mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 1 | __author__ = """Copyright Andy Whitcroft, Martin J. Bligh - 2006, 2007""" |
| 2 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 3 | import sys, os, subprocess, traceback, time, signal, pickle |
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 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 8 | def parallel(tasklist, timeout=None, return_results=False): |
| 9 | """Run a set of predefined subcommands in parallel. |
| 10 | If return_results is True instead of an exception being raised on |
| 11 | any error a list of the results/exceptions from the functions is |
| 12 | returned. |
| 13 | """ |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 14 | pids = [] |
| 15 | run_error = False |
| 16 | for task in tasklist: |
| 17 | task.fork_start() |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 18 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 19 | remaining_timeout = None |
| 20 | if timeout: |
| 21 | endtime = time.time() + timeout |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 22 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 23 | results = [] |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 24 | for task in tasklist: |
| 25 | if timeout: |
| 26 | remaining_timeout = max(endtime - time.time(), 1) |
| 27 | try: |
| 28 | status = task.fork_waitfor(remaining_timeout) |
| 29 | except error.AutoservSubcommandError: |
| 30 | run_error = True |
| 31 | else: |
| 32 | if status != 0: |
| 33 | run_error = True |
mbligh | 158ba7b | 2008-03-07 18:29:12 +0000 | [diff] [blame] | 34 | |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 35 | results.append(pickle.load(task.result_pickle)) |
| 36 | task.result_pickle.close() |
| 37 | |
| 38 | if return_results: |
| 39 | return results |
| 40 | elif run_error: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 41 | raise error.AutoservError('One or more subcommands failed') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 42 | |
| 43 | |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 44 | def parallel_simple(function, arglist, log=True, timeout=None): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 45 | """Each element in the arglist used to create a subcommand object, |
| 46 | where that arg is used both as a subdir name, and a single argument |
| 47 | to pass to "function". |
| 48 | We create a subcommand object for each element in the list, |
| 49 | then execute those subcommand objects in parallel.""" |
mbligh | dd3235b | 2008-01-14 16:44:19 +0000 | [diff] [blame] | 50 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 51 | # Bypass the multithreading if only one machine. |
| 52 | if len (arglist) == 1: |
| 53 | function(arglist[0]) |
| 54 | return |
| 55 | |
| 56 | subcommands = [] |
| 57 | for arg in arglist: |
| 58 | args = [arg] |
| 59 | if log: |
| 60 | subdir = str(arg) |
| 61 | else: |
| 62 | subdir = None |
| 63 | subcommands.append(subcommand(function, args, subdir)) |
| 64 | parallel(subcommands, timeout) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 65 | |
| 66 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 67 | def _where_art_thy_filehandles(): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 68 | os.system("ls -l /proc/%d/fd >> /dev/tty" % os.getpid()) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 69 | |
mbligh | dc735a2 | 2007-08-02 16:54:37 +0000 | [diff] [blame] | 70 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 71 | def _print_to_tty(string): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 72 | open('/dev/tty', 'w').write(string + '\n') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 73 | |
| 74 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 75 | def _redirect_stream(fd, output): |
mbligh | d4d8628 | 2009-02-26 00:46:34 +0000 | [diff] [blame] | 76 | newfd = os.open(output, os.O_WRONLY | os.O_CREAT | os.O_APPEND) |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 77 | os.dup2(newfd, fd) |
| 78 | os.close(newfd) |
| 79 | if fd == 1: |
| 80 | sys.stdout = os.fdopen(fd, 'w') |
| 81 | if fd == 2: |
| 82 | sys.stderr = os.fdopen(fd, 'w') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 83 | |
| 84 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 85 | def _redirect_stream_tee(fd, output, tag): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 86 | """Use the low-level fork & pipe operations here to get a fd, |
| 87 | not a filehandle. This ensures that we get both the |
| 88 | filehandle and fd for stdout/stderr redirected correctly.""" |
| 89 | r, w = os.pipe() |
| 90 | pid = os.fork() |
| 91 | if pid: # Parent |
| 92 | os.dup2(w, fd) |
| 93 | os.close(r) |
| 94 | os.close(w) |
| 95 | if fd == 1: |
| 96 | sys.stdout = os.fdopen(fd, 'w', 1) |
| 97 | if fd == 2: |
| 98 | sys.stderr = os.fdopen(fd, 'w', 1) |
| 99 | return |
| 100 | else: # Child |
jadmanski | f4a87ca | 2008-09-16 17:31:48 +0000 | [diff] [blame] | 101 | signal.signal(signal.SIGTERM, signal.SIG_DFL) # clear handler |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 102 | os.close(w) |
mbligh | d4d8628 | 2009-02-26 00:46:34 +0000 | [diff] [blame] | 103 | log = open(output, 'a') |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 104 | f = os.fdopen(r, 'r') |
| 105 | for line in iter(f.readline, ''): |
| 106 | # Tee straight to file |
| 107 | log.write(line) |
| 108 | log.flush() |
| 109 | # Prepend stdout with the tag |
| 110 | print tag + ' : ' + line, |
| 111 | sys.stdout.flush() |
| 112 | log.close() |
| 113 | os._exit(0) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 114 | |
| 115 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 116 | class subcommand(object): |
| 117 | fork_hooks, join_hooks = [], [] |
| 118 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 119 | def __init__(self, func, args, subdir = None, stdprint = True): |
| 120 | # func(args) - the subcommand to run |
| 121 | # subdir - the subdirectory to log results in |
| 122 | # stdprint - whether to print results to stdout/stderr |
| 123 | if subdir: |
| 124 | self.subdir = os.path.abspath(subdir) |
| 125 | if not os.path.exists(self.subdir): |
| 126 | os.mkdir(self.subdir) |
| 127 | self.debug = os.path.join(self.subdir, 'debug') |
| 128 | if not os.path.exists(self.debug): |
| 129 | os.mkdir(self.debug) |
| 130 | self.stdout = os.path.join(self.debug, 'stdout') |
| 131 | self.stderr = os.path.join(self.debug, 'stderr') |
| 132 | else: |
| 133 | self.subdir = None |
| 134 | self.debug = '/dev/null' |
| 135 | self.stdout = '/dev/null' |
| 136 | self.stderr = '/dev/null' |
mbligh | d7685d3 | 2007-08-10 22:08:42 +0000 | [diff] [blame] | 137 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 138 | self.func = func |
| 139 | self.args = args |
| 140 | self.lambda_function = lambda: func(*args) |
| 141 | self.pid = None |
| 142 | self.stdprint = stdprint |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 143 | self.returncode = None |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 144 | |
| 145 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 146 | @classmethod |
| 147 | def register_fork_hook(cls, hook): |
| 148 | """ Register a function to be called from the child process after |
| 149 | forking. """ |
| 150 | cls.fork_hooks.append(hook) |
| 151 | |
| 152 | |
| 153 | @classmethod |
| 154 | def register_join_hook(cls, hook): |
| 155 | """ Register a function to be called when from the child process |
| 156 | just before the child process terminates (joins to the parent). """ |
| 157 | cls.join_hooks.append(hook) |
| 158 | |
| 159 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 160 | def redirect_output(self): |
| 161 | if self.stdprint: |
| 162 | if self.subdir: |
| 163 | tag = os.path.basename(self.subdir) |
| 164 | _redirect_stream_tee(1, self.stdout, tag) |
| 165 | _redirect_stream_tee(2, self.stderr, tag) |
| 166 | else: |
| 167 | _redirect_stream(1, self.stdout) |
| 168 | _redirect_stream(2, self.stderr) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 169 | |
| 170 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 171 | def fork_start(self): |
| 172 | sys.stdout.flush() |
| 173 | sys.stderr.flush() |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 174 | r, w = os.pipe() |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 175 | self.pid = os.fork() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 176 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 177 | if self.pid: # I am the parent |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 178 | os.close(w) |
| 179 | self.result_pickle = os.fdopen(r, 'r') |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 180 | return |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 181 | else: |
| 182 | os.close(r) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 183 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 184 | # We are the child from this point on. Never return. |
| 185 | signal.signal(signal.SIGTERM, signal.SIG_DFL) # clear handler |
| 186 | if self.subdir: |
| 187 | os.chdir(self.subdir) |
| 188 | self.redirect_output() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 189 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 190 | try: |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 191 | for hook in self.fork_hooks: |
| 192 | hook(self) |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 193 | result = self.lambda_function() |
| 194 | os.write(w, pickle.dumps(result)) |
| 195 | except Exception, e: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 196 | traceback.print_exc() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 197 | exit_code = 1 |
mbligh | e183681 | 2009-04-17 18:25:14 +0000 | [diff] [blame] | 198 | os.write(w, pickle.dumps(e)) |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 199 | else: |
| 200 | exit_code = 0 |
| 201 | |
| 202 | try: |
| 203 | for hook in self.join_hooks: |
| 204 | hook(self) |
| 205 | finally: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 206 | sys.stdout.flush() |
| 207 | sys.stderr.flush() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 208 | os._exit(exit_code) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 209 | |
| 210 | |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 211 | def _handle_exitstatus(self, sts): |
| 212 | """ |
| 213 | This is partially borrowed from subprocess.Popen. |
| 214 | """ |
| 215 | if os.WIFSIGNALED(sts): |
| 216 | self.returncode = -os.WTERMSIG(sts) |
| 217 | elif os.WIFEXITED(sts): |
| 218 | self.returncode = os.WEXITSTATUS(sts) |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 219 | else: |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 220 | # Should never happen |
| 221 | raise RuntimeError("Unknown child exit status!") |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 222 | |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 223 | if self.returncode != 0: |
| 224 | print "subcommand failed pid %d" % self.pid |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 225 | print "%s" % (self.func,) |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 226 | print "rc=%d" % self.returncode |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 227 | print |
| 228 | if os.path.exists(self.stderr): |
| 229 | for line in open(self.stderr).readlines(): |
| 230 | print line, |
| 231 | print "\n--------------------------------------------\n" |
showard | c408c5e | 2009-01-08 23:30:53 +0000 | [diff] [blame] | 232 | raise error.AutoservSubcommandError(self.func, self.returncode) |
| 233 | |
| 234 | |
| 235 | def poll(self): |
| 236 | """ |
| 237 | This is borrowed from subprocess.Popen. |
| 238 | """ |
| 239 | if self.returncode is None: |
| 240 | try: |
| 241 | pid, sts = os.waitpid(self.pid, os.WNOHANG) |
| 242 | if pid == self.pid: |
| 243 | self._handle_exitstatus(sts) |
| 244 | except os.error: |
| 245 | pass |
| 246 | return self.returncode |
| 247 | |
| 248 | |
| 249 | def wait(self): |
| 250 | """ |
| 251 | This is borrowed from subprocess.Popen. |
| 252 | """ |
| 253 | if self.returncode is None: |
| 254 | pid, sts = os.waitpid(self.pid, 0) |
| 255 | self._handle_exitstatus(sts) |
| 256 | return self.returncode |
| 257 | |
| 258 | |
| 259 | def fork_waitfor(self, timeout=None): |
| 260 | if not timeout: |
| 261 | return self.wait() |
| 262 | else: |
| 263 | start_time = time.time() |
| 264 | while time.time() <= start_time + timeout: |
| 265 | self.poll() |
| 266 | if self.returncode is not None: |
| 267 | return self.returncode |
| 268 | time.sleep(1) |
| 269 | |
| 270 | utils.nuke_pid(self.pid) |
| 271 | print "subcommand failed pid %d" % self.pid |
| 272 | print "%s" % (self.func,) |
| 273 | print "timeout after %ds" % timeout |
| 274 | print |
| 275 | return None |