Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 1 | #!/usr/bin/python |
| 2 | |
Yunlian Jiang | 00cc30e | 2013-03-28 13:23:57 -0700 | [diff] [blame] | 3 | # Copyright (c) 2013 The Chromium OS Authors. All rights reserved. |
| 4 | # Use of this source code is governed by a BSD-style license that can be |
| 5 | # found in the LICENSE file. |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 6 | |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 7 | """The experiment setting module.""" |
| 8 | |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 9 | import os |
| 10 | import time |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 11 | |
cmtice | e5bc63b | 2015-05-27 16:59:37 -0700 | [diff] [blame] | 12 | import afe_lock_machine |
| 13 | |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 14 | from utils import logger |
Yunlian Jiang | 00cc30e | 2013-03-28 13:23:57 -0700 | [diff] [blame] | 15 | from utils import misc |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 16 | |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 17 | from benchmark_run import BenchmarkRun |
| 18 | from machine_manager import MachineManager |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 19 | from machine_manager import MockMachineManager |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 20 | import test_flag |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 21 | |
| 22 | |
| 23 | class Experiment(object): |
| 24 | """Class representing an Experiment to be run.""" |
| 25 | |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 26 | def __init__(self, name, remote, working_directory, |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 27 | chromeos_root, cache_conditions, labels, benchmarks, |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 28 | experiment_file, email_to, acquire_timeout, log_dir, |
cmtice | 5c09fc2 | 2015-04-22 09:25:53 -0700 | [diff] [blame] | 29 | log_level, share_cache, results_directory, locks_directory): |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 30 | self.name = name |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 31 | self.working_directory = working_directory |
| 32 | self.remote = remote |
| 33 | self.chromeos_root = chromeos_root |
| 34 | self.cache_conditions = cache_conditions |
| 35 | self.experiment_file = experiment_file |
Ahmad Sharif | f395c26 | 2012-10-09 17:48:09 -0700 | [diff] [blame] | 36 | self.email_to = email_to |
Yunlian Jiang | 00cc30e | 2013-03-28 13:23:57 -0700 | [diff] [blame] | 37 | if not results_directory: |
| 38 | self.results_directory = os.path.join(self.working_directory, |
| 39 | self.name + "_results") |
| 40 | else: |
| 41 | self.results_directory = misc.CanonicalizePath(results_directory) |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 42 | self.log_dir = log_dir |
cmtice | 1390924 | 2014-03-11 13:38:07 -0700 | [diff] [blame] | 43 | self.log_level = log_level |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 44 | self.labels = labels |
| 45 | self.benchmarks = benchmarks |
| 46 | self.num_complete = 0 |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 47 | self.num_run_complete = 0 |
cmtice | 1a22436 | 2014-10-16 15:49:56 -0700 | [diff] [blame] | 48 | self.share_cache = share_cache |
cmtice | 517dc98 | 2015-06-12 12:22:32 -0700 | [diff] [blame] | 49 | # If locks_directory (self.lock_dir) not blank, we will use the file |
| 50 | # locking mechanism; if it is blank then we will use the AFE server |
| 51 | # locking mechanism. |
| 52 | self.locks_dir = locks_directory |
cmtice | f3eb803 | 2015-07-27 13:55:52 -0700 | [diff] [blame] | 53 | self.locked_machines = [] |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 54 | |
| 55 | # We need one chromeos_root to run the benchmarks in, but it doesn't |
| 56 | # matter where it is, unless the ABIs are different. |
| 57 | if not chromeos_root: |
| 58 | for label in self.labels: |
| 59 | if label.chromeos_root: |
| 60 | chromeos_root = label.chromeos_root |
| 61 | if not chromeos_root: |
| 62 | raise Exception("No chromeos_root given and could not determine one from " |
| 63 | "the image path.") |
| 64 | |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 65 | if test_flag.GetTestMode(): |
cmtice | 1390924 | 2014-03-11 13:38:07 -0700 | [diff] [blame] | 66 | self.machine_manager = MockMachineManager(chromeos_root, acquire_timeout, |
cmtice | d96e457 | 2015-05-19 16:19:25 -0700 | [diff] [blame] | 67 | log_level, locks_directory) |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 68 | else: |
cmtice | 1390924 | 2014-03-11 13:38:07 -0700 | [diff] [blame] | 69 | self.machine_manager = MachineManager(chromeos_root, acquire_timeout, |
cmtice | 517dc98 | 2015-06-12 12:22:32 -0700 | [diff] [blame] | 70 | log_level, locks_directory) |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 71 | self.l = logger.GetLogger(log_dir) |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 72 | |
| 73 | for machine in remote: |
| 74 | self.machine_manager.AddMachine(machine) |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 75 | for label in labels: |
| 76 | self.machine_manager.ComputeCommonCheckSum(label) |
| 77 | self.machine_manager.ComputeCommonCheckSumString(label) |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 78 | |
| 79 | self.start_time = None |
| 80 | self.benchmark_runs = self._GenerateBenchmarkRuns() |
| 81 | |
| 82 | def _GenerateBenchmarkRuns(self): |
| 83 | """Generate benchmark runs from labels and benchmark defintions.""" |
| 84 | benchmark_runs = [] |
| 85 | for label in self.labels: |
| 86 | for benchmark in self.benchmarks: |
| 87 | for iteration in range(1, benchmark.iterations + 1): |
| 88 | |
| 89 | benchmark_run_name = "%s: %s (%s)" % (label.name, benchmark.name, |
| 90 | iteration) |
| 91 | full_name = "%s_%s_%s" % (label.name, benchmark.name, iteration) |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 92 | logger_to_use = logger.Logger(self.log_dir, |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 93 | "run.%s" % (full_name), |
cmtice | 7789294 | 2014-03-18 13:47:17 -0700 | [diff] [blame] | 94 | True) |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 95 | benchmark_run = BenchmarkRun(benchmark_run_name, |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 96 | benchmark, |
| 97 | label, |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 98 | iteration, |
| 99 | self.cache_conditions, |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 100 | self.machine_manager, |
Luis Lozano | f81680c | 2013-03-15 14:44:13 -0700 | [diff] [blame] | 101 | logger_to_use, |
cmtice | 1390924 | 2014-03-11 13:38:07 -0700 | [diff] [blame] | 102 | self.log_level, |
cmtice | 1a22436 | 2014-10-16 15:49:56 -0700 | [diff] [blame] | 103 | self.share_cache) |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 104 | |
| 105 | benchmark_runs.append(benchmark_run) |
| 106 | return benchmark_runs |
| 107 | |
| 108 | def Build(self): |
| 109 | pass |
| 110 | |
| 111 | def Terminate(self): |
| 112 | for t in self.benchmark_runs: |
| 113 | if t.isAlive(): |
| 114 | self.l.LogError("Terminating run: '%s'." % t.name) |
| 115 | t.Terminate() |
| 116 | |
| 117 | def IsComplete(self): |
| 118 | if self.active_threads: |
| 119 | for t in self.active_threads: |
| 120 | if t.isAlive(): |
| 121 | t.join(0) |
| 122 | if not t.isAlive(): |
| 123 | self.num_complete += 1 |
Ahmad Sharif | 4467f00 | 2012-12-20 12:09:49 -0800 | [diff] [blame] | 124 | if not t.cache_hit: |
| 125 | self.num_run_complete += 1 |
Ahmad Sharif | 0dcbc4b | 2012-02-02 16:37:18 -0800 | [diff] [blame] | 126 | self.active_threads.remove(t) |
| 127 | return False |
| 128 | return True |
| 129 | |
| 130 | def Run(self): |
| 131 | self.start_time = time.time() |
| 132 | self.active_threads = [] |
| 133 | for benchmark_run in self.benchmark_runs: |
| 134 | # Set threads to daemon so program exits when ctrl-c is pressed. |
| 135 | benchmark_run.daemon = True |
| 136 | benchmark_run.start() |
| 137 | self.active_threads.append(benchmark_run) |
| 138 | |
| 139 | def SetCacheConditions(self, cache_conditions): |
| 140 | for benchmark_run in self.benchmark_runs: |
| 141 | benchmark_run.SetCacheConditions(cache_conditions) |
| 142 | |
| 143 | def Cleanup(self): |
cmtice | e5bc63b | 2015-05-27 16:59:37 -0700 | [diff] [blame] | 144 | """Make sure all machines are unlocked.""" |
cmtice | 517dc98 | 2015-06-12 12:22:32 -0700 | [diff] [blame] | 145 | if self.locks_dir: |
| 146 | # We are using the file locks mechanism, so call machine_manager.Cleanup |
| 147 | # to unlock everything. |
| 148 | self.machine_manager.Cleanup() |
| 149 | else: |
cmtice | f3eb803 | 2015-07-27 13:55:52 -0700 | [diff] [blame] | 150 | all_machines = self.locked_machines |
| 151 | if not all_machines: |
| 152 | return |
| 153 | |
| 154 | # If we locked any machines earlier, make sure we unlock them now. |
cmtice | 517dc98 | 2015-06-12 12:22:32 -0700 | [diff] [blame] | 155 | lock_mgr = afe_lock_machine.AFELockManager(all_machines, "", |
| 156 | self.labels[0].chromeos_root, |
| 157 | None) |
| 158 | machine_states = lock_mgr.GetMachineStates("unlock") |
| 159 | for k, state in machine_states.iteritems(): |
| 160 | if state["locked"]: |
| 161 | lock_mgr.UpdateLockInAFE(False, k) |