def registerToMaster(self):
try:
bootmaster = StandingCall(copy_ref_with_timeout(self.master,config.BOOTSTRAP_CONTACT_TIMEOUT), 1, max_repeat=config.BOOTSTRAP_CONTACT_REPEAT)
self.wid,file_server_ior = bootmaster.registerWorker(self.uuid,1) # ordinary workers have capacity 1
file_server = self.program.orb.string_to_object(file_server_ior)
from diane.FileTransfer import FileTransferClient
self.ftc = FileTransferClient(file_server,self.uuid)
import diane.workspace
# collect and upload vcard
vcard = self.make_vcard()
file('vcard.txt','w').write(repr(vcard))
self.ftc.upload('vcard.txt',os.path.join(diane.workspace.workerSubdir(self.wid),'vcard.txt'))
#TEST - multiple registrations
#for i in range(1000):
# self.wid = self.master.registerWorker(self._this())
#logger.info('master ping %s',str(self.master.ping()))
self.heartbeat_thread = HeartbeatThread(self,self.master)
self.heartbeat_thread.start()
except CORBA.TRANSIENT:
logger.error('unable to establish connection to the master after timeout=%d seconds'%config.BOOTSTRAP_CONTACT_TIMEOUT)
raise
except DIANE_CORBA.XHangup:
logger.warning('I was refused a registration (my peer_id=%s)',self.uuid)
raise
def run(self):
import MSGWrap
from diane.config import log_configuration
log_configuration(title='initial configuration')
msg_data = { '_worker_uuid' : self.uuid }
try:
self.registerToMaster()
master = StandingCall(self.master, config.HEARTBEAT_DELAY, should_stop = self.should_stop)
app_boot,app_init = master.get_init_data(self.uuid) #(config.HEARTBEAT_DELAY,-1,self.should_stop,self.master,'get_init_data',self.uuid)
_boot = streamer.loads(app_boot)
msg_data['_master_uuid'] = _boot.master_uuid
msg_data['_runid'] = _boot.runid
import os
msg_data['ganga_job_uuid'] = self.ganga_job_uuid
# FIXME: if worker restart enabled, save diane.config.__all_configs and restore it after run has finished
MSGWrap.sendStatus('_worker_create_application_proxy_start', msg_data)
self.application = create_application_proxy(app_boot,app_init,agent=self)
MSGWrap.sendStatus('_worker_create_application_proxy_finish', msg_data)
self.program.registerAtExitHandler(self.finalize_application)
MSGWrap.sendStatus('_worker_initialize_start', msg_data)
app_init_output = self.application.initialize(app_init)
MSGWrap.sendStatus('_worker_initialize_finish', msg_data)
# config may have been updated and the value of config.HEARTBEAT_DELAY may have changed -> need to create the object again
# FIXME: use a REFERENCE to config.HEARTBEAT_DELAY
master = StandingCall(self.master, config.HEARTBEAT_DELAY, should_stop = self.should_stop)
master.put_init_result(self.uuid,app_init_output,0) #(config.HEARTBEAT_DELAY,-1,self.should_stop,self.master,'put_init_result',self.uuid,app_init_output,0)
while not self.should_stop():
time.sleep(config.PULL_REQUEST_DELAY) # PENDING: this parameter should be dynamically controlled by the master
tid,task_data = master.get_task_data(self.uuid) #(config.HEARTBEAT_DELAY,-1,self.should_stop,self.master,'get_task_data',self.uuid)
try:
msg_data['tid'] = tid
MSGWrap.sendStatus('_worker_do_work_start', msg_data)
task_result = self.application.do_work(task_data)
MSGWrap.sendStatus('_worker_do_work_finish', msg_data)
error = 0
except diane.application.ApplicationFailure,x: # recoverable problem
task_result = streamer.dumps(x)
error = 1
#FIXME: reporting failure is not yet well-defined
master.put_task_result(self.uuid,tid,task_result,error) #(config.HEARTBEAT_DELAY,-1,self.should_stop,self.master,'put_task_result',self.uuid,tid,task_result,error)
except diane.application.ApplicationFailure,x: # recoverable problem but raised by the application init
pass
