Fix for issue #127. The "got_revision" property in perforce is now set after a P4...
[buildbot.git] / buildbot / slave / bot.py
blob665996cf243e42235238156860ab9672aebd5cf4
2 import os.path
4 import buildbot
6 from twisted.spread import pb
7 from twisted.python import log
8 from twisted.internet import reactor, defer
9 from twisted.application import service, internet
10 from twisted.cred import credentials
12 from buildbot.util import now
13 from buildbot.pbutil import ReconnectingPBClientFactory
14 from buildbot.slave import registry
15 # make sure the standard commands get registered. This import is performed
16 # for its side-effects.
17 from buildbot.slave import commands
18 # and make pyflakes think we aren't being stupid
19 commands = commands
21 class NoCommandRunning(pb.Error):
22 pass
23 class WrongCommandRunning(pb.Error):
24 pass
25 class UnknownCommand(pb.Error):
26 pass
28 class Master:
29 def __init__(self, host, port, username, password):
30 self.host = host
31 self.port = port
32 self.username = username
33 self.password = password
35 class SlaveBuild:
37 """This is an object that can hold state from one step to another in the
38 same build. All SlaveCommands have access to it.
39 """
40 def __init__(self, builder):
41 self.builder = builder
43 class SlaveBuilder(pb.Referenceable, service.Service):
45 """This is the local representation of a single Builder: it handles a
46 single kind of build (like an all-warnings build). It has a name and a
47 home directory. The rest of its behavior is determined by the master.
48 """
50 stopCommandOnShutdown = True
52 # remote is a ref to the Builder object on the master side, and is set
53 # when they attach. We use it to detect when the connection to the master
54 # is severed.
55 remote = None
57 # .build points to a SlaveBuild object, a new one for each build
58 build = None
60 # .command points to a SlaveCommand instance, and is set while the step
61 # is running. We use it to implement the stopBuild method.
62 command = None
64 # .remoteStep is a ref to the master-side BuildStep object, and is set
65 # when the step is started
66 remoteStep = None
68 def __init__(self, name, not_really):
69 #service.Service.__init__(self) # Service has no __init__ method
70 self.setName(name)
71 self.not_really = not_really
73 def __repr__(self):
74 return "<SlaveBuilder '%s' at %d>" % (self.name, id(self))
76 def setServiceParent(self, parent):
77 service.Service.setServiceParent(self, parent)
78 self.bot = self.parent
79 # note that self.parent will go away when the buildmaster's config
80 # file changes and this Builder is removed (possibly because it has
81 # been changed, so the Builder will be re-added again in a moment).
82 # This may occur during a build, while a step is running.
84 def setBuilddir(self, builddir):
85 assert self.parent
86 self.builddir = builddir
87 self.basedir = os.path.join(self.bot.basedir, self.builddir)
88 if not os.path.isdir(self.basedir):
89 os.mkdir(self.basedir)
91 def stopService(self):
92 service.Service.stopService(self)
93 if self.stopCommandOnShutdown:
94 self.stopCommand()
96 def activity(self):
97 bot = self.parent
98 if bot:
99 buildslave = bot.parent
100 if buildslave:
101 bf = buildslave.bf
102 bf.activity()
104 def remote_setMaster(self, remote):
105 self.remote = remote
106 self.remote.notifyOnDisconnect(self.lostRemote)
107 def remote_print(self, message):
108 log.msg("SlaveBuilder.remote_print(%s): message from master: %s" %
109 (self.name, message))
110 if message == "ping":
111 return self.remote_ping()
113 def remote_ping(self):
114 log.msg("SlaveBuilder.remote_ping(%s)" % self)
115 if self.bot and self.bot.parent:
116 debugOpts = self.bot.parent.debugOpts
117 if debugOpts.get("stallPings"):
118 log.msg(" debug_stallPings")
119 timeout, timers = debugOpts["stallPings"]
120 d = defer.Deferred()
121 t = reactor.callLater(timeout, d.callback, None)
122 timers.append(t)
123 return d
124 if debugOpts.get("failPingOnce"):
125 log.msg(" debug_failPingOnce")
126 class FailPingError(pb.Error): pass
127 del debugOpts['failPingOnce']
128 raise FailPingError("debug_failPingOnce means we should fail")
130 def lostRemote(self, remote):
131 log.msg("lost remote")
132 self.remote = None
134 def lostRemoteStep(self, remotestep):
135 log.msg("lost remote step")
136 self.remoteStep = None
137 if self.stopCommandOnShutdown:
138 self.stopCommand()
140 # the following are Commands that can be invoked by the master-side
141 # Builder
142 def remote_startBuild(self):
143 """This is invoked before the first step of any new build is run. It
144 creates a new SlaveBuild object, which holds slave-side state from
145 one step to the next."""
146 self.build = SlaveBuild(self)
147 log.msg("%s.startBuild" % self)
149 def remote_startCommand(self, stepref, stepId, command, args):
151 This gets invoked by L{buildbot.process.step.RemoteCommand.start}, as
152 part of various master-side BuildSteps, to start various commands
153 that actually do the build. I return nothing. Eventually I will call
154 .commandComplete() to notify the master-side RemoteCommand that I'm
155 done.
158 self.activity()
160 if self.command:
161 log.msg("leftover command, dropping it")
162 self.stopCommand()
164 try:
165 factory, version = registry.commandRegistry[command]
166 except KeyError:
167 raise UnknownCommand, "unrecognized SlaveCommand '%s'" % command
168 self.command = factory(self, stepId, args)
170 log.msg(" startCommand:%s [id %s]" % (command,stepId))
171 self.remoteStep = stepref
172 self.remoteStep.notifyOnDisconnect(self.lostRemoteStep)
173 d = self.command.doStart()
174 d.addCallback(lambda res: None)
175 d.addBoth(self.commandComplete)
176 return None
178 def remote_interruptCommand(self, stepId, why):
179 """Halt the current step."""
180 log.msg("asked to interrupt current command: %s" % why)
181 self.activity()
182 if not self.command:
183 # TODO: just log it, a race could result in their interrupting a
184 # command that wasn't actually running
185 log.msg(" .. but none was running")
186 return
187 self.command.doInterrupt()
190 def stopCommand(self):
191 """Make any currently-running command die, with no further status
192 output. This is used when the buildslave is shutting down or the
193 connection to the master has been lost. Interrupt the command,
194 silence it, and then forget about it."""
195 if not self.command:
196 return
197 log.msg("stopCommand: halting current command %s" % self.command)
198 self.command.doInterrupt() # shut up! and die!
199 self.command = None # forget you!
201 # sendUpdate is invoked by the Commands we spawn
202 def sendUpdate(self, data):
203 """This sends the status update to the master-side
204 L{buildbot.process.step.RemoteCommand} object, giving it a sequence
205 number in the process. It adds the update to a queue, and asks the
206 master to acknowledge the update so it can be removed from that
207 queue."""
209 if not self.running:
210 # .running comes from service.Service, and says whether the
211 # service is running or not. If we aren't running, don't send any
212 # status messages.
213 return
214 # the update[1]=0 comes from the leftover 'updateNum', which the
215 # master still expects to receive. Provide it to avoid significant
216 # interoperability issues between new slaves and old masters.
217 if self.remoteStep:
218 update = [data, 0]
219 updates = [update]
220 d = self.remoteStep.callRemote("update", updates)
221 d.addCallback(self.ackUpdate)
222 d.addErrback(self._ackFailed, "SlaveBuilder.sendUpdate")
224 def ackUpdate(self, acknum):
225 self.activity() # update the "last activity" timer
227 def ackComplete(self, dummy):
228 self.activity() # update the "last activity" timer
230 def _ackFailed(self, why, where):
231 log.msg("SlaveBuilder._ackFailed:", where)
232 #log.err(why) # we don't really care
235 # this is fired by the Deferred attached to each Command
236 def commandComplete(self, failure):
237 if failure:
238 log.msg("SlaveBuilder.commandFailed", self.command)
239 log.err(failure)
240 # failure, if present, is a failure.Failure. To send it across
241 # the wire, we must turn it into a pb.CopyableFailure.
242 failure = pb.CopyableFailure(failure)
243 failure.unsafeTracebacks = True
244 else:
245 # failure is None
246 log.msg("SlaveBuilder.commandComplete", self.command)
247 self.command = None
248 if not self.running:
249 log.msg(" but we weren't running, quitting silently")
250 return
251 if self.remoteStep:
252 self.remoteStep.dontNotifyOnDisconnect(self.lostRemoteStep)
253 d = self.remoteStep.callRemote("complete", failure)
254 d.addCallback(self.ackComplete)
255 d.addErrback(self._ackFailed, "sendComplete")
256 self.remoteStep = None
259 def remote_shutdown(self):
260 print "slave shutting down on command from master"
261 reactor.stop()
264 class Bot(pb.Referenceable, service.MultiService):
265 """I represent the slave-side bot."""
266 usePTY = None
267 name = "bot"
269 def __init__(self, basedir, usePTY, not_really=0):
270 service.MultiService.__init__(self)
271 self.basedir = basedir
272 self.usePTY = usePTY
273 self.not_really = not_really
274 self.builders = {}
276 def startService(self):
277 assert os.path.isdir(self.basedir)
278 service.MultiService.startService(self)
280 def remote_getDirs(self):
281 return filter(lambda d: os.path.isdir(d), os.listdir(self.basedir))
283 def remote_getCommands(self):
284 commands = {}
285 for name, (factory, version) in registry.commandRegistry.items():
286 commands[name] = version
287 return commands
289 def remote_setBuilderList(self, wanted):
290 retval = {}
291 wanted_dirs = ["info"]
292 for (name, builddir) in wanted:
293 wanted_dirs.append(builddir)
294 b = self.builders.get(name, None)
295 if b:
296 if b.builddir != builddir:
297 log.msg("changing builddir for builder %s from %s to %s" \
298 % (name, b.builddir, builddir))
299 b.setBuilddir(builddir)
300 else:
301 b = SlaveBuilder(name, self.not_really)
302 b.usePTY = self.usePTY
303 b.setServiceParent(self)
304 b.setBuilddir(builddir)
305 self.builders[name] = b
306 retval[name] = b
307 for name in self.builders.keys():
308 if not name in map(lambda a: a[0], wanted):
309 log.msg("removing old builder %s" % name)
310 self.builders[name].disownServiceParent()
311 del(self.builders[name])
313 for d in os.listdir(self.basedir):
314 if os.path.isdir(d):
315 if d not in wanted_dirs:
316 log.msg("I have a leftover directory '%s' that is not "
317 "being used by the buildmaster: you can delete "
318 "it now" % d)
319 return retval
321 def remote_print(self, message):
322 log.msg("message from master:", message)
324 def remote_getSlaveInfo(self):
325 """This command retrieves data from the files in SLAVEDIR/info/* and
326 sends the contents to the buildmaster. These are used to describe
327 the slave and its configuration, and should be created and
328 maintained by the slave administrator. They will be retrieved each
329 time the master-slave connection is established.
332 files = {}
333 basedir = os.path.join(self.basedir, "info")
334 if not os.path.isdir(basedir):
335 return files
336 for f in os.listdir(basedir):
337 filename = os.path.join(basedir, f)
338 if os.path.isfile(filename):
339 files[f] = open(filename, "r").read()
340 return files
342 class BotFactory(ReconnectingPBClientFactory):
343 # 'keepaliveInterval' serves two purposes. The first is to keep the
344 # connection alive: it guarantees that there will be at least some
345 # traffic once every 'keepaliveInterval' seconds, which may help keep an
346 # interposed NAT gateway from dropping the address mapping because it
347 # thinks the connection has been abandoned. The second is to put an upper
348 # limit on how long the buildmaster might have gone away before we notice
349 # it. For this second purpose, we insist upon seeing *some* evidence of
350 # the buildmaster at least once every 'keepaliveInterval' seconds.
351 keepaliveInterval = None # None = do not use keepalives
353 # 'keepaliveTimeout' seconds before the interval expires, we will send a
354 # keepalive request, both to add some traffic to the connection, and to
355 # prompt a response from the master in case all our builders are idle. We
356 # don't insist upon receiving a timely response from this message: a slow
357 # link might put the request at the wrong end of a large build message.
358 keepaliveTimeout = 30 # how long we will go without a response
360 keepaliveTimer = None
361 activityTimer = None
362 lastActivity = 0
363 unsafeTracebacks = 1
364 perspective = None
366 def __init__(self, keepaliveInterval, keepaliveTimeout):
367 ReconnectingPBClientFactory.__init__(self)
368 self.keepaliveInterval = keepaliveInterval
369 self.keepaliveTimeout = keepaliveTimeout
371 def startedConnecting(self, connector):
372 ReconnectingPBClientFactory.startedConnecting(self, connector)
373 self.connector = connector
375 def gotPerspective(self, perspective):
376 ReconnectingPBClientFactory.gotPerspective(self, perspective)
377 self.perspective = perspective
378 try:
379 perspective.broker.transport.setTcpKeepAlive(1)
380 except:
381 log.msg("unable to set SO_KEEPALIVE")
382 if not self.keepaliveInterval:
383 self.keepaliveInterval = 10*60
384 self.activity()
385 if self.keepaliveInterval:
386 log.msg("sending application-level keepalives every %d seconds" \
387 % self.keepaliveInterval)
388 self.startTimers()
390 def clientConnectionFailed(self, connector, reason):
391 self.connector = None
392 ReconnectingPBClientFactory.clientConnectionFailed(self,
393 connector, reason)
395 def clientConnectionLost(self, connector, reason):
396 self.connector = None
397 self.stopTimers()
398 self.perspective = None
399 ReconnectingPBClientFactory.clientConnectionLost(self,
400 connector, reason)
402 def startTimers(self):
403 assert self.keepaliveInterval
404 assert not self.keepaliveTimer
405 assert not self.activityTimer
406 # Insist that doKeepalive fires before checkActivity. Really, it
407 # needs to happen at least one RTT beforehand.
408 assert self.keepaliveInterval > self.keepaliveTimeout
410 # arrange to send a keepalive a little while before our deadline
411 when = self.keepaliveInterval - self.keepaliveTimeout
412 self.keepaliveTimer = reactor.callLater(when, self.doKeepalive)
413 # and check for activity too
414 self.activityTimer = reactor.callLater(self.keepaliveInterval,
415 self.checkActivity)
417 def stopTimers(self):
418 if self.keepaliveTimer:
419 self.keepaliveTimer.cancel()
420 self.keepaliveTimer = None
421 if self.activityTimer:
422 self.activityTimer.cancel()
423 self.activityTimer = None
425 def activity(self, res=None):
426 self.lastActivity = now()
428 def doKeepalive(self):
429 # send the keepalive request. If it fails outright, the connection
430 # was already dropped, so just log and ignore.
431 self.keepaliveTimer = None
432 log.msg("sending app-level keepalive")
433 d = self.perspective.callRemote("keepalive")
434 d.addCallback(self.activity)
435 d.addErrback(self.keepaliveLost)
437 def keepaliveLost(self, f):
438 log.msg("BotFactory.keepaliveLost")
440 def checkActivity(self):
441 self.activityTimer = None
442 if self.lastActivity + self.keepaliveInterval < now():
443 log.msg("BotFactory.checkActivity: nothing from master for "
444 "%d secs" % (now() - self.lastActivity))
445 self.perspective.broker.transport.loseConnection()
446 return
447 self.startTimers()
449 def stopFactory(self):
450 ReconnectingPBClientFactory.stopFactory(self)
451 self.stopTimers()
454 class BuildSlave(service.MultiService):
455 botClass = Bot
457 # debugOpts is a dictionary used during unit tests.
459 # debugOpts['stallPings'] can be set to a tuple of (timeout, []). Any
460 # calls to remote_print will stall for 'timeout' seconds before
461 # returning. The DelayedCalls used to implement this are stashed in the
462 # list so they can be cancelled later.
464 # debugOpts['failPingOnce'] can be set to True to make the slaveping fail
465 # exactly once.
467 def __init__(self, buildmaster_host, port, name, passwd, basedir,
468 keepalive, usePTY, keepaliveTimeout=30, umask=None,
469 debugOpts={}):
470 log.msg("Creating BuildSlave -- buildbot.version: %s" % buildbot.version)
471 service.MultiService.__init__(self)
472 self.debugOpts = debugOpts.copy()
473 bot = self.botClass(basedir, usePTY)
474 bot.setServiceParent(self)
475 self.bot = bot
476 if keepalive == 0:
477 keepalive = None
478 self.umask = umask
479 bf = self.bf = BotFactory(keepalive, keepaliveTimeout)
480 bf.startLogin(credentials.UsernamePassword(name, passwd), client=bot)
481 self.connection = c = internet.TCPClient(buildmaster_host, port, bf)
482 c.setServiceParent(self)
484 def waitUntilDisconnected(self):
485 # utility method for testing. Returns a Deferred that will fire when
486 # we lose the connection to the master.
487 if not self.bf.perspective:
488 return defer.succeed(None)
489 d = defer.Deferred()
490 self.bf.perspective.notifyOnDisconnect(lambda res: d.callback(None))
491 return d
493 def startService(self):
494 if self.umask is not None:
495 os.umask(self.umask)
496 service.MultiService.startService(self)
498 def stopService(self):
499 self.bf.continueTrying = 0
500 self.bf.stopTrying()
501 service.MultiService.stopService(self)
502 # now kill the TCP connection
503 # twisted >2.0.1 does this for us, and leaves _connection=None
504 if self.connection._connection:
505 self.connection._connection.disconnect()