mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 1 | __author__ = """Copyright Andy Whitcroft, Martin J. Bligh - 2006, 2007""" |
| 2 | |
mbligh | 589bf32 | 2008-05-27 21:28:15 +0000 | [diff] [blame] | 3 | import sys, os, subprocess, traceback, time, signal |
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 | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 8 | def parallel(tasklist, timeout=None): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 9 | """Run an set of predefined subcommands in parallel""" |
| 10 | pids = [] |
| 11 | run_error = False |
| 12 | for task in tasklist: |
| 13 | task.fork_start() |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 14 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 15 | remaining_timeout = None |
| 16 | if timeout: |
| 17 | endtime = time.time() + timeout |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 18 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 19 | for task in tasklist: |
| 20 | if timeout: |
| 21 | remaining_timeout = max(endtime - time.time(), 1) |
| 22 | try: |
| 23 | status = task.fork_waitfor(remaining_timeout) |
| 24 | except error.AutoservSubcommandError: |
| 25 | run_error = True |
| 26 | else: |
| 27 | if status != 0: |
| 28 | run_error = True |
mbligh | 158ba7b | 2008-03-07 18:29:12 +0000 | [diff] [blame] | 29 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 30 | if run_error: |
| 31 | raise error.AutoservError('One or more subcommands failed') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 32 | |
| 33 | |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 34 | def parallel_simple(function, arglist, log=True, timeout=None): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 35 | """Each element in the arglist used to create a subcommand object, |
| 36 | where that arg is used both as a subdir name, and a single argument |
| 37 | to pass to "function". |
| 38 | We create a subcommand object for each element in the list, |
| 39 | then execute those subcommand objects in parallel.""" |
mbligh | dd3235b | 2008-01-14 16:44:19 +0000 | [diff] [blame] | 40 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 41 | # Bypass the multithreading if only one machine. |
| 42 | if len (arglist) == 1: |
| 43 | function(arglist[0]) |
| 44 | return |
| 45 | |
| 46 | subcommands = [] |
| 47 | for arg in arglist: |
| 48 | args = [arg] |
| 49 | if log: |
| 50 | subdir = str(arg) |
| 51 | else: |
| 52 | subdir = None |
| 53 | subcommands.append(subcommand(function, args, subdir)) |
| 54 | parallel(subcommands, timeout) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 55 | |
| 56 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 57 | def _where_art_thy_filehandles(): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 58 | os.system("ls -l /proc/%d/fd >> /dev/tty" % os.getpid()) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 59 | |
mbligh | dc735a2 | 2007-08-02 16:54:37 +0000 | [diff] [blame] | 60 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 61 | def _print_to_tty(string): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 62 | open('/dev/tty', 'w').write(string + '\n') |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 63 | |
| 64 | |
mbligh | cee25b1 | 2007-08-31 08:53:05 +0000 | [diff] [blame] | 65 | def _redirect_stream(fd, output): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 66 | newfd = os.open(output, os.O_WRONLY | os.O_CREAT) |
| 67 | os.dup2(newfd, fd) |
| 68 | os.close(newfd) |
| 69 | if fd == 1: |
| 70 | sys.stdout = os.fdopen(fd, 'w') |
| 71 | if fd == 2: |
| 72 | sys.stderr = os.fdopen(fd, 'w') |
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_tee(fd, output, tag): |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 76 | """Use the low-level fork & pipe operations here to get a fd, |
| 77 | not a filehandle. This ensures that we get both the |
| 78 | filehandle and fd for stdout/stderr redirected correctly.""" |
| 79 | r, w = os.pipe() |
| 80 | pid = os.fork() |
| 81 | if pid: # Parent |
| 82 | os.dup2(w, fd) |
| 83 | os.close(r) |
| 84 | os.close(w) |
| 85 | if fd == 1: |
| 86 | sys.stdout = os.fdopen(fd, 'w', 1) |
| 87 | if fd == 2: |
| 88 | sys.stderr = os.fdopen(fd, 'w', 1) |
| 89 | return |
| 90 | else: # Child |
jadmanski | f4a87ca | 2008-09-16 17:31:48 +0000 | [diff] [blame] | 91 | signal.signal(signal.SIGTERM, signal.SIG_DFL) # clear handler |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 92 | os.close(w) |
| 93 | log = open(output, 'w') |
| 94 | f = os.fdopen(r, 'r') |
| 95 | for line in iter(f.readline, ''): |
| 96 | # Tee straight to file |
| 97 | log.write(line) |
| 98 | log.flush() |
| 99 | # Prepend stdout with the tag |
| 100 | print tag + ' : ' + line, |
| 101 | sys.stdout.flush() |
| 102 | log.close() |
| 103 | os._exit(0) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 104 | |
| 105 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 106 | class subcommand(object): |
| 107 | fork_hooks, join_hooks = [], [] |
| 108 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 109 | def __init__(self, func, args, subdir = None, stdprint = True): |
| 110 | # func(args) - the subcommand to run |
| 111 | # subdir - the subdirectory to log results in |
| 112 | # stdprint - whether to print results to stdout/stderr |
| 113 | if subdir: |
| 114 | self.subdir = os.path.abspath(subdir) |
| 115 | if not os.path.exists(self.subdir): |
| 116 | os.mkdir(self.subdir) |
| 117 | self.debug = os.path.join(self.subdir, 'debug') |
| 118 | if not os.path.exists(self.debug): |
| 119 | os.mkdir(self.debug) |
| 120 | self.stdout = os.path.join(self.debug, 'stdout') |
| 121 | self.stderr = os.path.join(self.debug, 'stderr') |
| 122 | else: |
| 123 | self.subdir = None |
| 124 | self.debug = '/dev/null' |
| 125 | self.stdout = '/dev/null' |
| 126 | self.stderr = '/dev/null' |
mbligh | d7685d3 | 2007-08-10 22:08:42 +0000 | [diff] [blame] | 127 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 128 | self.func = func |
| 129 | self.args = args |
| 130 | self.lambda_function = lambda: func(*args) |
| 131 | self.pid = None |
| 132 | self.stdprint = stdprint |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 133 | |
| 134 | |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 135 | @classmethod |
| 136 | def register_fork_hook(cls, hook): |
| 137 | """ Register a function to be called from the child process after |
| 138 | forking. """ |
| 139 | cls.fork_hooks.append(hook) |
| 140 | |
| 141 | |
| 142 | @classmethod |
| 143 | def register_join_hook(cls, hook): |
| 144 | """ Register a function to be called when from the child process |
| 145 | just before the child process terminates (joins to the parent). """ |
| 146 | cls.join_hooks.append(hook) |
| 147 | |
| 148 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 149 | def redirect_output(self): |
| 150 | if self.stdprint: |
| 151 | if self.subdir: |
| 152 | tag = os.path.basename(self.subdir) |
| 153 | _redirect_stream_tee(1, self.stdout, tag) |
| 154 | _redirect_stream_tee(2, self.stderr, tag) |
| 155 | else: |
| 156 | _redirect_stream(1, self.stdout) |
| 157 | _redirect_stream(2, self.stderr) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 158 | |
| 159 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 160 | def fork_start(self): |
| 161 | sys.stdout.flush() |
| 162 | sys.stderr.flush() |
| 163 | self.pid = os.fork() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 164 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 165 | if self.pid: # I am the parent |
| 166 | return |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 167 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 168 | # We are the child from this point on. Never return. |
| 169 | signal.signal(signal.SIGTERM, signal.SIG_DFL) # clear handler |
| 170 | if self.subdir: |
| 171 | os.chdir(self.subdir) |
| 172 | self.redirect_output() |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 173 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 174 | try: |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 175 | for hook in self.fork_hooks: |
| 176 | hook(self) |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 177 | self.lambda_function() |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 178 | except: |
| 179 | traceback.print_exc() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 180 | exit_code = 1 |
| 181 | else: |
| 182 | exit_code = 0 |
| 183 | |
| 184 | try: |
| 185 | for hook in self.join_hooks: |
| 186 | hook(self) |
| 187 | finally: |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 188 | sys.stdout.flush() |
| 189 | sys.stderr.flush() |
jadmanski | 550fdc2 | 2008-11-20 16:32:08 +0000 | [diff] [blame] | 190 | os._exit(exit_code) |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 191 | |
| 192 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 193 | def fork_waitfor(self, timeout=None): |
| 194 | if not timeout: |
| 195 | (pid, status) = os.waitpid(self.pid, 0) |
| 196 | else: |
| 197 | pid = None |
| 198 | start_time = time.time() |
| 199 | while time.time() <= start_time + timeout: |
| 200 | (pid, status) = os.waitpid(self.pid, os.WNOHANG) |
| 201 | if pid: |
| 202 | break |
| 203 | time.sleep(1) |
mbligh | c3aee0f | 2008-01-17 16:26:39 +0000 | [diff] [blame] | 204 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 205 | if not pid: |
| 206 | utils.nuke_pid(self.pid) |
| 207 | print "subcommand failed pid %d" % self.pid |
| 208 | print "%s" % (self.func,) |
| 209 | print "timeout after %ds" % timeout |
| 210 | print |
| 211 | return None |
mbligh | b0fab82 | 2007-07-25 16:40:19 +0000 | [diff] [blame] | 212 | |
jadmanski | 0afbb63 | 2008-06-06 21:10:57 +0000 | [diff] [blame] | 213 | if status != 0: |
| 214 | print "subcommand failed pid %d" % pid |
| 215 | print "%s" % (self.func,) |
| 216 | print "rc=%d" % status |
| 217 | print |
| 218 | if os.path.exists(self.stderr): |
| 219 | for line in open(self.stderr).readlines(): |
| 220 | print line, |
| 221 | print "\n--------------------------------------------\n" |
| 222 | raise error.AutoservSubcommandError(self.func, status) |
| 223 | return status |