIR) | |
476 self.test_runner.EnsureOutputDirectory(self.result_folder_name) | |
477 self.test_runner.EnsureOutputDirectory(os.path.join( | |
478 'old', self.result_folder_name)) | |
479 os.chdir(DART_REPO_LOC) | |
480 if not self.test_runner.no_test: | |
481 self.tester.RunTests() | |
482 | |
483 os.chdir(os.path.join(TOP_LEVEL_DIR, 'tools', 'testing', 'perf_testing')) | |
484 | |
485 files = os.listdir(self.result_folder_name) | |
486 post_success = True | |
487 for afile in files: | |
488 if not afile.startswith('.'): | |
489 should_move_file = self.file_processor.ProcessFile(afile, True) | |
490 if should_move_file: | |
491 shutil.move(os.path.join(self.result_folder_name, afile), | |
492 os.path.join('old', self.result_folder_name, afile)) | |
493 else: | |
494 post_success = False | |
495 | |
496 return post_success | |
497 | |
498 | |
499 class Tester(object): | |
500 """The base level visitor class that runs tests. It contains convenience | |
501 methods that many Tester objects use. Any class that would like to be a | |
502 TesterVisitor must implement the RunTests() method.""" | |
503 | |
504 def __init__(self, test): | |
505 self.test = test | |
506 | |
507 def Prepare(self): | |
508 """Perform any initial setup required before the test is run.""" | |
509 pass | |
510 | |
511 def AddSvnRevisionToTrace(self, outfile, browser = None): | |
512 """Add the svn version number to the provided tracefile.""" | |
513 def get_dartium_revision(): | |
514 version_file_name = os.path.join(DART_REPO_LOC, 'client', 'tests', | |
515 'dartium', 'LAST_VERSION') | |
516 try: | |
517 version_file = open(version_file_name, 'r') | |
518 version = version_file.read().split('.')[-3].split('-')[-1] | |
519 version_file.close() | |
520 return version | |
521 except IOError as e: | |
522 dartium_dir = os.path.join(DART_REPO_LOC, 'client', 'tests', 'dartium') | |
523 if (os.path.exists(os.path.join(dartium_dir, 'Chromium.app', 'Contents', | |
524 'MacOS', 'Chromium') or os.path.exists(os.path.join(dartium_dir, | |
525 'chrome.exe'))) or | |
526 os.path.exists(os.path.join(dartium_dir, 'chrome'))): | |
527 print "Error: VERSION file wasn't found." | |
528 return SearchForRevision() | |
529 else: | |
530 raise | |
531 | |
532 if browser and browser == 'dartium': | |
533 revision = get_dartium_revision() | |
534 self.test.test_runner.RunCmd(['echo', 'Revision: ' + revision], outfile) | |
535 else: | |
536 revision = SearchForRevision() | |
537 self.test.test_runner.RunCmd(['echo', 'Revision: ' + revision], outfile) | |
538 | |
539 | |
540 class Processor(object): | |
541 """The base level vistor class that processes tests. It contains convenience | |
542 methods that many File Processor objects use. Any class that would like to be | |
543 a ProcessorVisitor must implement the ProcessFile() method.""" | |
544 | |
545 SCORE = 'Score' | |
546 COMPILE_TIME = 'CompileTime' | |
547 CODE_SIZE = 'CodeSize' | |
548 | |
549 def __init__(self, test): | |
550 self.test = test | |
551 | |
552 def Prepare(self): | |
553 """Perform any initial setup required before the test is run.""" | |
554 pass | |
555 | |
556 def OpenTraceFile(self, afile, not_yet_uploaded): | |
557 """Find the correct location for the trace file, and open it. | |
558 Args: | |
559 afile: The tracefile name. | |
560 not_yet_uploaded: True if this file is to be found in a directory that | |
561 contains un-uploaded data. | |
562 Returns: A file object corresponding to the given file name.""" | |
563 file_path = os.path.join(self.test.result_folder_name, afile) | |
564 if not not_yet_uploaded: | |
565 file_path = os.path.join('old', file_path) | |
566 return open(file_path) | |
567 | |
568 def ReportResults(self, benchmark_name, score, platform, variant, | |
569 revision_number, metric): | |
570 """Store the results of the benchmark run. | |
571 Args: | |
572 benchmark_name: The name of the individual benchmark. | |
573 score: The numerical value of this benchmark. | |
574 platform: The platform the test was run on (firefox, command line, etc). | |
575 variant: Specifies whether the data was about generated Frog, js, a | |
576 combination of both, or Dart depending on the test. | |
577 revision_number: The revision of the code (and sometimes the revision of | |
578 dartium). | |
579 | |
580 Returns: True if the post was successful file.""" | |
581 return post_results.report_results(benchmark_name, score, platform, variant, | |
582 revision_number, metric) | |
583 | |
584 def CalculateGeometricMean(self, platform, variant, svn_revision): | |
585 """Calculate the aggregate geometric mean for JS and dart2js benchmark sets, | |
586 given two benchmark dictionaries.""" | |
587 geo_mean = 0 | |
588 if self.test.IsValidCombination(platform, variant): | |
589 for benchmark in self.test.values_list: | |
590 if not self.test.values_dict[platform][variant][benchmark]: | |
591 print 'Error determining mean for %s %s %s' % (platform, variant, | |
592 benchmark) | |
593 continue | |
594 geo_mean += math.log( | |
595 self.test.values_dict[platform][variant][benchmark][-1]) | |
596 | |
597 self.test.values_dict[platform][variant]['Geo-Mean'] += \ | |
598 [math.pow(math.e, geo_mean / len(self.test.values_list))] | |
599 self.test.revision_dict[platform][variant]['Geo-Mean'] += [svn_revision] | |
600 | |
601 def GetScoreType(self, benchmark_name): | |
602 """Determine the type of score for posting -- default is 'Score' (aka | |
603 Runtime), other options are CompileTime and CodeSize.""" | |
604 return self.SCORE | |
605 | |
606 | |
607 class RuntimePerformanceTest(Test): | |
608 """Super class for all runtime performance testing.""" | |
609 | |
610 def __init__(self, result_folder_name, platform_list, platform_type, | |
611 versions, benchmarks, test_runner, tester, file_processor): | |
612 """Args: | |
613 result_folder_name: The name of the folder where a tracefile of | |
614 performance results will be stored. | |
615 platform_list: A list containing the platform(s) that our data has been | |
616 run on. (command line, firefox, chrome, etc) | |
617 variants: A list specifying whether we hold data about Frog | |
618 generated code, plain JS code, or a combination of both, or | |
619 Dart depending on the test. | |
620 values_list: A list containing the type of data we will be graphing | |
621 (benchmarks, percentage passing, etc). | |
622 test_runner: Reference to the parent test runner object that notifies a | |
623 test when to run. | |
624 tester: The visitor that actually performs the test running mechanics. | |
625 file_processor: The visitor that processes files in the format | |
626 appropriate for this test. | |
627 extra_metrics: A list of any additional measurements we wish to keep | |
628 track of (such as the geometric mean of a set, the sum, etc).""" | |
629 super(RuntimePerformanceTest, self).__init__(result_folder_name, | |
630 platform_list, versions, benchmarks, test_runner, tester, | |
631 file_processor) | |
632 self.platform_list = platform_list | |
633 self.platform_type = platform_type | |
634 self.versions = versions | |
635 self.benchmarks = benchmarks | |
636 | |
637 | |
638 class BrowserTester(Tester): | |
639 @staticmethod | |
640 def GetBrowsers(add_dartium=True): | |
641 browsers = ['ff', 'chrome'] | |
642 if add_dartium: | |
643 browsers += ['dartium'] | |
644 has_shell = False | |
645 if platform.system() == 'Darwin': | |
646 browsers += ['safari'] | |
647 if platform.system() == 'Windows': | |
648 browsers += ['ie'] | |
649 has_shell = True | |
650 return browsers | |
651 | |
652 | |
653 class DromaeoTester(Tester): | |
654 DROMAEO_BENCHMARKS = { | |
655 'attr': ('attributes', [ | |
656 'getAttribute', | |
657 'element.property', | |
658 'setAttribute', | |
659 'element.property = value']), | |
660 'modify': ('modify', [ | |
661 'createElement', | |
662 'createTextNode', | |
663 'innerHTML', | |
664 'cloneNode', | |
665 'appendChild', | |
666 'insertBefore']), | |
667 'query': ('query', [ | |
668 'getElementById', | |
669 'getElementById (not in document)', | |
670 'getElementsByTagName(div)', | |
671 'getElementsByTagName(p)', | |
672 'getElementsByTagName(a)', | |
673 'getElementsByTagName(*)', | |
674 'getElementsByTagName (not in document)', | |
675 'getElementsByName', | |
676 'getElementsByName (not in document)']), | |
677 'traverse': ('traverse', [ | |
678 'firstChild', | |
679 'lastChild', | |
680 'nextSibling', | |
681 'previousSibling', | |
682 'childNodes']) | |
683 } | |
684 | |
685 # Use filenames that don't have unusual characters for benchmark names. | |
686 @staticmethod | |
687 def LegalizeFilename(str): | |
688 remap = { | |
689 ' ': '_', | |
690 '(': '_', | |
691 ')': '_', | |
692 '*': 'ALL', | |
693 '=': 'ASSIGN', | |
694 } | |
695 for (old, new) in remap.iteritems(): | |
696 str = str.replace(old, new) | |
697 return str | |
698 | |
699 # TODO(vsm): This is a hack to skip breaking tests. Triage this | |
700 # failure properly. The modify suite fails on 32-bit chrome, which | |
701 # is the default on mac and win. | |
702 @staticmethod | |
703 def GetValidDromaeoTags(): | |
704 tags = [tag for (tag, _) in DromaeoTester.DROMAEO_BENCHMARKS.values()] | |
705 if platform.system() == 'Darwin' or platform.system() == 'Windows': | |
706 tags.remove('modify') | |
707 return tags | |
708 | |
709 @staticmethod | |
710 def GetDromaeoBenchmarks(): | |
711 valid = DromaeoTester.GetValidDromaeoTags() | |
712 benchmarks = reduce(lambda l1,l2: l1+l2, | |
713 [tests for (tag, tests) in | |
714 DromaeoTester.DROMAEO_BENCHMARKS.values() | |
715 if tag in valid]) | |
716 return map(DromaeoTester.LegalizeFilename, benchmarks) | |
717 | |
718 @staticmethod | |
719 def GetDromaeoVersions(): | |
720 return ['js', 'dart2js_html', 'dart_html'] | |
721 | |
722 | |
723 class DromaeoTest(RuntimePerformanceTest): | |
724 """Runs Dromaeo tests, in the browser.""" | |
725 def __init__(self, test_runner): | |
726 super(DromaeoTest, self).__init__( | |
727 self.Name(), | |
728 BrowserTester.GetBrowsers(True), | |
729 'browser', | |
730 DromaeoTester.GetDromaeoVersions(), | |
731 DromaeoTester.GetDromaeoBenchmarks(), test_runner, | |
732 self.DromaeoPerfTester(self), | |
733 self.DromaeoFileProcessor(self)) | |
734 | |
735 @staticmethod | |
736 def Name(): | |
737 return 'dromaeo' | |
738 | |
739 class DromaeoPerfTester(DromaeoTester): | |
740 def RunTests(self): | |
741 """Run dromaeo in the browser.""" | |
742 success, _, _ = self.test.test_runner.GetArchive('dartium') | |
743 if not success: | |
744 # Unable to download dartium. Try later. | |
745 return | |
746 | |
747 # Build tests. | |
748 current_path = os.getcwd() | |
749 os.chdir(os.path.join(DART_REPO_LOC, 'samples', 'third_party', | |
750 'dromaeo')) | |
751 # Note: This uses debug on purpose, so that we can also run performance | |
752 # tests on pure Dart applications in Dartium. Pub --debug simply also | |
753 # moves the .dart files to the build directory. To ensure effective | |
754 # comparison, though, ensure that minify: true is set in your transformer | |
755 # compilation step in your pubspec. | |
756 stdout, _ = self.test.test_runner.RunCmd([os.path.join(DART_REPO_LOC, | |
757 utils.GetBuildRoot(utils.GuessOS(), 'release', 'ia32'), | |
758 'dart-sdk', 'bin', 'pub'), 'build', '--mode=debug']) | |
759 os.chdir(current_path) | |
760 if 'failed' in stdout: | |
761 return | |
762 | |
763 versions = DromaeoTester.GetDromaeoVersions() | |
764 | |
765 for browser in BrowserTester.GetBrowsers(): | |
766 for version_name in versions: | |
767 if not self.test.IsValidCombination(browser, version_name): | |
768 continue | |
769 version = DromaeoTest.DromaeoPerfTester.GetDromaeoUrlQuery( | |
770 browser, version_name) | |
771 self.test.trace_file = os.path.join(TOP_LEVEL_DIR, | |
772 'tools', 'testing', 'perf_testing', self.test.result_folder_name, | |
773 'dromaeo-%s-%s-%s' % (self.test.cur_time, browser, version_name)) | |
774 self.AddSvnRevisionToTrace(self.test.trace_file, browser) | |
775 url_path = '/'.join(['/code_root', 'build', 'web', 'index%s.html?%s'%( | |
776 '-dart' if version_name == 'dart_html' else '-js', | |
777 version)]) | |
778 | |
779 self.test.test_runner.RunBrowserPerfRunnerCmd(browser, url_path, | |
780 os.path.join(DART_REPO_LOC, 'samples', 'third_party', 'dromaeo'), | |
781 self.test.trace_file) | |
782 | |
783 @staticmethod | |
784 def GetDromaeoUrlQuery(browser, version): | |
785 version = version.replace('_','AND') | |
786 tags = DromaeoTester.GetValidDromaeoTags() | |
787 return 'OR'.join([ '%sAND%s' % (version, tag) for tag in tags]) | |
788 | |
789 | |
790 class DromaeoFileProcessor(Processor): | |
791 def ProcessFile(self, afile, should_post_file): | |
792 """Comb through the html to find the performance results. | |
793 Returns: True if we successfully posted our data to storage.""" | |
794 parts = afile.split('-') | |
795 browser = parts[2] | |
796 version = parts[3] | |
797 | |
798 bench_dict = self.test.values_dict[browser][version] | |
799 | |
800 f = self.OpenTraceFile(afile, should_post_file) | |
801 lines = f.readlines() | |
802 i = 0 | |
803 revision_num = 0 | |
804 revision_pattern = r'Revision: (\d+)' | |
805 suite_pattern = r'<div class="result-item done">(.+?)</ol></div>' | |
806 result_pattern = r'<b>(.+?)</b>(.+?)<small> runs/s(.+)' | |
807 | |
808 upload_success = True | |
809 for line in lines: | |
810 rev = re.match(revision_pattern, line.strip().replace('"', '')) | |
811 if rev: | |
812 revision_num = int(rev.group(1)) | |
813 continue | |
814 | |
815 suite_results = re.findall(suite_pattern, line) | |
816 if suite_results: | |
817 for suite_result in suite_results: | |
818 results = re.findall(r'<li>(.*?)</li>', suite_result) | |
819 if results: | |
820 for result in results: | |
821 r = re.match(result_pattern, result) | |
822 name = DromaeoTester.LegalizeFilename(r.group(1).strip(':')) | |
823 score = float(r.group(2)) | |
824 bench_dict[name] += [float(score)] | |
825 self.test.revision_dict[browser][version][name] += \ | |
826 [revision_num] | |
827 if not self.test.test_runner.no_upload and should_post_file: | |
828 upload_success = upload_success and self.ReportResults( | |
829 name, score, browser, version, revision_num, | |
830 self.GetScoreType(name)) | |
831 else: | |
832 upload_success = False | |
833 | |
834 f.close() | |
835 self.CalculateGeometricMean(browser, version, revision_num) | |
836 return upload_success | |
837 | |
838 class TodoMVCTester(BrowserTester): | |
839 @staticmethod | |
840 def GetVersions(): | |
841 return ['js', 'dart2js_html', 'dart_html'] | |
842 | |
843 @staticmethod | |
844 def GetBenchmarks(): | |
845 return ['TodoMVCstartup'] | |
846 | |
847 class TodoMVCStartupTest(RuntimePerformanceTest): | |
848 """Start up TodoMVC and see how long it takes to start.""" | |
849 def __init__(self, test_runner): | |
850 super(TodoMVCStartupTest, self).__init__( | |
851 self.Name(), | |
852 BrowserTester.GetBrowsers(True), | |
853 'browser', | |
854 TodoMVCTester.GetVersions(), | |
855 TodoMVCTester.GetBenchmarks(), test_runner, | |
856 self.TodoMVCStartupTester(self), | |
857 self.TodoMVCFileProcessor(self)) | |
858 | |
859 @staticmethod | |
860 def Name(): | |
861 return 'todoMvcStartup' | |
862 | |
863 class TodoMVCStartupTester(BrowserTester): | |
864 def RunTests(self): | |
865 """Run dromaeo in the browser.""" | |
866 success, _, _ = self.test.test_runner.GetArchive('dartium') | |
867 if not success: | |
868 # Unable to download dartium. Try later. | |
869 return | |
870 | |
871 dromaeo_path = os.path.join('samples', 'third_party', 'dromaeo') | |
872 current_path = os.getcwd() | |
873 | |
874 os.chdir(os.path.join(DART_REPO_LOC, 'samples', 'third_party', | |
875 'todomvc_performance')) | |
876 self.test.test_runner.RunCmd([os.path.join(DART_REPO_LOC, | |
877 utils.GetBuildRoot(utils.GuessOS(), 'release', 'ia32'), | |
878 'dart-sdk', 'bin', 'pub'), 'build', '--mode=debug']) | |
879 os.chdir('js_todomvc'); | |
880 self.test.test_runner.RunCmd([os.path.join(DART_REPO_LOC, | |
881 utils.GetBuildRoot(utils.GuessOS(), 'release', 'ia32'), | |
882 'dart-sdk', 'bin', 'pub'), 'get']) | |
883 | |
884 versions = TodoMVCTester.GetVersions() | |
885 | |
886 for browser in BrowserTester.GetBrowsers(): | |
887 for version_name in versions: | |
888 if not self.test.IsValidCombination(browser, version_name): | |
889 continue | |
890 self.test.trace_file = os.path.join(TOP_LEVEL_DIR, | |
891 'tools', 'testing', 'perf_testing', self.test.result_folder_name, | |
892 'todoMvcStartup-%s-%s-%s' % (self.test.cur_time, browser, | |
893 version_name)) | |
894 self.AddSvnRevisionToTrace(self.test.trace_file, browser) | |
895 | |
896 if version_name == 'js': | |
897 code_root = os.path.join(DART_REPO_LOC, 'samples', 'third_party', | |
898 'todomvc_performance', 'js_todomvc') | |
899 self.test.test_runner.RunBrowserPerfRunnerCmd(browser, | |
900 '/code_root/index.html', code_root, self.test.trace_file, | |
901 code_root) | |
902 else: | |
903 self.test.test_runner.RunBrowserPerfRunnerCmd(browser, | |
904 '/code_root/build/web/startup-performance.html', os.path.join( | |
905 DART_REPO_LOC, 'samples', 'third_party', 'todomvc_performance'), | |
906 self.test.trace_file) | |
907 | |
908 class TodoMVCFileProcessor(Processor): | |
909 def ProcessFile(self, afile, should_post_file): | |
910 """Comb through the html to find the performance results. | |
911 Returns: True if we successfully posted our data to storage.""" | |
912 parts = afile.split('-') | |
913 browser = parts[2] | |
914 version = parts[3] | |
915 | |
916 bench_dict = self.test.values_dict[browser][version] | |
917 | |
918 f = self.OpenTraceFile(afile, should_post_file) | |
919 lines = f.readlines() | |
920 i = 0 | |
921 revision_num = 0 | |
922 revision_pattern = r'Revision: (\d+)' | |
923 result_pattern = r'The startup time is (\d+)' | |
924 | |
925 upload_success = True | |
926 for line in lines: | |
927 rev = re.match(revision_pattern, line.strip().replace('"', '')) | |
928 if rev: | |
929 revision_num = int(rev.group(1)) | |
930 continue | |
931 | |
932 results = re.search(result_pattern, line) | |
933 if results: | |
934 score = float(results.group(1)) | |
935 name = TodoMVCTester.GetBenchmarks()[0] | |
936 bench_dict[name] += [float(score)] | |
937 self.test.revision_dict[browser][version][name] += \ | |
938 [revision_num] | |
939 if not self.test.test_runner.no_upload and should_post_file: | |
940 upload_success = upload_success and self.ReportResults( | |
941 name, score, browser, version, revision_num, | |
942 self.GetScoreType(name)) | |
943 | |
944 f.close() | |
945 self.CalculateGeometricMean(browser, version, revision_num) | |
946 return upload_success | |
947 | |
948 | |
949 class TestBuilder(object): | |
950 """Construct the desired test object.""" | |
951 available_suites = dict((suite.Name(), suite) for suite in [ | |
952 DromaeoTest, TodoMVCStartupTest]) | |
953 | |
954 @staticmethod | |
955 def MakeTest(test_name, test_runner): | |
956 return TestBuilder.available_suites[test_name](test_runner) | |
957 | |
958 @staticmethod | |
959 def AvailableSuiteNames(): | |
960 return TestBuilder.available_suites.keys() | |
961 | |
962 | |
963 def SearchForRevision(directory = None): | |
964 """Find the current revision number in the desired directory. If directory is | |
965 None, find the revision number in the current directory.""" | |
966 def FindRevision(svn_info_command): | |
967 p = subprocess.Popen(svn_info_command, stdout = subprocess.PIPE, | |
968 stderr = subprocess.STDOUT, | |
969 shell = (platform.system() == 'Windows')) | |
970 output, _ = p.communicate() | |
971 for line in output.split('\n'): | |
972 if 'Revision' in line: | |
973 return int(line.split()[1]) | |
974 return -1 | |
975 | |
976 cwd = os.getcwd() | |
977 if not directory: | |
978 directory = cwd | |
979 os.chdir(directory) | |
980 revision_num = int(FindRevision(['svn', 'info'])) | |
981 if revision_num == -1: | |
982 revision_num = int(FindRevision(['git', 'svn', 'info'])) | |
983 os.chdir(cwd) | |
984 return str(revision_num) | |
985 | |
986 | |
987 def UpdateSetOfDoneCls(revision_num=None): | |
988 """Update the set of CLs that do not need additional performance runs. | |
989 Args: | |
990 revision_num: an additional number to be added to the 'done set' | |
991 """ | |
992 filename = os.path.join(TOP_LEVEL_DIR, 'cached_results.txt') | |
993 if not os.path.exists(filename): | |
994 f = open(filename, 'w') | |
995 results = set() | |
996 pickle.dump(results, f) | |
997 f.close() | |
998 f = open(filename, 'r+') | |
999 result_set = pickle.load(f) | |
1000 if revision_num: | |
1001 f.seek(0) | |
1002 result_set.add(revision_num) | |
1003 pickle.dump(result_set, f) | |
1004 f.close() | |
1005 return result_set | |
1006 | |
1007 | |
1008 def FillInBackHistory(results_set, runner): | |
1009 """Fill in back history performance data. This is done one of two ways, with | |
1010 equal probability of trying each way (falling back on the sequential version | |
1011 as our data becomes more densely populated).""" | |
1012 revision_num = int(SearchForRevision(DART_REPO_LOC)) | |
1013 has_run_extra = False | |
1014 | |
1015 def TryToRunAdditional(revision_number): | |
1016 """Determine the number of results we have stored for a particular revision | |
1017 number, and if it is less than 10, run some extra tests. | |
1018 Args: | |
1019 - revision_number: the revision whose performance we want to potentially | |
1020 test. | |
1021 Returns: True if we successfully ran some additional tests.""" | |
1022 if not runner.HasInterestingCode(revision_number)[0]: | |
1023 results_set = UpdateSetOfDoneCls(revision_number) | |
1024 return False | |
1025 a_test = TestBuilder.MakeTest(runner.suite_names[0], runner) | |
1026 benchmark_name = a_test.values_list[0] | |
1027 platform_name = a_test.platform_list[0] | |
1028 variant = a_test.values_dict[platform_name].keys()[0] | |
1029 num_results = post_results.get_num_results(benchmark_name, | |
1030 platform_name, variant, revision_number, | |
1031 a_test.file_processor.GetScoreType(benchmark_name)) | |
1032 if num_results < 10: | |
1033 # Run at most two more times. | |
1034 if num_results > 8: | |
1035 reruns = 10 - num_results | |
1036 else: | |
1037 reruns = 2 | |
1038 run = runner.RunTestSequence(revision_num=str(revision_number), | |
1039 num_reruns=reruns) | |
1040 if num_results >= 10 or run == 0 and num_results + reruns >= 10: | |
1041 results_set = UpdateSetOfDoneCls(revision_number) | |
1042 elif run != 0: | |
1043 return False | |
1044 return True | |
1045 | |
1046 # Try to get up to 10 runs of each CL, starting with the most recent | |
1047 # CL that does not yet have 10 runs. But only perform a set of extra | |
1048 # runs at most 2 at a time before checking to see if new code has been | |
1049 # checked in. | |
1050 while revision_num > EARLIEST_REVISION and not has_run_extra: | |
1051 if revision_num not in results_set: | |
1052 has_run_extra = TryToRunAdditional(revision_num) | |
1053 revision_num -= 1 | |
1054 if not has_run_extra: | |
1055 # No more extra back-runs to do (for now). Wait for new code. | |
1056 time.sleep(200) | |
1057 return results_set | |
1058 | |
1059 | |
1060 def main(): | |
1061 runner = TestRunner() | |
1062 continuous = runner.ParseArgs() | |
1063 | |
1064 if not os.path.exists(DART_REPO_LOC): | |
1065 os.mkdir(dirname(DART_REPO_LOC)) | |
1066 os.chdir(dirname(DART_REPO_LOC)) | |
1067 p = subprocess.Popen('gclient config https://dart.googlecode.com/svn/' + | |
1068 'branches/bleeding_edge/deps/all.deps', | |
1069 stdout=subprocess.PIPE, stderr=subprocess.PIPE, | |
1070 shell=True) | |
1071 p.communicate() | |
1072 if continuous: | |
1073 while True: | |
1074 results_set = UpdateSetOfDoneCls() | |
1075 (is_interesting, interesting_rev_num) = runner.HasInterestingCode() | |
1076 if is_interesting: | |
1077 runner.RunTestSequence(interesting_rev_num) | |
1078 else: | |
1079 if runner.backfill: | |
1080 results_set = FillInBackHistory(results_set, runner) | |
1081 else: | |
1082 time.sleep(200) | |
1083 else: | |
1084 runner.RunTestSequence() | |
1085 | |
1086 if __name__ == '__main__': | |
1087 main() | |
OLD | NEW |