mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 1 | __author__ = """Copyright Andy Whitcroft, Martin J. Bligh - 2006, 2007""" |
| 2 | |
mbligh | 26f0d88 | 2009-06-22 18:30:01 +0000 | [diff] [blame^] | 3 | import sys, os, subprocess, traceback, time, signal, pickle, logging |
jadmanski | d93d7d2 | 2008-05-29 21:37:29 +0000 | [diff] [blame] | 4 | |
| 5 | from autotest_lib.client.common_lib import error, utils |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 6 | |
| 7 | |
showard | 75cdfee | 2009-06-10 17:40:41 +0000 | [diff] [blame] | 8 | # entry points that use subcommand must set this to their logging manager |
| 9 | # to get log redirection for subcommands |
| 10 | logging_manager_object = None |
| 11 | |
| 12 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 13 | def parallel(tasklist, timeout=None, return_results=False): |
mbligh | 415dc21 | 2009-06-15 21:53:34 +0000 | [diff] [blame] | 14 | """ |
| 15 | Run a set of predefined subcommands in parallel. |
| 16 | |
| 17 | @param tasklist: A list of subcommand instances to execute. |
| 18 | @param timeout: Number of seconds after which the commands should timeout. |
| 19 | @param return_results: If True instead of an AutoServError being raised |
| 20 | on any error a list of the results|exceptions from the tasks is |
| 21 | returned. [default: False] |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 22 | """ |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 23 | pids = [] |
| 24 | run_error = False |
| 25 | for task in tasklist: |
| 26 | task.fork_start() |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 27 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 28 | remaining_timeout = None |
| 29 | if timeout: |
| 30 | endtime = time.time() + timeout |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 31 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 32 | results = [] |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 33 | for task in tasklist: |
| 34 | if timeout: |
| 35 | remaining_timeout = max(endtime - time.time(), 1) |
| 36 | try: |
| 37 | status = task.fork_waitfor(remaining_timeout) |
| 38 | except error.AutoservSubcommandError: |
| 39 | run_error = True |
| 40 | else: |
| 41 | if status != 0: |
| 42 | run_error = True |
mbligh | 158ba7b | 2008-03-07 18:29:12 +0000 | [diff] [blame] | 43 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 44 | results.append(pickle.load(task.result_pickle)) |
| 45 | task.result_pickle.close() |
| 46 | |
| 47 | if return_results: |
| 48 | return results |
| 49 | elif run_error: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 50 | raise error.AutoservError('One or more subcommands failed') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 51 | |
| 52 | |
mbligh | 415dc21 | 2009-06-15 21:53:34 +0000 | [diff] [blame] | 53 | def parallel_simple(function, arglist, log=True, timeout=None, |
| 54 | return_results=False): |
| 55 | """ |
| 56 | Each element in the arglist used to create a subcommand object, |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 57 | where that arg is used both as a subdir name, and a single argument |
| 58 | to pass to "function". |
mbligh | dd3235b | 2008-01-14 16:44:19 +0000 | [diff] [blame] | 59 | |
mbligh | 415dc21 | 2009-06-15 21:53:34 +0000 | [diff] [blame] | 60 | We create a subcommand object for each element in the list, |
| 61 | then execute those subcommand objects in parallel. |
| 62 | |
| 63 | NOTE: As an optimization, if len(arglist) == 1 a subcommand is not used. |
| 64 | |
| 65 | @param function: A callable to run in parallel once per arg in arglist. |
| 66 | @param arglist: A list of single arguments to be used one per subcommand; |
| 67 | typically a list of machine names. |
| 68 | @param log: If True, output will be written to output in a subdirectory |
| 69 | named after each subcommand's arg. |
| 70 | @param timeout: Number of seconds after which the commands should timeout. |
| 71 | @param return_results: If True instead of an AutoServError being raised |
| 72 | on any error a list of the results|exceptions from the function |
| 73 | called on each arg is returned. [default: False] |
| 74 | |
| 75 | @returns None or a list of results/exceptions. |
| 76 | """ |
mbligh | 26f0d88 | 2009-06-22 18:30:01 +0000 | [diff] [blame^] | 77 | if not arglist: |
| 78 | logging.warn("parallel_simple was called with an empty arglist, " |
| 79 | "did you forget to pass in a list of machines?") |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 80 | # Bypass the multithreading if only one machine. |
mbligh | 415dc21 | 2009-06-15 21:53:34 +0000 | [diff] [blame] | 81 | if len(arglist) == 1: |
| 82 | arg = arglist[0] |
| 83 | if return_results: |
| 84 | try: |
| 85 | result = function(arg) |
| 86 | except Exception, e: |
| 87 | return [e] |
| 88 | return [result] |
| 89 | else: |
| 90 | function(arg) |
| 91 | return |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 92 | |
| 93 | subcommands = [] |
| 94 | for arg in arglist: |
| 95 | args = [arg] |
| 96 | if log: |
| 97 | subdir = str(arg) |
| 98 | else: |
| 99 | subdir = None |
| 100 | subcommands.append(subcommand(function, args, subdir)) |
mbligh | 415dc21 | 2009-06-15 21:53:34 +0000 | [diff] [blame] | 101 | return parallel(subcommands, timeout, return_results=return_results) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 102 | |
| 103 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 104 | class subcommand(object): |
| 105 | fork_hooks, join_hooks = [], [] |
| 106 | |
showard | 75cdfee | 2009-06-10 17:40:41 +0000 | [diff] [blame] | 107 | def __init__(self, func, args, subdir = None): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 108 | # func(args) - the subcommand to run |
| 109 | # subdir - the subdirectory to log results in |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 110 | if subdir: |
| 111 | self.subdir = os.path.abspath(subdir) |
| 112 | if not os.path.exists(self.subdir): |
| 113 | os.mkdir(self.subdir) |
| 114 | self.debug = os.path.join(self.subdir, 'debug') |
| 115 | if not os.path.exists(self.debug): |
| 116 | os.mkdir(self.debug) |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 117 | else: |
| 118 | self.subdir = None |
showard | 75cdfee | 2009-06-10 17:40:41 +0000 | [diff] [blame] | 119 | self.debug = None |
mbligh | d7685d3 | 2007-08-10 22:08:42 +0000 | [diff] [blame] | 120 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 121 | self.func = func |
| 122 | self.args = args |
| 123 | self.lambda_function = lambda: func(*args) |
| 124 | self.pid = None |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 125 | self.returncode = None |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 126 | |
| 127 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 128 | @classmethod |
| 129 | def register_fork_hook(cls, hook): |
| 130 | """ Register a function to be called from the child process after |
| 131 | forking. """ |
| 132 | cls.fork_hooks.append(hook) |
| 133 | |
| 134 | |
| 135 | @classmethod |
| 136 | def register_join_hook(cls, hook): |
| 137 | """ Register a function to be called when from the child process |
| 138 | just before the child process terminates (joins to the parent). """ |
| 139 | cls.join_hooks.append(hook) |
| 140 | |
| 141 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 142 | def redirect_output(self): |
showard | 75cdfee | 2009-06-10 17:40:41 +0000 | [diff] [blame] | 143 | if self.subdir and logging_manager_object: |
| 144 | tag = os.path.basename(self.subdir) |
| 145 | logging_manager_object.tee_redirect_debug_dir(self.debug, tag=tag) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 146 | |
| 147 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 148 | def fork_start(self): |
| 149 | sys.stdout.flush() |
| 150 | sys.stderr.flush() |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 151 | r, w = os.pipe() |
jadmanski | 77e8da8 | 2009-04-21 14:26:40 +0000 | [diff] [blame] | 152 | self.returncode = None |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 153 | self.pid = os.fork() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 154 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 155 | if self.pid: # I am the parent |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 156 | os.close(w) |
| 157 | self.result_pickle = os.fdopen(r, 'r') |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 158 | return |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 159 | else: |
| 160 | os.close(r) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 161 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 162 | # We are the child from this point on. Never return. |
| 163 | signal.signal(signal.SIGTERM, signal.SIG_DFL) # clear handler |
| 164 | if self.subdir: |
| 165 | os.chdir(self.subdir) |
| 166 | self.redirect_output() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 167 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 168 | try: |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 169 | for hook in self.fork_hooks: |
| 170 | hook(self) |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 171 | result = self.lambda_function() |
| 172 | os.write(w, pickle.dumps(result)) |
| 173 | except Exception, e: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 174 | traceback.print_exc() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 175 | exit_code = 1 |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 176 | os.write(w, pickle.dumps(e)) |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 177 | else: |
| 178 | exit_code = 0 |
| 179 | |
| 180 | try: |
| 181 | for hook in self.join_hooks: |
| 182 | hook(self) |
| 183 | finally: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 184 | sys.stdout.flush() |
| 185 | sys.stderr.flush() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 186 | os._exit(exit_code) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 187 | |
| 188 | |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 189 | def _handle_exitstatus(self, sts): |
| 190 | """ |
| 191 | This is partially borrowed from subprocess.Popen. |
| 192 | """ |
| 193 | if os.WIFSIGNALED(sts): |
| 194 | self.returncode = -os.WTERMSIG(sts) |
| 195 | elif os.WIFEXITED(sts): |
| 196 | self.returncode = os.WEXITSTATUS(sts) |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 197 | else: |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 198 | # Should never happen |
| 199 | raise RuntimeError("Unknown child exit status!") |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 200 | |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 201 | if self.returncode != 0: |
| 202 | print "subcommand failed pid %d" % self.pid |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 203 | print "%s" % (self.func,) |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 204 | print "rc=%d" % self.returncode |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 205 | print |
showard | 75cdfee | 2009-06-10 17:40:41 +0000 | [diff] [blame] | 206 | if self.debug: |
| 207 | stderr_file = os.path.join(self.debug, 'autoserv.stderr') |
| 208 | if os.path.exists(stderr_file): |
| 209 | for line in open(stderr_file).readlines(): |
| 210 | print line, |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 211 | print "\n--------------------------------------------\n" |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 212 | raise error.AutoservSubcommandError(self.func, self.returncode) |
| 213 | |
| 214 | |
| 215 | def poll(self): |
| 216 | """ |
| 217 | This is borrowed from subprocess.Popen. |
| 218 | """ |
| 219 | if self.returncode is None: |
| 220 | try: |
| 221 | pid, sts = os.waitpid(self.pid, os.WNOHANG) |
| 222 | if pid == self.pid: |
| 223 | self._handle_exitstatus(sts) |
| 224 | except os.error: |
| 225 | pass |
| 226 | return self.returncode |
| 227 | |
| 228 | |
| 229 | def wait(self): |
| 230 | """ |
| 231 | This is borrowed from subprocess.Popen. |
| 232 | """ |
| 233 | if self.returncode is None: |
| 234 | pid, sts = os.waitpid(self.pid, 0) |
| 235 | self._handle_exitstatus(sts) |
| 236 | return self.returncode |
| 237 | |
| 238 | |
| 239 | def fork_waitfor(self, timeout=None): |
| 240 | if not timeout: |
| 241 | return self.wait() |
| 242 | else: |
| 243 | start_time = time.time() |
| 244 | while time.time() <= start_time + timeout: |
| 245 | self.poll() |
| 246 | if self.returncode is not None: |
| 247 | return self.returncode |
| 248 | time.sleep(1) |
| 249 | |
| 250 | utils.nuke_pid(self.pid) |
| 251 | print "subcommand failed pid %d" % self.pid |
| 252 | print "%s" % (self.func,) |
| 253 | print "timeout after %ds" % timeout |
| 254 | print |
| 255 | return None |