예제 #1
0
    def __init__(self, cf, opts, type, args, workdir, quitMinutes, sharedWorker=False):
        """Initialize the object as a client, given a configuration.
           cf =  the configuration class
		   opts = dictionary with options
           type = the worker plugin name to start.
           args = arguments for the worker plugin."""
        # the run lock and condition variable
        self.runLock = threading.Lock()
        self.runCondVar = threading.Condition(self.runLock)
        self.conf = cf
        self.opts = opts
        self.type = type
        self.args = args
        self.quit = False
        self.isShared = sharedWorker
        self.id = "%s-%d" % (self.conf.getHostName(), os.getpid())
        # the number of seconds after which to quit if there is no work:
        self.quitSeconds = None
        if quitMinutes is not None:
            self.quitSeconds = 60 * quitMinutes
        log.info("Worker ID: %s." % self.id)
        # Process (untar) the run request into a directory name
        # that is unique for this process+hostname, and worker job iteration
        if workdir is None:
            self.workerTopDir = self.conf.getRunDir()
            createTopDir = True
            # self.mainDir=os.path.join(self.conf.getRunDir(), self.id)
        else:
            if not os.path.exists(workdir):
                raise WorkerError("Given run directory %s does not exist." % (workdir))
            self.workerTopDir = workdir
            createTopDir = False

        # fix the path so the server knows where it is
        if not os.path.isabs(self.workerTopDir):
            self.workerTopDir = os.path.join(os.getcwd(), self.workerTopDir)
            isabs = False
        else:
            isabs = True

        # set the actual path being used.
        self.mainDir = os.path.join(self.workerTopDir, self.id)

        try:
            log.debug("Creating work directory %s" % (self.mainDir))
            if createTopDir:
                if not os.path.exists(self.workerTopDir):
                    os.mkdir(self.workerTopDir)
                self.workerTopDirCreated = True
            else:
                self.workerTopDirCreated = False
            os.mkdir(self.mainDir)
        except:
            if isabs:
                absn = ""
            else:
                absn = "in the current working directory"
            log.error("Can't create the directory '%s' %s." % (self.mainDir, absn))
            log.error("cpc-worker must be able to write in a temporary directory")
            log.error("Run cpc-worker from a user-writeable directory (e.g. /tmp)")
            raise WorkerError("Can't create directory '%s' %s." % (self.mainDir, absn))
        self.heartbeat = heartbeat.HeartbeatSender(self, self.runCondVar)  # self.id, self.mainDir,
        # First get our architecture(s) (hw + sw) from the plugin
        self.plugin = PlatformPlugin(self.type, self.mainDir, self.conf)
        canRun = self.plugin.canRun()
        if not canRun:
            raise WorkerError("Plugin can't run.")
        self.platforms = self._getPlatforms(self.plugin)

        # we make copies to be able to subtract our usage off the max resources
        self.remainingPlatforms = copy.deepcopy(self.platforms)
        # Then check what executables we already have
        self._getExecutables()
        # start without workloads

        if len(self.exelist.findAllByPlattform(self.type)) == 0:
            print "No executables found for platform %s" % self.type
            sys.exit(1)

        self._printAvailableExes()
        self.workloads = []
        self.iteration = 0
        self.acceptCommands = True
        # install the signal handler
        signalHandlerAddWorker(self)
예제 #2
0
    def __init__(self,
                 cf,
                 opts,
                 type,
                 args,
                 workdir,
                 quitMinutes,
                 sharedWorker=False):
        """Initialize the object as a client, given a configuration.
           cf =  the configuration class
		   opts = dictionary with options
           type = the worker plugin name to start.
           args = arguments for the worker plugin."""
        # the run lock and condition variable
        self.runLock = threading.Lock()
        self.runCondVar = threading.Condition(self.runLock)
        self.conf = cf
        self.opts = opts
        self.type = type
        self.args = args
        self.quit = False
        self.isShared = sharedWorker
        self.id = "%s-%d" % (self.conf.getHostName(), os.getpid())
        # the number of seconds after which to quit if there is no work:
        self.quitSeconds = None
        if quitMinutes is not None:
            self.quitSeconds = 60 * quitMinutes
        log.info("Worker ID: %s." % self.id)
        # Process (untar) the run request into a directory name
        # that is unique for this process+hostname, and worker job iteration
        if workdir is None:
            self.workerTopDir = self.conf.getRunDir()
            createTopDir = True
            #self.mainDir=os.path.join(self.conf.getRunDir(), self.id)
        else:
            if not os.path.exists(workdir):
                raise WorkerError("Given run directory %s does not exist." %
                                  (workdir))
            self.workerTopDir = workdir
            createTopDir = False

        # fix the path so the server knows where it is
        if not os.path.isabs(self.workerTopDir):
            self.workerTopDir = os.path.join(os.getcwd(), self.workerTopDir)
            isabs = False
        else:
            isabs = True

        # set the actual path being used.
        self.mainDir = os.path.join(self.workerTopDir, self.id)

        try:
            log.debug("Creating work directory %s" % (self.mainDir))
            if createTopDir:
                if not os.path.exists(self.workerTopDir):
                    os.mkdir(self.workerTopDir)
                self.workerTopDirCreated = True
            else:
                self.workerTopDirCreated = False
            os.mkdir(self.mainDir)
        except:
            if isabs:
                absn = ""
            else:
                absn = "in the current working directory"
            log.error("Can't create the directory '%s' %s." %
                      (self.mainDir, absn))
            log.error(
                "cpc-worker must be able to write in a temporary directory")
            log.error(
                "Run cpc-worker from a user-writeable directory (e.g. /tmp)")
            raise WorkerError("Can't create directory '%s' %s." %
                              (self.mainDir, absn))
        self.heartbeat = heartbeat.HeartbeatSender(
            self,  #self.id, self.mainDir,
            self.runCondVar)
        # First get our architecture(s) (hw + sw) from the plugin
        self.plugin = PlatformPlugin(self.type, self.mainDir, self.conf)
        canRun = self.plugin.canRun()
        if not canRun:
            raise WorkerError("Plugin can't run.")
        self.platforms = self._getPlatforms(self.plugin)

        # we make copies to be able to subtract our usage off the max resources
        self.remainingPlatforms = copy.deepcopy(self.platforms)
        # Then check what executables we already have
        self._getExecutables()
        # start without workloads

        if len(self.exelist.findAllByPlattform(self.type)) == 0:
            print "No executables found for platform %s" % self.type
            sys.exit(1)

        self._printAvailableExes()
        self.workloads = []
        self.iteration = 0
        self.acceptCommands = True
        # install the signal handler
        signalHandlerAddWorker(self)
예제 #3
0
class Worker(object):
    """The worker class creates a worker client that contacts a server
       and asks for tasks."""

    def __init__(self, cf, opts, type, args, workdir, quitMinutes, sharedWorker=False):
        """Initialize the object as a client, given a configuration.
           cf =  the configuration class
		   opts = dictionary with options
           type = the worker plugin name to start.
           args = arguments for the worker plugin."""
        # the run lock and condition variable
        self.runLock = threading.Lock()
        self.runCondVar = threading.Condition(self.runLock)
        self.conf = cf
        self.opts = opts
        self.type = type
        self.args = args
        self.quit = False
        self.isShared = sharedWorker
        self.id = "%s-%d" % (self.conf.getHostName(), os.getpid())
        # the number of seconds after which to quit if there is no work:
        self.quitSeconds = None
        if quitMinutes is not None:
            self.quitSeconds = 60 * quitMinutes
        log.info("Worker ID: %s." % self.id)
        # Process (untar) the run request into a directory name
        # that is unique for this process+hostname, and worker job iteration
        if workdir is None:
            self.workerTopDir = self.conf.getRunDir()
            createTopDir = True
            # self.mainDir=os.path.join(self.conf.getRunDir(), self.id)
        else:
            if not os.path.exists(workdir):
                raise WorkerError("Given run directory %s does not exist." % (workdir))
            self.workerTopDir = workdir
            createTopDir = False

        # fix the path so the server knows where it is
        if not os.path.isabs(self.workerTopDir):
            self.workerTopDir = os.path.join(os.getcwd(), self.workerTopDir)
            isabs = False
        else:
            isabs = True

        # set the actual path being used.
        self.mainDir = os.path.join(self.workerTopDir, self.id)

        try:
            log.debug("Creating work directory %s" % (self.mainDir))
            if createTopDir:
                if not os.path.exists(self.workerTopDir):
                    os.mkdir(self.workerTopDir)
                self.workerTopDirCreated = True
            else:
                self.workerTopDirCreated = False
            os.mkdir(self.mainDir)
        except:
            if isabs:
                absn = ""
            else:
                absn = "in the current working directory"
            log.error("Can't create the directory '%s' %s." % (self.mainDir, absn))
            log.error("cpc-worker must be able to write in a temporary directory")
            log.error("Run cpc-worker from a user-writeable directory (e.g. /tmp)")
            raise WorkerError("Can't create directory '%s' %s." % (self.mainDir, absn))
        self.heartbeat = heartbeat.HeartbeatSender(self, self.runCondVar)  # self.id, self.mainDir,
        # First get our architecture(s) (hw + sw) from the plugin
        self.plugin = PlatformPlugin(self.type, self.mainDir, self.conf)
        canRun = self.plugin.canRun()
        if not canRun:
            raise WorkerError("Plugin can't run.")
        self.platforms = self._getPlatforms(self.plugin)

        # we make copies to be able to subtract our usage off the max resources
        self.remainingPlatforms = copy.deepcopy(self.platforms)
        # Then check what executables we already have
        self._getExecutables()
        # start without workloads

        if len(self.exelist.findAllByPlattform(self.type)) == 0:
            print "No executables found for platform %s" % self.type
            sys.exit(1)

        self._printAvailableExes()
        self.workloads = []
        self.iteration = 0
        self.acceptCommands = True
        # install the signal handler
        signalHandlerAddWorker(self)

    def getID(self):
        return self.id

    def getWorkerDir(self):
        return self.mainDir

    def _getWorkloads(self):
        """Get the list of workloads, assuming a locked runCondVar."""
        return self.workloads

    def killWorkload(self, cmdID):
        """Kill a workload by command ID."""
        with self.runCondVar:
            try:
                # first try to find it
                workload = None
                for nworkload in self.workloads:
                    if nworkload.cmd.id == cmdID:
                        workload = nworkload
                        break
                if workload is not None:
                    log.info("Found workload %s to kill" % cmdID)
                    workload.killLocked()
                else:
                    log.debug("Error killWorkload %s not found" % cmdID)
            except:
                log.debug("Error in killWorkload %s" % cmdID)

    def run(self):
        """Ask for tasks until told to quit."""
        noWorkSeconds = 0
        while not self.quit:
            # send a request for a command to run
            startWaitingTime = time.time()
            with self.runCondVar:
                acceptCommands = self.acceptCommands
            if acceptCommands:
                resp = self._obtainCommands()
                # and extract the command and run directory
                workloads = self._extractCommands(resp)
                log.info("Got %d commands." % len(workloads))
                for workload in workloads:
                    log.info("cmd ID=%s" % workload.cmd.id)
                    # Check whether we have the neccesary executable.
                    # (If not, we need to ask for it)
                    if workload.executable is None:
                        # TODO: implement getting the executable
                        log.error("Found no executable!")
                        raise WorkerError("Executable not found")
                    workload.reservePlatform()
                if len(workloads) > 0:
                    # We first prepare
                    self._prepareWorkloads(workloads)
                    # add the new workloads to our lists
                    self.workloads.extend(workloads)
                    self.heartbeat.addWorkloads(workloads)
                    # Now we run.
                    log.debug("Running workloads: %d" % len(self.workloads))
                    # hb=heartbeat.Heartbeat(cmd.id, origServer, rundir)
                    # just before starting to run, we again check whether we
                    # should.
                    with self.runCondVar:
                        acceptCommands = self.acceptCommands
                    if acceptCommands:
                        for workload in workloads:
                            workload.run(self.plugin, self.args)
            # now wait until a workload finishes
            finishedWorkloads = []
            self.runCondVar.acquire()
            continueWaiting = True
            while continueWaiting:
                haveFinishedWorkloads = False
                for workload in self.workloads:
                    if not workload.running:
                        haveFinishedWorkloads = True
                        break
                if self.acceptCommands and not haveFinishedWorkloads:
                    haveRemainingResources = self._haveRemainingResources()
                    if haveRemainingResources:
                        log.info("Have free resources. Waiting 30 seconds")
                        self.runCondVar.wait(30)
                        continueWaiting = False
                    else:
                        # we can't ask for new jobs, so we wait indefinitely
                        self.runCondVar.wait()
                else:
                    continueWaiting = False
                # now sleep one second to make sure that jobs stopping around
                # the same time are reported back at once.
                time.sleep(1)
                # loop over all workloads
                for workload in self.workloads:
                    if not workload.running:
                        finishedWorkloads.append(workload)
                        log.info("Command id %s finished" % workload.cmd.id)
                        continueWaiting = False
                    # else:
                    #    log.debug("Command %s still running"%workload.cmd.id)
                # log.debug("End of waiting loop")
            # log.debug("Out of waiting loop")
            self.runCondVar.release()
            stopWaitingTime = time.time()
            # check whether there was work to do. If not, start counting
            # the amount of time we waited.
            if len(self.workloads) == 0:
                noWorkSeconds += stopWaitingTime - startWaitingTime
                if (self.quitSeconds is not None) and (noWorkSeconds > self.quitSeconds):
                    with self.runCondVar:
                        # signal quit.
                        self.acceptCommands = False
            else:
                noWorkSeconds = 0
            # now deal with finished workloads.
            for workload in finishedWorkloads:
                workload.finish(self.plugin, self.args)
                workload.returnResults()
                workload.releasePlatform()
            if len(finishedWorkloads) > 0:
                self.heartbeat.delWorkloads(finishedWorkloads)
                for workload in finishedWorkloads:
                    self.workloads.remove(workload)
            with self.runCondVar:
                acceptCommands = self.acceptCommands
            if not acceptCommands and len(self.workloads) == 0:
                self.quit = True
        self.heartbeat.stop()

    def cleanup(self):
        shutil.rmtree(self.mainDir)
        # now clean up the worker top dir. This might be in use by other workers
        # so we use rmDir
        try:
            if self.workerTopDirCreated:
                os.rmdir(self.workerTopDir)
        except:
            log.debug("Couldn't erase worker top dir %s" % self.workerTopDir)

    def _printAvailableExes(self):

        print "Available executables for platform %s:" % self.type
        for exe in self.exelist.findAllByPlattform(self.type):
            print "%s %s" % (exe.name, exe.version.getStr())

    def _getPlatforms(self, plugin):
        """Get the list of platforms als an XML string from the run plugin."""
        # make an empty platform reservation
        plr = PlatformReservation(self.mainDir)
        plugin_retmsg = plugin.run(".", "platform", self.args, str(plr.printXML()))
        if plugin_retmsg[0] != 0:
            log.error("Platform plugin failed: %s" % plugin_retmsg[1])
            raise WorkerError("Platform plugin failed: %s" % plugin_retmsg[1])
        log.debug("From platform plugin, platform cmd: '%s'" % plugin_retmsg[1])
        pfr = cpc.command.PlatformReader()
        # we also parse it for later.
        pfr.readString(plugin_retmsg[1], ("Platform description from platform plugin %s" % plugin.name))
        platforms = pfr.getPlatforms()
        return platforms

    def _getExecutables(self):
        """Get a list of executables as an ExecutableList object."""
        execdirs = self.conf.getExecutablesPath()
        self.exelist = cpc.command.ExecutableList()
        for execdir in execdirs:
            self.exelist.readDir(execdir, self.platforms)
        self.exelist.genIDs()

        log.debug("Found %d executables." % (len(self.exelist.executables)))

    def _obtainCommands(self):
        """Obtain a command from the up-most server given a list of
           platforms and exelist. Returns the client response object."""
        # Send a run request with our arch+binaries
        req = u'<?xml version="1.0"?>\n'
        req += u"<worker-request>\n"
        req += u"<worker-arch-capabilities>\n"
        for platform in self.remainingPlatforms:
            req += platform.printXML()
        req += "\n"
        req += self.exelist.printPartialXML()
        req += u"\n</worker-arch-capabilities>"
        # Append optional project specifics
        req += u"\n<worker-requirements>\n"
        if "project" in self.opts:
            req += u'  <option key="project" value="%s"/>\n' % self.opts["project"]
        req += u"</worker-requirements>\n"
        req += u"</worker-request>\n"
        log.debug("request string is: %s" % req)
        runreq_clnt = WorkerMessage()
        resp = runreq_clnt.workerRequest(self.id, req)
        # print "Got %s"%(resp.read(len(resp)))
        return resp

    def _extractCommands(self, resp):
        """Extract a command and a run directory from a server response.
            Returns a list of Workloads."""
        workloads = []
        log.debug("Response type=%s" % resp.getType())
        if resp.getType() == "application/x-tar":
            if resp.headers.has_key("originating-server-id"):
                origServer = resp.headers["originating-server-id"]
            else:
                raise WorkerError(
                    "No originating-server-id not found in "
                    "header, "
                    "The worker will not now where this "
                    "workload is coming from"
                )

            log.debug("Originating server: %s" % origServer)
            rundir = os.path.join(self.mainDir, "%d" % self.iteration)
            log.debug("run directory: %s" % rundir)
            # os.mkdir(rundir)
            cpc.util.file.extractSafely(rundir, fileobj=resp.getRawData())
            # get the commands.
            i = 0
            for subdir in os.listdir(rundir):
                cmddir = os.path.join(rundir, subdir)
                if os.path.exists(os.path.join(cmddir, "command.xml")):
                    log.debug("trying command directory: %s" % cmddir)
                    # there is a command here. Get the command.
                    cr = cpc.command.CommandReader()
                    commandFilename = os.path.join(cmddir, "command.xml")
                    cr.read(commandFilename)
                    # write log
                    inf = open(commandFilename, "r")
                    log.debug("Received job. Command is: %s" % inf.read())
                    inf.close()
                    cmd = cr.getCommands()[0]
                    (exe, pf) = self._findExecutable(cmd)
                    if exe is None:
                        raise WorkerError("Executable not found")
                    id = "%d/%d" % (self.iteration, i)
                    workloads.append(
                        workload.WorkLoad(self.mainDir, cmd, cmddir, origServer, exe, pf, id, self.runCondVar)
                    )
                    i += 1
            resp.close()
        self.iteration += 1
        return workloads

    def _findExecutable(self, cmd):
        """Find the right executable for a command given the list of platforms.
           cmd = the command
           returns tuple with the executable and the platform
           """
        for platform in self.remainingPlatforms:
            # we iterate in the order we got from the run plugin. This
            # might be important: it should return in the order it thinks
            # goes from most to least optimal.

            log.debug("Using platform %s for executable search" % platform.name)
            exe = self.exelist.find(cmd.executable, platform, cmd.minVersion, cmd.maxVersion)
            if exe is not None:
                log.debug("Found matching executable")
                return (exe, platform)
        return (None, None)

    def _prepareWorkloads(self, workloadlist):
        """Prepare the workloads (by joining, for example)."""
        # do a join
        joinableWorkloads = []
        for workload in workloadlist:
            if workload.platform.isJoinPrefered() and workload.executable.isJoinable():
                joinableWorkloads.append(workload)
        while len(joinableWorkloads) > 0:
            joinTo = joinableWorkloads[0]
            joinableWorkloads.remove(joinTo)
            join = []
            for i in xrange(len(joinableWorkloads)):
                if joinTo.canJoin(joinableWorkloads[i]):
                    log.debug("Joining command %s and %s" % (joinTo.cmd.id, joinableWorkloads[i].cmd.id))
                    join.append(joinableWorkloads[i])
            # and do the actual joining
            if len(join) > 0:
                joinTo.join(join)
                for j in join:
                    # now remove those from the original lists
                    joinableWorkloads.remove(j)
                    workloadlist.remove(j)

    def _haveRemainingResources(self):
        """Check whether any of the resources has been depleted.
           returns: True if none of the resources have been depleted, False
                    otherwise
           """
        for platform in self.remainingPlatforms:
            for rsrc in platform.getMaxResources().itervalues():
                if rsrc.value <= 0:
                    return False
        return True

    def shutdown(self):
        """Shut down this worker cleanly. This must be called from a thread,
           not directly from a signal handler."""
        # log.log(cpc.util.log.TRACE,"Received shutdown signal")
        log.debug("Received shutdown signal")
        # now set the variable and notify
        self.runCondVar.acquire()
        self.acceptCommands = False
        self.runCondVar.notifyAll()
        self.runCondVar.release()
        log.debug("release")
예제 #4
0
class Worker(object):
    """The worker class creates a worker client that contacts a server
       and asks for tasks."""
    def __init__(self,
                 cf,
                 opts,
                 type,
                 args,
                 workdir,
                 quitMinutes,
                 sharedWorker=False):
        """Initialize the object as a client, given a configuration.
           cf =  the configuration class
		   opts = dictionary with options
           type = the worker plugin name to start.
           args = arguments for the worker plugin."""
        # the run lock and condition variable
        self.runLock = threading.Lock()
        self.runCondVar = threading.Condition(self.runLock)
        self.conf = cf
        self.opts = opts
        self.type = type
        self.args = args
        self.quit = False
        self.isShared = sharedWorker
        self.id = "%s-%d" % (self.conf.getHostName(), os.getpid())
        # the number of seconds after which to quit if there is no work:
        self.quitSeconds = None
        if quitMinutes is not None:
            self.quitSeconds = 60 * quitMinutes
        log.info("Worker ID: %s." % self.id)
        # Process (untar) the run request into a directory name
        # that is unique for this process+hostname, and worker job iteration
        if workdir is None:
            self.workerTopDir = self.conf.getRunDir()
            createTopDir = True
            #self.mainDir=os.path.join(self.conf.getRunDir(), self.id)
        else:
            if not os.path.exists(workdir):
                raise WorkerError("Given run directory %s does not exist." %
                                  (workdir))
            self.workerTopDir = workdir
            createTopDir = False

        # fix the path so the server knows where it is
        if not os.path.isabs(self.workerTopDir):
            self.workerTopDir = os.path.join(os.getcwd(), self.workerTopDir)
            isabs = False
        else:
            isabs = True

        # set the actual path being used.
        self.mainDir = os.path.join(self.workerTopDir, self.id)

        try:
            log.debug("Creating work directory %s" % (self.mainDir))
            if createTopDir:
                if not os.path.exists(self.workerTopDir):
                    os.mkdir(self.workerTopDir)
                self.workerTopDirCreated = True
            else:
                self.workerTopDirCreated = False
            os.mkdir(self.mainDir)
        except:
            if isabs:
                absn = ""
            else:
                absn = "in the current working directory"
            log.error("Can't create the directory '%s' %s." %
                      (self.mainDir, absn))
            log.error(
                "cpc-worker must be able to write in a temporary directory")
            log.error(
                "Run cpc-worker from a user-writeable directory (e.g. /tmp)")
            raise WorkerError("Can't create directory '%s' %s." %
                              (self.mainDir, absn))
        self.heartbeat = heartbeat.HeartbeatSender(
            self,  #self.id, self.mainDir,
            self.runCondVar)
        # First get our architecture(s) (hw + sw) from the plugin
        self.plugin = PlatformPlugin(self.type, self.mainDir, self.conf)
        canRun = self.plugin.canRun()
        if not canRun:
            raise WorkerError("Plugin can't run.")
        self.platforms = self._getPlatforms(self.plugin)

        # we make copies to be able to subtract our usage off the max resources
        self.remainingPlatforms = copy.deepcopy(self.platforms)
        # Then check what executables we already have
        self._getExecutables()
        # start without workloads

        if len(self.exelist.findAllByPlattform(self.type)) == 0:
            print "No executables found for platform %s" % self.type
            sys.exit(1)

        self._printAvailableExes()
        self.workloads = []
        self.iteration = 0
        self.acceptCommands = True
        # install the signal handler
        signalHandlerAddWorker(self)

    def getID(self):
        return self.id

    def getWorkerDir(self):
        return self.mainDir

    def _getWorkloads(self):
        """Get the list of workloads, assuming a locked runCondVar."""
        return self.workloads

    def killWorkload(self, cmdID):
        """Kill a workload by command ID."""
        with self.runCondVar:
            try:
                # first try to find it
                workload = None
                for nworkload in self.workloads:
                    if nworkload.cmd.id == cmdID:
                        workload = nworkload
                        break
                if workload is not None:
                    log.info("Found workload %s to kill" % cmdID)
                    workload.killLocked()
                else:
                    log.debug("Error killWorkload %s not found" % cmdID)
            except:
                log.debug("Error in killWorkload %s" % cmdID)

    def run(self):
        """Ask for tasks until told to quit."""
        noWorkSeconds = 0
        while not self.quit:
            # send a request for a command to run
            startWaitingTime = time.time()
            with self.runCondVar:
                acceptCommands = self.acceptCommands
            if acceptCommands:
                resp = self._obtainCommands()
                # and extract the command and run directory
                workloads = self._extractCommands(resp)
                log.info("Got %d commands." % len(workloads))
                for workload in workloads:
                    log.info("cmd ID=%s" % workload.cmd.id)
                    # Check whether we have the neccesary executable.
                    # (If not, we need to ask for it)
                    if workload.executable is None:
                        # TODO: implement getting the executable
                        log.error("Found no executable!")
                        raise WorkerError("Executable not found")
                    workload.reservePlatform()
                if len(workloads) > 0:
                    # We first prepare
                    self._prepareWorkloads(workloads)
                    # add the new workloads to our lists
                    self.workloads.extend(workloads)
                    self.heartbeat.addWorkloads(workloads)
                    # Now we run.
                    log.debug("Running workloads: %d" % len(self.workloads))
                    #hb=heartbeat.Heartbeat(cmd.id, origServer, rundir)
                    # just before starting to run, we again check whether we
                    # should.
                    with self.runCondVar:
                        acceptCommands = self.acceptCommands
                    if acceptCommands:
                        for workload in workloads:
                            workload.run(self.plugin, self.args)
            # now wait until a workload finishes
            finishedWorkloads = []
            self.runCondVar.acquire()
            continueWaiting = True
            while continueWaiting:
                haveFinishedWorkloads = False
                for workload in self.workloads:
                    if not workload.running:
                        haveFinishedWorkloads = True
                        break
                if self.acceptCommands and not haveFinishedWorkloads:
                    haveRemainingResources = self._haveRemainingResources()
                    if haveRemainingResources:
                        log.info("Have free resources. Waiting 30 seconds")
                        self.runCondVar.wait(30)
                        continueWaiting = False
                    else:
                        # we can't ask for new jobs, so we wait indefinitely
                        self.runCondVar.wait()
                else:
                    continueWaiting = False
                # now sleep one second to make sure that jobs stopping around
                # the same time are reported back at once.
                time.sleep(1)
                # loop over all workloads
                for workload in self.workloads:
                    if not workload.running:
                        finishedWorkloads.append(workload)
                        log.info("Command id %s finished" % workload.cmd.id)
                        continueWaiting = False
                    #else:
                    #    log.debug("Command %s still running"%workload.cmd.id)
                #log.debug("End of waiting loop")
            #log.debug("Out of waiting loop")
            self.runCondVar.release()
            stopWaitingTime = time.time()
            # check whether there was work to do. If not, start counting
            # the amount of time we waited.
            if len(self.workloads) == 0:
                noWorkSeconds += stopWaitingTime - startWaitingTime
                if ((self.quitSeconds is not None)
                        and (noWorkSeconds > self.quitSeconds)):
                    with self.runCondVar:
                        # signal quit.
                        self.acceptCommands = False
            else:
                noWorkSeconds = 0
            # now deal with finished workloads.
            for workload in finishedWorkloads:
                workload.finish(self.plugin, self.args)
                workload.returnResults()
                workload.releasePlatform()
            if len(finishedWorkloads) > 0:
                self.heartbeat.delWorkloads(finishedWorkloads)
                for workload in finishedWorkloads:
                    self.workloads.remove(workload)
            with self.runCondVar:
                acceptCommands = self.acceptCommands
            if not acceptCommands and len(self.workloads) == 0:
                self.quit = True
        self.heartbeat.stop()

    def cleanup(self):
        shutil.rmtree(self.mainDir)
        # now clean up the worker top dir. This might be in use by other workers
        # so we use rmDir
        try:
            if self.workerTopDirCreated:
                os.rmdir(self.workerTopDir)
        except:
            log.debug("Couldn't erase worker top dir %s" % self.workerTopDir)

    def _printAvailableExes(self):

        print "Available executables for platform %s:" % self.type
        for exe in self.exelist.findAllByPlattform(self.type):
            print "%s %s" % (exe.name, exe.version.getStr())

    def _getPlatforms(self, plugin):
        """Get the list of platforms als an XML string from the run plugin."""
        # make an empty platform reservation
        plr = PlatformReservation(self.mainDir)
        plugin_retmsg = plugin.run(".", "platform", self.args,
                                   str(plr.printXML()))
        if plugin_retmsg[0] != 0:
            log.error("Platform plugin failed: %s" % plugin_retmsg[1])
            raise WorkerError("Platform plugin failed: %s" % plugin_retmsg[1])
        log.debug("From platform plugin, platform cmd: '%s'" %
                  plugin_retmsg[1])
        pfr = cpc.command.PlatformReader()
        # we also parse it for later.
        pfr.readString(
            plugin_retmsg[1],
            ("Platform description from platform plugin %s" % plugin.name))
        platforms = pfr.getPlatforms()
        return platforms

    def _getExecutables(self):
        """Get a list of executables as an ExecutableList object."""
        execdirs = self.conf.getExecutablesPath()
        self.exelist = cpc.command.ExecutableList()
        for execdir in execdirs:
            self.exelist.readDir(execdir, self.platforms)
        self.exelist.genIDs()

        log.debug("Found %d executables." % (len(self.exelist.executables)))

    def _obtainCommands(self):
        """Obtain a command from the up-most server given a list of
           platforms and exelist. Returns the client response object."""
        # Send a run request with our arch+binaries
        req = u'<?xml version="1.0"?>\n'
        req += u'<worker-request>\n'
        req += u'<worker-arch-capabilities>\n'
        for platform in self.remainingPlatforms:
            req += platform.printXML()
        req += '\n'
        req += self.exelist.printPartialXML()
        req += u'\n</worker-arch-capabilities>'
        #Append optional project specifics
        req += u'\n<worker-requirements>\n'
        if "project" in self.opts:
            req += u'  <option key="project" value="%s"/>\n' % self.opts[
                'project']
        req += u'</worker-requirements>\n'
        req += u'</worker-request>\n'
        log.debug('request string is: %s' % req)
        runreq_clnt = WorkerMessage()
        resp = runreq_clnt.workerRequest(self.id, req)
        #print "Got %s"%(resp.read(len(resp)))
        return resp

    def _extractCommands(self, resp):
        """Extract a command and a run directory from a server response.
            Returns a list of Workloads."""
        workloads = []
        log.debug("Response type=%s" % resp.getType())
        if resp.getType() == "application/x-tar":
            if resp.headers.has_key('originating-server-id'):
                origServer = resp.headers['originating-server-id']
            else:
                raise WorkerError("No originating-server-id not found in "
                                  "header, "
                                  "The worker will not now where this "
                                  "workload is coming from")

            log.debug("Originating server: %s" % origServer)
            rundir = os.path.join(self.mainDir, "%d" % self.iteration)
            log.debug("run directory: %s" % rundir)
            #os.mkdir(rundir)
            cpc.util.file.extractSafely(rundir, fileobj=resp.getRawData())
            # get the commands.
            i = 0
            for subdir in os.listdir(rundir):
                cmddir = os.path.join(rundir, subdir)
                if os.path.exists(os.path.join(cmddir, "command.xml")):
                    log.debug("trying command directory: %s" % cmddir)
                    # there is a command here. Get the command.
                    cr = cpc.command.CommandReader()
                    commandFilename = os.path.join(cmddir, "command.xml")
                    cr.read(commandFilename)
                    # write log
                    inf = open(commandFilename, "r")
                    log.debug("Received job. Command is: %s" % inf.read())
                    inf.close()
                    cmd = cr.getCommands()[0]
                    (exe, pf) = self._findExecutable(cmd)
                    if (exe is None):
                        raise WorkerError("Executable not found")
                    id = "%d/%d" % (self.iteration, i)
                    workloads.append(
                        workload.WorkLoad(self.mainDir, cmd, cmddir,
                                          origServer, exe, pf, id,
                                          self.runCondVar))
                    i += 1
            resp.close()
        self.iteration += 1
        return workloads

    def _findExecutable(self, cmd):
        """Find the right executable for a command given the list of platforms.
           cmd = the command
           returns tuple with the executable and the platform
           """
        for platform in self.remainingPlatforms:
            # we iterate in the order we got from the run plugin. This
            # might be important: it should return in the order it thinks
            # goes from most to least optimal.

            log.debug("Using platform %s for executable search" %
                      platform.name)
            exe = self.exelist.find(cmd.executable, platform, cmd.minVersion,
                                    cmd.maxVersion)
            if exe is not None:
                log.debug("Found matching executable")
                return (exe, platform)
        return (None, None)

    def _prepareWorkloads(self, workloadlist):
        """Prepare the workloads (by joining, for example)."""
        # do a join
        joinableWorkloads = []
        for workload in workloadlist:
            if (workload.platform.isJoinPrefered()
                    and workload.executable.isJoinable()):
                joinableWorkloads.append(workload)
        while len(joinableWorkloads) > 0:
            joinTo = joinableWorkloads[0]
            joinableWorkloads.remove(joinTo)
            join = []
            for i in xrange(len(joinableWorkloads)):
                if joinTo.canJoin(joinableWorkloads[i]):
                    log.debug("Joining command %s and %s" %
                              (joinTo.cmd.id, joinableWorkloads[i].cmd.id))
                    join.append(joinableWorkloads[i])
            # and do the actual joining
            if len(join) > 0:
                joinTo.join(join)
                for j in join:
                    # now remove those from the original lists
                    joinableWorkloads.remove(j)
                    workloadlist.remove(j)

    def _haveRemainingResources(self):
        """Check whether any of the resources has been depleted.
           returns: True if none of the resources have been depleted, False
                    otherwise
           """
        for platform in self.remainingPlatforms:
            for rsrc in platform.getMaxResources().itervalues():
                if rsrc.value <= 0:
                    return False
        return True

    def shutdown(self):
        """Shut down this worker cleanly. This must be called from a thread,
           not directly from a signal handler."""
        #log.log(cpc.util.log.TRACE,"Received shutdown signal")
        log.debug("Received shutdown signal")
        # now set the variable and notify
        self.runCondVar.acquire()
        self.acceptCommands = False
        self.runCondVar.notifyAll()
        self.runCondVar.release()
        log.debug("release")