# Copyright 2015 ARM Limited # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # pylint: disable=attribute-defined-outside-init import os import re import csv import math import shutil import json from zipfile import is_zipfile, ZipFile from collections import defaultdict try: import pandas as pd except ImportError: pd = None from wlauto import Workload, Parameter from wlauto.exceptions import WorkloadError, ConfigError from wlauto.utils.misc import check_output, get_null, get_meansd from wlauto.utils.types import numeric, identifier RESULT_REGEX = re.compile(r'RESULT ([^:]+): ([^=]+)\s*=\s*' # preamble and test/metric name r'(\[([^\]]+)\]|(\S+))' # value r'\s*(\S+)') # units TRACE_REGEX = re.compile(r'Trace saved as ([^\n]+)') # Trace event that signifies rendition of a Frame FRAME_EVENT = 'SwapBuffersLatency' class Telemetry(Workload): name = 'telemetry' description = """ Executes Google's Telemetery benchmarking framework (must be installed). Url: https://www.chromium.org/developers/telemetry From the web site: Telemetry is Chrome's performance testing framework. It allows you to perform arbitrary actions on a set of web pages and report metrics about it. The framework abstracts: - Launching a browser with arbitrary flags on any platform. - Opening a tab and navigating to the page under test. - Fetching data via the Inspector timeline and traces. - Using Web Page Replay to cache real-world websites so they don't change when used in benchmarks. Design Principles - Write one performance test that runs on all platforms - Windows, Mac, Linux, Chrome OS, and Android for both Chrome and ContentShell. - Runs on browser binaries, without a full Chromium checkout, and without having to build the browser yourself. - Use WebPageReplay to get repeatable test results. - Clean architecture for writing benchmarks that keeps measurements and use cases separate. - Run on non-Chrome browsers for comparative studies. This instrument runs telemetry via its ``run_benchmark`` script (which must be in PATH or specified using ``run_benchmark_path`` parameter) and parses metrics from the resulting output. **device setup** The device setup will depend on whether you're running a test image (in which case little or no setup should be necessary) """ parameters = [ Parameter('run_benchmark_path', default='run_benchmark', description=""" This is the path to run_benchmark script which runs a Telemetry benchmark. If not specified, the assumption will be that it is in path (i.e. with be invoked as ``run_benchmark``). """), Parameter('test', default='page_cycler.top_10_mobile', description=""" Specifies with of the the telemetry tests is to be run. """), Parameter('run_benchmark_params', default='', description=""" Additional paramters to be passed to ``run_benchmark``. """), Parameter('run_timeout', kind=int, default=900, description=""" Timeout for execution of the test. """), Parameter('extract_fps', kind=bool, default=False, description=""" if ``True``, FPS for the run will be computed from the trace (must be enabled). """), ] def validate(self): ret = os.system('{} > {} 2>&1'.format(self.run_benchmark_path, get_null())) if ret > 255: pass # telemetry found and appears to be installed properly. elif ret == 127: raise WorkloadError('run_benchmark not found (did you specify correct run_benchmark_path?)') else: raise WorkloadError('Unexected error from run_benchmark: {}'.format(ret)) if self.extract_fps and 'trace' not in self.run_benchmark_params: raise ConfigError('"trace" profiler must be enabled in order to extract FPS for Telemetry') def setup(self, context): self.raw_output = None self.command = self.build_command() def run(self, context): self.logger.debug(self.command) self.raw_output, _ = check_output(self.command, shell=True, timeout=self.run_timeout, ignore=1) def update_result(self, context): # pylint: disable=too-many-locals if not self.raw_output: self.logger.warning('Did not get run_benchmark output.') return raw_outfile = os.path.join(context.output_directory, 'telemetry_raw.out') with open(raw_outfile, 'w') as wfh: wfh.write(self.raw_output) context.add_artifact('telemetry-raw', raw_outfile, kind='raw') results, artifacts = parse_telemetry_results(raw_outfile) csv_outfile = os.path.join(context.output_directory, 'telemetry.csv') averages = defaultdict(list) with open(csv_outfile, 'wb') as wfh: writer = csv.writer(wfh) writer.writerow(['kind', 'url', 'iteration', 'value', 'units']) for result in results: name_template = identifier('{}_{}_{{}}'.format(result.url, result.kind)) averages[result.kind].append(result.average) context.result.add_metric(name_template.format('avg'), result.average, result.units, lower_is_better=True) context.result.add_metric(name_template.format('sd'), result.std, result.units, lower_is_better=True) writer.writerows(result.rows) context.add_artifact('telemetry', csv_outfile, kind='data') for kind, values in averages.iteritems(): context.result.add_metric(kind, special_average(values), lower_is_better=True) for idx, artifact in enumerate(artifacts): if is_zipfile(artifact): zf = ZipFile(artifact) for item in zf.infolist(): zf.extract(item, context.output_directory) zf.close() context.add_artifact('telemetry_trace_{}'.format(idx), path=item.filename, kind='data') else: # not a zip archive wa_path = os.path.join(context.output_directory, os.path.basename(artifact)) shutil.copy(artifact, wa_path) context.add_artifact('telemetry_artifact_{}'.format(idx), path=wa_path, kind='data') if self.extract_fps: self.logger.debug('Extracting FPS...') _extract_fps(context) def teardown(self, context): pass def build_command(self): device_opts = '' if self.device.platform == 'chromeos': if '--remote' not in self.run_benchmark_params: device_opts += '--remote={} '.format(self.device.host) if '--browser' not in self.run_benchmark_params: device_opts += '--browser=cros-chrome ' elif self.device.platform == 'android': if '--device' not in self.run_benchmark_params and self.device.adb_name: device_opts += '--device={} '.format(self.device.adb_name) if '--browser' not in self.run_benchmark_params: device_opts += '--browser=android-webview-shell ' else: raise WorkloadError('Currently, telemetry workload supports only ChromeOS or Android devices.') return '{} {} {} {}'.format(self.run_benchmark_path, self.test, device_opts, self.run_benchmark_params) def _extract_fps(context): trace_files = [a.path for a in context.iteration_artifacts if a.name.startswith('telemetry_trace_')] for tf in trace_files: name = os.path.splitext(os.path.basename(tf))[0] fps_file = os.path.join(context.output_directory, name + '-fps.csv') with open(tf) as fh: data = json.load(fh) events = pd.Series([e['ts'] for e in data['traceEvents'] if FRAME_EVENT == e['name']]) fps = (1000000 / (events - events.shift(1))) fps.index = events df = fps.dropna().reset_index() df.columns = ['timestamp', 'fps'] with open(fps_file, 'w') as wfh: df.to_csv(wfh, index=False) context.add_artifact('{}_fps'.format(name), fps_file, kind='data') context.result.add_metric('{} FPS'.format(name), df.fps.mean(), units='fps') context.result.add_metric('{} FPS (std)'.format(name), df.fps.std(), units='fps', lower_is_better=True) class TelemetryResult(object): @property def average(self): return get_meansd(self.values)[0] @property def std(self): return get_meansd(self.values)[1] @property def rows(self): for i, v in enumerate(self.values): yield [self.kind, self.url, i, v, self.units] def __init__(self, kind=None, url=None, values=None, units=None): self.kind = kind self.url = url self.values = values or [] self.units = units def __str__(self): return 'TR({kind},{url},{values},{units})'.format(**self.__dict__) __repr__ = __str__ def parse_telemetry_results(filepath): results = [] artifacts = [] with open(filepath) as fh: for line in fh: match = RESULT_REGEX.search(line) if match: result = TelemetryResult() result.kind = match.group(1) result.url = match.group(2) if match.group(4): result.values = map(numeric, match.group(4).split(',')) else: result.values = [numeric(match.group(5))] result.units = match.group(6) results.append(result) match = TRACE_REGEX.search(line) if match: artifacts.append(match.group(1)) return results, artifacts def special_average(values): """Overall score calculation. Tries to accound for large differences between different pages.""" negs = [v < 0 for v in values] abs_logs = [(av and math.log(av, 10) or av) for av in map(abs, values)] signed_logs = [] for lv, n in zip(abs_logs, negs): if n: signed_logs.append(-lv) else: signed_logs.append(lv) return get_meansd(signed_logs)[0] if __name__ == '__main__': import sys from pprint import pprint path = sys.argv[1] pprint(parse_telemetry_results(path))