Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 1 | #!/usr/bin/env python2.7 |
| 2 | # Copyright 2016, Google Inc. |
| 3 | # All rights reserved. |
| 4 | # |
| 5 | # Redistribution and use in source and binary forms, with or without |
| 6 | # modification, are permitted provided that the following conditions are |
| 7 | # met: |
| 8 | # |
| 9 | # * Redistributions of source code must retain the above copyright |
| 10 | # notice, this list of conditions and the following disclaimer. |
| 11 | # * Redistributions in binary form must reproduce the above |
| 12 | # copyright notice, this list of conditions and the following disclaimer |
| 13 | # in the documentation and/or other materials provided with the |
| 14 | # distribution. |
| 15 | # * Neither the name of Google Inc. nor the names of its |
| 16 | # contributors may be used to endorse or promote products derived from |
| 17 | # this software without specific prior written permission. |
| 18 | # |
| 19 | # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 20 | # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 21 | # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 22 | # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 23 | # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 24 | # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 25 | # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 26 | # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 27 | # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 28 | # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 29 | # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 30 | |
| 31 | """Run performance tests locally or remotely.""" |
| 32 | |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 33 | from __future__ import print_function |
| 34 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 35 | import argparse |
Craig Tiller | accf16b | 2016-09-15 09:08:32 -0700 | [diff] [blame] | 36 | import collections |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 37 | import itertools |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 38 | import jobset |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 39 | import json |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 40 | import multiprocessing |
| 41 | import os |
Craig Tiller | accf16b | 2016-09-15 09:08:32 -0700 | [diff] [blame] | 42 | import performance.scenario_config as scenario_config |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 43 | import pipes |
Jan Tattermusch | 38becc2 | 2016-04-14 08:00:35 -0700 | [diff] [blame] | 44 | import re |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 45 | import report_utils |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 46 | import subprocess |
| 47 | import sys |
| 48 | import tempfile |
| 49 | import time |
Jan Tattermusch | ee9032c | 2016-04-14 08:35:51 -0700 | [diff] [blame] | 50 | import traceback |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 51 | import uuid |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 52 | import report_utils |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 53 | |
| 54 | |
| 55 | _ROOT = os.path.abspath(os.path.join(os.path.dirname(sys.argv[0]), '../..')) |
| 56 | os.chdir(_ROOT) |
| 57 | |
| 58 | |
| 59 | _REMOTE_HOST_USERNAME = 'jenkins' |
| 60 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 61 | _PERF_REPORT_OUTPUT_DIR = 'perf_reports' |
| 62 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 63 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 64 | class QpsWorkerJob: |
| 65 | """Encapsulates a qps worker server job.""" |
| 66 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 67 | def __init__(self, spec, language, host_and_port, perf_file_base_name=None): |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 68 | self._spec = spec |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 69 | self.language = language |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 70 | self.host_and_port = host_and_port |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 71 | self._job = None |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 72 | self.perf_file_base_name = perf_file_base_name |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 73 | |
| 74 | def start(self): |
Craig Tiller | c197ec1 | 2016-09-15 09:19:33 -0700 | [diff] [blame] | 75 | self._job = jobset.Job(self._spec, newline_on_success=True, travis=True, add_env={}) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 76 | |
| 77 | def is_running(self): |
| 78 | """Polls a job and returns True if given job is still running.""" |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 79 | return self._job and self._job.state() == jobset._RUNNING |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 80 | |
| 81 | def kill(self): |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 82 | if self._job: |
| 83 | self._job.kill() |
| 84 | self._job = None |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 85 | |
| 86 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 87 | def create_qpsworker_job(language, shortname=None, port=10000, remote_host=None, perf_cmd=None): |
| 88 | cmdline = (language.worker_cmdline() + ['--driver_port=%s' % port]) |
| 89 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 90 | if remote_host: |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 91 | host_and_port='%s:%s' % (remote_host, port) |
| 92 | else: |
| 93 | host_and_port='localhost:%s' % port |
| 94 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 95 | perf_file_base_name = None |
| 96 | if perf_cmd: |
| 97 | perf_file_base_name = '%s-%s' % (host_and_port, shortname) |
| 98 | # specify -o output file so perf.data gets collected when worker stopped |
| 99 | cmdline = perf_cmd + ['-o', '%s-perf.data' % perf_file_base_name] + cmdline |
| 100 | |
| 101 | if remote_host: |
| 102 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, remote_host) |
| 103 | ssh_cmd = ['ssh'] |
| 104 | ssh_cmd.extend([str(user_at_host), 'cd ~/performance_workspace/grpc/ && %s' % ' '.join(cmdline)]) |
| 105 | cmdline = ssh_cmd |
| 106 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 107 | jobspec = jobset.JobSpec( |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 108 | cmdline=cmdline, |
| 109 | shortname=shortname, |
Jan Tattermusch | 447548b | 2016-10-17 12:04:56 +0200 | [diff] [blame] | 110 | timeout_seconds=5*60, # workers get restarted after each scenario |
| 111 | verbose_success=True) |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 112 | return QpsWorkerJob(jobspec, language, host_and_port, perf_file_base_name) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 113 | |
| 114 | |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 115 | def create_scenario_jobspec(scenario_json, workers, remote_host=None, |
| 116 | bq_result_table=None): |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 117 | """Runs one scenario using QPS driver.""" |
| 118 | # setting QPS_WORKERS env variable here makes sure it works with SSH too. |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 119 | cmd = 'QPS_WORKERS="%s" ' % ','.join(workers) |
| 120 | if bq_result_table: |
| 121 | cmd += 'BQ_RESULT_TABLE="%s" ' % bq_result_table |
| 122 | cmd += 'tools/run_tests/performance/run_qps_driver.sh ' |
| 123 | cmd += '--scenarios_json=%s ' % pipes.quote(json.dumps({'scenarios': [scenario_json]})) |
Vijay Pai | c23d33b | 2016-07-19 11:19:12 -0700 | [diff] [blame] | 124 | cmd += '--scenario_result_file=scenario_result.json' |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 125 | if remote_host: |
| 126 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, remote_host) |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 127 | cmd = 'ssh %s "cd ~/performance_workspace/grpc/ && "%s' % (user_at_host, pipes.quote(cmd)) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 128 | |
| 129 | return jobset.JobSpec( |
| 130 | cmdline=[cmd], |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 131 | shortname='qps_json_driver.%s' % scenario_json['name'], |
| 132 | timeout_seconds=3*60, |
| 133 | shell=True, |
| 134 | verbose_success=True) |
| 135 | |
| 136 | |
| 137 | def create_quit_jobspec(workers, remote_host=None): |
| 138 | """Runs quit using QPS driver.""" |
| 139 | # setting QPS_WORKERS env variable here makes sure it works with SSH too. |
Craig Tiller | 025972d | 2016-09-15 09:26:50 -0700 | [diff] [blame] | 140 | cmd = 'QPS_WORKERS="%s" bins/opt/qps_json_driver --quit' % ','.join(w.host_and_port for w in workers) |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 141 | if remote_host: |
| 142 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, remote_host) |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 143 | cmd = 'ssh %s "cd ~/performance_workspace/grpc/ && "%s' % (user_at_host, pipes.quote(cmd)) |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 144 | |
| 145 | return jobset.JobSpec( |
| 146 | cmdline=[cmd], |
vjpai | 29089c7 | 2016-04-20 12:38:16 -0700 | [diff] [blame] | 147 | shortname='qps_json_driver.quit', |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 148 | timeout_seconds=3*60, |
| 149 | shell=True, |
| 150 | verbose_success=True) |
| 151 | |
| 152 | |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 153 | def create_netperf_jobspec(server_host='localhost', client_host=None, |
| 154 | bq_result_table=None): |
| 155 | """Runs netperf benchmark.""" |
| 156 | cmd = 'NETPERF_SERVER_HOST="%s" ' % server_host |
| 157 | if bq_result_table: |
| 158 | cmd += 'BQ_RESULT_TABLE="%s" ' % bq_result_table |
Jan Tattermusch | ad17bf7 | 2016-05-11 12:41:37 -0700 | [diff] [blame] | 159 | if client_host: |
| 160 | # If netperf is running remotely, the env variables populated by Jenkins |
| 161 | # won't be available on the client, but we need them for uploading results |
| 162 | # to BigQuery. |
| 163 | jenkins_job_name = os.getenv('JOB_NAME') |
| 164 | if jenkins_job_name: |
| 165 | cmd += 'JOB_NAME="%s" ' % jenkins_job_name |
| 166 | jenkins_build_number = os.getenv('BUILD_NUMBER') |
| 167 | if jenkins_build_number: |
| 168 | cmd += 'BUILD_NUMBER="%s" ' % jenkins_build_number |
| 169 | |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 170 | cmd += 'tools/run_tests/performance/run_netperf.sh' |
| 171 | if client_host: |
| 172 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, client_host) |
| 173 | cmd = 'ssh %s "cd ~/performance_workspace/grpc/ && "%s' % (user_at_host, pipes.quote(cmd)) |
| 174 | |
| 175 | return jobset.JobSpec( |
| 176 | cmdline=[cmd], |
| 177 | shortname='netperf', |
| 178 | timeout_seconds=60, |
| 179 | shell=True, |
| 180 | verbose_success=True) |
| 181 | |
| 182 | |
Jan Tattermusch | de874a1 | 2016-04-18 09:21:37 -0700 | [diff] [blame] | 183 | def archive_repo(languages): |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 184 | """Archives local version of repo including submodules.""" |
Jan Tattermusch | de874a1 | 2016-04-18 09:21:37 -0700 | [diff] [blame] | 185 | cmdline=['tar', '-cf', '../grpc.tar', '../grpc/'] |
| 186 | if 'java' in languages: |
| 187 | cmdline.append('../grpc-java') |
| 188 | if 'go' in languages: |
| 189 | cmdline.append('../grpc-go') |
| 190 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 191 | archive_job = jobset.JobSpec( |
Jan Tattermusch | de874a1 | 2016-04-18 09:21:37 -0700 | [diff] [blame] | 192 | cmdline=cmdline, |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 193 | shortname='archive_repo', |
| 194 | timeout_seconds=3*60) |
| 195 | |
| 196 | jobset.message('START', 'Archiving local repository.', do_newline=True) |
| 197 | num_failures, _ = jobset.run( |
| 198 | [archive_job], newline_on_success=True, maxjobs=1) |
| 199 | if num_failures == 0: |
| 200 | jobset.message('SUCCESS', |
Jan Tattermusch | de874a1 | 2016-04-18 09:21:37 -0700 | [diff] [blame] | 201 | 'Archive with local repository created successfully.', |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 202 | do_newline=True) |
| 203 | else: |
| 204 | jobset.message('FAILED', 'Failed to archive local repository.', |
| 205 | do_newline=True) |
| 206 | sys.exit(1) |
| 207 | |
| 208 | |
Jan Tattermusch | 1408920 | 2016-04-27 17:55:27 -0700 | [diff] [blame] | 209 | def prepare_remote_hosts(hosts, prepare_local=False): |
| 210 | """Prepares remote hosts (and maybe prepare localhost as well).""" |
| 211 | prepare_timeout = 5*60 |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 212 | prepare_jobs = [] |
| 213 | for host in hosts: |
| 214 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, host) |
| 215 | prepare_jobs.append( |
| 216 | jobset.JobSpec( |
| 217 | cmdline=['tools/run_tests/performance/remote_host_prepare.sh'], |
| 218 | shortname='remote_host_prepare.%s' % host, |
| 219 | environ = {'USER_AT_HOST': user_at_host}, |
Jan Tattermusch | 1408920 | 2016-04-27 17:55:27 -0700 | [diff] [blame] | 220 | timeout_seconds=prepare_timeout)) |
| 221 | if prepare_local: |
| 222 | # Prepare localhost as well |
| 223 | prepare_jobs.append( |
| 224 | jobset.JobSpec( |
| 225 | cmdline=['tools/run_tests/performance/kill_workers.sh'], |
| 226 | shortname='local_prepare', |
| 227 | timeout_seconds=prepare_timeout)) |
| 228 | jobset.message('START', 'Preparing hosts.', do_newline=True) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 229 | num_failures, _ = jobset.run( |
| 230 | prepare_jobs, newline_on_success=True, maxjobs=10) |
| 231 | if num_failures == 0: |
| 232 | jobset.message('SUCCESS', |
Jan Tattermusch | 1408920 | 2016-04-27 17:55:27 -0700 | [diff] [blame] | 233 | 'Prepare step completed successfully.', |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 234 | do_newline=True) |
| 235 | else: |
| 236 | jobset.message('FAILED', 'Failed to prepare remote hosts.', |
| 237 | do_newline=True) |
| 238 | sys.exit(1) |
| 239 | |
| 240 | |
Craig Tiller | d92d5c5 | 2016-04-04 13:49:29 -0700 | [diff] [blame] | 241 | def build_on_remote_hosts(hosts, languages=scenario_config.LANGUAGES.keys(), build_local=False): |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 242 | """Builds performance worker on remote hosts (and maybe also locally).""" |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 243 | build_timeout = 15*60 |
| 244 | build_jobs = [] |
| 245 | for host in hosts: |
| 246 | user_at_host = '%s@%s' % (_REMOTE_HOST_USERNAME, host) |
| 247 | build_jobs.append( |
| 248 | jobset.JobSpec( |
Craig Tiller | 7797e3f | 2016-04-01 07:41:05 -0700 | [diff] [blame] | 249 | cmdline=['tools/run_tests/performance/remote_host_build.sh'] + languages, |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 250 | shortname='remote_host_build.%s' % host, |
| 251 | environ = {'USER_AT_HOST': user_at_host, 'CONFIG': 'opt'}, |
| 252 | timeout_seconds=build_timeout)) |
| 253 | if build_local: |
| 254 | # Build locally as well |
| 255 | build_jobs.append( |
| 256 | jobset.JobSpec( |
Craig Tiller | 7797e3f | 2016-04-01 07:41:05 -0700 | [diff] [blame] | 257 | cmdline=['tools/run_tests/performance/build_performance.sh'] + languages, |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 258 | shortname='local_build', |
| 259 | environ = {'CONFIG': 'opt'}, |
| 260 | timeout_seconds=build_timeout)) |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 261 | jobset.message('START', 'Building.', do_newline=True) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 262 | num_failures, _ = jobset.run( |
| 263 | build_jobs, newline_on_success=True, maxjobs=10) |
| 264 | if num_failures == 0: |
| 265 | jobset.message('SUCCESS', |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 266 | 'Built successfully.', |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 267 | do_newline=True) |
| 268 | else: |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 269 | jobset.message('FAILED', 'Build failed.', |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 270 | do_newline=True) |
| 271 | sys.exit(1) |
| 272 | |
| 273 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 274 | def create_qpsworkers(languages, worker_hosts, perf_cmd=None): |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 275 | """Creates QPS workers (but does not start them).""" |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 276 | if not worker_hosts: |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 277 | # run two workers locally (for each language) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 278 | workers=[(None, 10000), (None, 10010)] |
| 279 | elif len(worker_hosts) == 1: |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 280 | # run two workers on the remote host (for each language) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 281 | workers=[(worker_hosts[0], 10000), (worker_hosts[0], 10010)] |
| 282 | else: |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 283 | # run one worker per each remote host (for each language) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 284 | workers=[(worker_host, 10000) for worker_host in worker_hosts] |
| 285 | |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 286 | return [create_qpsworker_job(language, |
| 287 | shortname= 'qps_worker_%s_%s' % (language, |
| 288 | worker_idx), |
| 289 | port=worker[1] + language.worker_port_offset(), |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 290 | remote_host=worker[0], |
| 291 | perf_cmd=perf_cmd) |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 292 | for language in languages |
| 293 | for worker_idx, worker in enumerate(workers)] |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 294 | |
| 295 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 296 | def perf_report_processor_job(worker_host, perf_base_name, output_filename): |
| 297 | print('Creating perf report collection job for %s' % worker_host) |
| 298 | cmd = '' |
| 299 | if worker_host != 'localhost': |
| 300 | user_at_host = "%s@%s" % (_REMOTE_HOST_USERNAME, worker_host) |
| 301 | cmd = "USER_AT_HOST=%s OUTPUT_FILENAME=%s OUTPUT_DIR=%s PERF_BASE_NAME=%s\ |
| 302 | tools/run_tests/performance/process_remote_perf_flamegraphs.sh" \ |
| 303 | % (user_at_host, output_filename, _PERF_REPORT_OUTPUT_DIR, perf_base_name) |
| 304 | else: |
| 305 | cmd = "OUTPUT_FILENAME=%s OUTPUT_DIR=%s PERF_BASE_NAME=%s\ |
| 306 | tools/run_tests/performance/process_local_perf_flamegraphs.sh" \ |
| 307 | % (output_filename, _PERF_REPORT_OUTPUT_DIR, perf_base_name) |
| 308 | |
| 309 | return jobset.JobSpec(cmdline=cmd, |
| 310 | timeout_seconds=3*60, |
| 311 | shell=True, |
| 312 | verbose_success=True, |
| 313 | shortname='process perf report') |
| 314 | |
| 315 | |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 316 | Scenario = collections.namedtuple('Scenario', 'jobspec workers name') |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 317 | |
| 318 | |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 319 | def create_scenarios(languages, workers_by_lang, remote_host=None, regex='.*', |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 320 | category='all', bq_result_table=None, |
| 321 | netperf=False, netperf_hosts=[]): |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 322 | """Create jobspecs for scenarios to run.""" |
Ken Payson | 0482c10 | 2016-04-19 12:08:34 -0700 | [diff] [blame] | 323 | all_workers = [worker |
| 324 | for workers in workers_by_lang.values() |
| 325 | for worker in workers] |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 326 | scenarios = [] |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 327 | _NO_WORKERS = [] |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 328 | |
| 329 | if netperf: |
| 330 | if not netperf_hosts: |
| 331 | netperf_server='localhost' |
| 332 | netperf_client=None |
| 333 | elif len(netperf_hosts) == 1: |
| 334 | netperf_server=netperf_hosts[0] |
| 335 | netperf_client=netperf_hosts[0] |
| 336 | else: |
| 337 | netperf_server=netperf_hosts[0] |
| 338 | netperf_client=netperf_hosts[1] |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 339 | scenarios.append(Scenario( |
| 340 | create_netperf_jobspec(server_host=netperf_server, |
| 341 | client_host=netperf_client, |
| 342 | bq_result_table=bq_result_table), |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 343 | _NO_WORKERS, 'netperf')) |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 344 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 345 | for language in languages: |
Craig Tiller | 0bda0b3 | 2016-03-03 12:51:53 -0800 | [diff] [blame] | 346 | for scenario_json in language.scenarios(): |
Jan Tattermusch | 38becc2 | 2016-04-14 08:00:35 -0700 | [diff] [blame] | 347 | if re.search(args.regex, scenario_json['name']): |
Craig Tiller | b6df247 | 2016-09-13 09:41:26 -0700 | [diff] [blame] | 348 | categories = scenario_json.get('CATEGORIES', ['scalable', 'smoketest']) |
| 349 | if category in categories or category == 'all': |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 350 | workers = workers_by_lang[str(language)][:] |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 351 | # 'SERVER_LANGUAGE' is an indicator for this script to pick |
| 352 | # a server in different language. |
| 353 | custom_server_lang = scenario_json.get('SERVER_LANGUAGE', None) |
Jan Tattermusch | 37a907e | 2016-05-13 13:49:43 -0700 | [diff] [blame] | 354 | custom_client_lang = scenario_json.get('CLIENT_LANGUAGE', None) |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 355 | scenario_json = scenario_config.remove_nonproto_fields(scenario_json) |
Jan Tattermusch | 37a907e | 2016-05-13 13:49:43 -0700 | [diff] [blame] | 356 | if custom_server_lang and custom_client_lang: |
| 357 | raise Exception('Cannot set both custom CLIENT_LANGUAGE and SERVER_LANGUAGE' |
| 358 | 'in the same scenario') |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 359 | if custom_server_lang: |
| 360 | if not workers_by_lang.get(custom_server_lang, []): |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 361 | print('Warning: Skipping scenario %s as' % scenario_json['name']) |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 362 | print('SERVER_LANGUAGE is set to %s yet the language has ' |
| 363 | 'not been selected with -l' % custom_server_lang) |
| 364 | continue |
| 365 | for idx in range(0, scenario_json['num_servers']): |
| 366 | # replace first X workers by workers of a different language |
| 367 | workers[idx] = workers_by_lang[custom_server_lang][idx] |
Jan Tattermusch | 37a907e | 2016-05-13 13:49:43 -0700 | [diff] [blame] | 368 | if custom_client_lang: |
| 369 | if not workers_by_lang.get(custom_client_lang, []): |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 370 | print('Warning: Skipping scenario %s as' % scenario_json['name']) |
Jan Tattermusch | 37a907e | 2016-05-13 13:49:43 -0700 | [diff] [blame] | 371 | print('CLIENT_LANGUAGE is set to %s yet the language has ' |
| 372 | 'not been selected with -l' % custom_client_lang) |
| 373 | continue |
| 374 | for idx in range(scenario_json['num_servers'], len(workers)): |
| 375 | # replace all client workers by workers of a different language, |
| 376 | # leave num_server workers as they are server workers. |
| 377 | workers[idx] = workers_by_lang[custom_client_lang][idx] |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 378 | scenario = Scenario( |
| 379 | create_scenario_jobspec(scenario_json, |
| 380 | [w.host_and_port for w in workers], |
| 381 | remote_host=remote_host, |
| 382 | bq_result_table=bq_result_table), |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 383 | workers, |
| 384 | scenario_json['name']) |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 385 | scenarios.append(scenario) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 386 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 387 | return scenarios |
| 388 | |
| 389 | |
| 390 | def finish_qps_workers(jobs): |
| 391 | """Waits for given jobs to finish and eventually kills them.""" |
| 392 | retries = 0 |
Alexander Polcyn | 4979667 | 2016-10-17 10:01:37 -0700 | [diff] [blame] | 393 | num_killed = 0 |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 394 | while any(job.is_running() for job in jobs): |
| 395 | for job in qpsworker_jobs: |
| 396 | if job.is_running(): |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 397 | print('QPS worker "%s" is still running.' % job.host_and_port) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 398 | if retries > 10: |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 399 | print('Killing all QPS workers.') |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 400 | for job in jobs: |
| 401 | job.kill() |
Alexander Polcyn | 4979667 | 2016-10-17 10:01:37 -0700 | [diff] [blame] | 402 | num_killed += 1 |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 403 | retries += 1 |
| 404 | time.sleep(3) |
siddharthshukla | 0589e53 | 2016-07-07 16:08:01 +0200 | [diff] [blame] | 405 | print('All QPS workers finished.') |
Alexander Polcyn | 4979667 | 2016-10-17 10:01:37 -0700 | [diff] [blame] | 406 | return num_killed |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 407 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 408 | profile_output_files = [] |
| 409 | |
| 410 | # Collect perf text reports and flamegraphs if perf_cmd was used |
| 411 | # Note the base names of perf text reports are used when creating and processing |
| 412 | # perf data. The scenario name uniqifies the output name in the final |
| 413 | # perf reports directory. |
| 414 | # Alos, the perf profiles need to be fetched and processed after each scenario |
| 415 | # in order to avoid clobbering the output files. |
| 416 | def run_collect_perf_profile_jobs(hosts_and_base_names, scenario_name): |
| 417 | perf_report_jobs = [] |
| 418 | global profile_output_files |
| 419 | for host_and_port in hosts_and_base_names: |
| 420 | perf_base_name = hosts_and_base_names[host_and_port] |
| 421 | output_filename = '%s-%s' % (scenario_name, perf_base_name) |
| 422 | # from the base filename, create .svg output filename |
| 423 | host = host_and_port.split(':')[0] |
| 424 | profile_output_files.append('%s.svg' % output_filename) |
| 425 | perf_report_jobs.append(perf_report_processor_job(host, perf_base_name, output_filename)) |
| 426 | |
| 427 | jobset.message('START', 'Collecting perf reports from qps workers', do_newline=True) |
| 428 | failures, _ = jobset.run(perf_report_jobs, newline_on_success=True, maxjobs=1) |
| 429 | jobset.message('END', 'Collecting perf reports from qps workers', do_newline=True) |
| 430 | return failures |
| 431 | |
| 432 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 433 | argp = argparse.ArgumentParser(description='Run performance tests.') |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 434 | argp.add_argument('-l', '--language', |
Craig Tiller | d92d5c5 | 2016-04-04 13:49:29 -0700 | [diff] [blame] | 435 | choices=['all'] + sorted(scenario_config.LANGUAGES.keys()), |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 436 | nargs='+', |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 437 | required=True, |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 438 | help='Languages to benchmark.') |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 439 | argp.add_argument('--remote_driver_host', |
| 440 | default=None, |
| 441 | help='Run QPS driver on given host. By default, QPS driver is run locally.') |
| 442 | argp.add_argument('--remote_worker_host', |
| 443 | nargs='+', |
| 444 | default=[], |
| 445 | help='Worker hosts where to start QPS workers.') |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 446 | argp.add_argument('--dry_run', |
| 447 | default=False, |
| 448 | action='store_const', |
| 449 | const=True, |
| 450 | help='Just list scenarios to be run, but don\'t run them.') |
Jan Tattermusch | 38becc2 | 2016-04-14 08:00:35 -0700 | [diff] [blame] | 451 | argp.add_argument('-r', '--regex', default='.*', type=str, |
| 452 | help='Regex to select scenarios to run.') |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 453 | argp.add_argument('--bq_result_table', default=None, type=str, |
| 454 | help='Bigquery "dataset.table" to upload results to.') |
Jan Tattermusch | 427699b | 2016-05-05 18:10:14 -0700 | [diff] [blame] | 455 | argp.add_argument('--category', |
Craig Tiller | 6388da5 | 2016-09-07 17:06:29 -0700 | [diff] [blame] | 456 | choices=['smoketest','all','scalable','sweep'], |
Jan Tattermusch | 37a907e | 2016-05-13 13:49:43 -0700 | [diff] [blame] | 457 | default='all', |
| 458 | help='Select a category of tests to run.') |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 459 | argp.add_argument('--netperf', |
| 460 | default=False, |
| 461 | action='store_const', |
| 462 | const=True, |
| 463 | help='Run netperf benchmark as one of the scenarios.') |
Jan Tattermusch | 88818ae | 2016-11-18 14:21:33 +0100 | [diff] [blame] | 464 | argp.add_argument('-x', '--xml_report', default='report.xml', type=str, |
| 465 | help='Name of XML report file to generate.') |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 466 | argp.add_argument('--perf_args', |
| 467 | help=('Example usage: "--perf_args=record -F 99 -g". ' |
| 468 | 'Wrap QPS workers in a perf command ' |
| 469 | 'with the arguments to perf specified here. ' |
| 470 | '".svg" flame graph profiles will be ' |
| 471 | 'created for each Qps Worker on each scenario. ' |
| 472 | 'Files will output to "<repo_root>/perf_reports" ' |
| 473 | 'directory. Output files from running the worker ' |
| 474 | 'under perf are saved in the repo root where its ran. ' |
| 475 | 'Note that the perf "-g" flag is necessary for ' |
| 476 | 'flame graphs generation to work (assuming the binary ' |
| 477 | 'being profiled uses frame pointers, check out ' |
| 478 | '"--call-graph dwarf" option using libunwind otherwise.) ' |
| 479 | 'Also note that the entire "--perf_args=<arg(s)>" must ' |
| 480 | 'be wrapped in quotes as in the example usage. ' |
| 481 | 'If the "--perg_args" is unspecified, "perf" will ' |
| 482 | 'not be used at all. ' |
| 483 | 'See http://www.brendangregg.com/perf.html ' |
| 484 | 'for more general perf examples.')) |
| 485 | argp.add_argument('--skip_generate_flamegraphs', |
| 486 | default=False, |
| 487 | action='store_const', |
| 488 | const=True, |
| 489 | help=('Turn flame graph generation off. ' |
| 490 | 'May be useful if "perf_args" arguments do not make sense for ' |
| 491 | 'generating flamegraphs (e.g., "--perf_args=stat ...")')) |
| 492 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 493 | |
| 494 | args = argp.parse_args() |
| 495 | |
Craig Tiller | d92d5c5 | 2016-04-04 13:49:29 -0700 | [diff] [blame] | 496 | languages = set(scenario_config.LANGUAGES[l] |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 497 | for l in itertools.chain.from_iterable( |
Craig Tiller | d92d5c5 | 2016-04-04 13:49:29 -0700 | [diff] [blame] | 498 | scenario_config.LANGUAGES.iterkeys() if x == 'all' else [x] |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 499 | for x in args.language)) |
| 500 | |
Jan Tattermusch | 6d7fa55 | 2016-04-14 17:42:54 -0700 | [diff] [blame] | 501 | |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 502 | # Put together set of remote hosts where to run and build |
| 503 | remote_hosts = set() |
| 504 | if args.remote_worker_host: |
| 505 | for host in args.remote_worker_host: |
| 506 | remote_hosts.add(host) |
| 507 | if args.remote_driver_host: |
| 508 | remote_hosts.add(args.remote_driver_host) |
| 509 | |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 510 | if not args.dry_run: |
| 511 | if remote_hosts: |
| 512 | archive_repo(languages=[str(l) for l in languages]) |
| 513 | prepare_remote_hosts(remote_hosts, prepare_local=True) |
| 514 | else: |
| 515 | prepare_remote_hosts([], prepare_local=True) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 516 | |
| 517 | build_local = False |
| 518 | if not args.remote_driver_host: |
| 519 | build_local = True |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 520 | if not args.dry_run: |
| 521 | build_on_remote_hosts(remote_hosts, languages=[str(l) for l in languages], build_local=build_local) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 522 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 523 | perf_cmd = None |
| 524 | if args.perf_args: |
| 525 | # Expect /usr/bin/perf to be installed here, as is usual |
| 526 | perf_cmd = ['/usr/bin/perf'] |
| 527 | perf_cmd.extend(re.split('\s+', args.perf_args)) |
| 528 | |
| 529 | qpsworker_jobs = create_qpsworkers(languages, args.remote_worker_host, perf_cmd=perf_cmd) |
Jan Tattermusch | 38becc2 | 2016-04-14 08:00:35 -0700 | [diff] [blame] | 530 | |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 531 | # get list of worker addresses for each language. |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 532 | workers_by_lang = dict([(str(language), []) for language in languages]) |
Jan Tattermusch | bb1a453 | 2016-03-30 18:04:01 -0700 | [diff] [blame] | 533 | for job in qpsworker_jobs: |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 534 | workers_by_lang[str(job.language)].append(job) |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 535 | |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 536 | scenarios = create_scenarios(languages, |
Craig Tiller | d82fccc | 2016-09-15 09:15:23 -0700 | [diff] [blame] | 537 | workers_by_lang=workers_by_lang, |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 538 | remote_host=args.remote_driver_host, |
| 539 | regex=args.regex, |
| 540 | category=args.category, |
| 541 | bq_result_table=args.bq_result_table, |
| 542 | netperf=args.netperf, |
| 543 | netperf_hosts=args.remote_worker_host) |
Jan Tattermusch | 4de2c32 | 2016-05-10 14:33:07 -0700 | [diff] [blame] | 544 | |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 545 | if not scenarios: |
| 546 | raise Exception('No scenarios to run') |
Jan Tattermusch | b275844 | 2016-03-28 09:32:20 -0700 | [diff] [blame] | 547 | |
Alex Polcyn | cac93f6 | 2016-10-19 09:27:57 -0700 | [diff] [blame] | 548 | total_scenario_failures = 0 |
Alexander Polcyn | 898a2e9 | 2016-10-22 17:41:23 -0700 | [diff] [blame] | 549 | qps_workers_killed = 0 |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 550 | merged_resultset = {} |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 551 | perf_report_failures = 0 |
| 552 | |
Craig Tiller | c1b54f2 | 2016-09-15 08:57:14 -0700 | [diff] [blame] | 553 | for scenario in scenarios: |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 554 | if args.dry_run: |
| 555 | print(scenario.name) |
| 556 | else: |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 557 | scenario_failures = 0 |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 558 | try: |
| 559 | for worker in scenario.workers: |
| 560 | worker.start() |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 561 | scenario_failures, resultset = jobset.run([scenario.jobspec, |
| 562 | create_quit_jobspec(scenario.workers, remote_host=args.remote_driver_host)], |
| 563 | newline_on_success=True, maxjobs=1) |
Alex Polcyn | cac93f6 | 2016-10-19 09:27:57 -0700 | [diff] [blame] | 564 | total_scenario_failures += scenario_failures |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 565 | merged_resultset = dict(itertools.chain(merged_resultset.iteritems(), |
| 566 | resultset.iteritems())) |
Craig Tiller | 677966a | 2016-09-26 07:37:28 -0700 | [diff] [blame] | 567 | finally: |
Alexander Polcyn | 898a2e9 | 2016-10-22 17:41:23 -0700 | [diff] [blame] | 568 | # Consider qps workers that need to be killed as failures |
| 569 | qps_workers_killed += finish_qps_workers(scenario.workers) |
Alexander Polcyn | 4979667 | 2016-10-17 10:01:37 -0700 | [diff] [blame] | 570 | |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 571 | if perf_cmd and scenario_failures == 0 and not args.skip_generate_flamegraphs: |
| 572 | workers_and_base_names = {} |
| 573 | for worker in scenario.workers: |
| 574 | if not worker.perf_file_base_name: |
| 575 | raise Exception('using perf buf perf report filename is unspecified') |
| 576 | workers_and_base_names[worker.host_and_port] = worker.perf_file_base_name |
| 577 | perf_report_failures += run_collect_perf_profile_jobs(workers_and_base_names, scenario.name) |
| 578 | |
| 579 | |
| 580 | # Still write the index.html even if some scenarios failed. |
| 581 | # 'profile_output_files' will only have names for scenarios that passed |
| 582 | if perf_cmd and not args.skip_generate_flamegraphs: |
| 583 | # write the index fil to the output dir, with all profiles from all scenarios/workers |
| 584 | report_utils.render_perf_profiling_results('%s/index.html' % _PERF_REPORT_OUTPUT_DIR, profile_output_files) |
| 585 | |
| 586 | if total_scenario_failures > 0 or qps_workers_killed > 0: |
| 587 | print('%s scenarios failed and %s qps worker jobs killed' % (total_scenario_failures, qps_workers_killed)) |
| 588 | sys.exit(1) |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 589 | |
Jan Tattermusch | 88818ae | 2016-11-18 14:21:33 +0100 | [diff] [blame] | 590 | report_utils.render_junit_xml_report(merged_resultset, args.xml_report, |
Jan Tattermusch | 94d40cb | 2016-10-24 21:06:40 +0200 | [diff] [blame] | 591 | suite_name='benchmarks') |
Alexander Polcyn | 9f08d11 | 2016-10-24 12:25:02 -0700 | [diff] [blame^] | 592 | if perf_report_failures > 0: |
| 593 | print('%s perf profile collection jobs failed' % perf_report_failures) |
Alex Polcyn | cac93f6 | 2016-10-19 09:27:57 -0700 | [diff] [blame] | 594 | sys.exit(1) |