def client_proc(job_id, data_file, rtask, task=None):
# send input file to rtask.location; this will be saved to dispycos process's
# working directory
if (yield pycos.Pycos().send_file(rtask.location, data_file,
timeout=10)) < 0:
print('Could not send input data to %s' % rtask.location)
# terminate remote task
rtask.send(None)
raise StopIteration(-1)
# send info about input
obj = C(job_id, data_file, random.uniform(5, 8), task)
if (yield rtask.deliver(obj)) != 1:
print('Could not send input to %s' % rtask.location)
raise StopIteration(-1)
# rtask sends result to this task as message
result = yield task.receive()
if not result.result_file:
print('Processing %s failed' % obj.i)
raise StopIteration(-1)
# rtask saves results file at this client, which is saved in pycos's
# dest_path, not current working directory!
result_file = os.path.join(pycos.Pycos().dest_path, result.result_file)
# move file to cwd
target = os.path.join(os.getcwd(), os.path.basename(result_file))
os.rename(result_file, target)
print(' job %s output is in %s' % (obj.i, target))
def shutdown(self, wait=True):
"""This method should be called by user program to close the
http server. If 'wait' is True the server waits for poll_sec
so the http client gets all the updates before server is
closed.
"""
if wait:
pycos.logger.info(
'HTTP server waiting for %s seconds for client updates '
'before quitting', self._poll_sec)
if pycos.Pycos().cur_task():
def _shutdown(task=None):
yield task.sleep(self._poll_sec + 0.5)
self._server.shutdown()
self._server.server_close()
pycos.Task(_shutdown)
else:
time.sleep(self._poll_sec + 0.5)
self._server.shutdown()
self._server.server_close()
else:
self._server.shutdown()
self._server.server_close()
def rtask_proc(task=None):
import os
# receive object from client_proc task
cobj = yield task.receive()
if not cobj:
raise StopIteration
# Input file is already copied at where this rtask is running (by client).
# For given input file, create an output file with each line in the output
# file computed as length of corresponding line in input file
cobj.result_file = 'result-%s' % cobj.data_file
with open(cobj.data_file, 'r') as data_fd:
with open(cobj.result_file, 'w') as result_fd:
for lineno, line in enumerate(data_fd, start=1):
result_fd.write('%d: %d\n' % (lineno, len(line) - 1))
# 'sleep' to simulate computing
yield task.sleep(cobj.n)
# transfer the result file to client
status = yield pycos.Pycos().send_file(cobj.client.location,
cobj.result_file,
overwrite=True,
timeout=30)
if status:
print('Could not send %s to %s' %
(cobj.result_file, cobj.client.location))
cobj.result_file = None
cobj.client.send(cobj)
os.remove(cobj.data_file)
os.remove(cobj.result_file)
def server_available(location, data_file, task=None):
import os
# 'server_available' is executed locally (at client) when a server process
# is available. 'location' is Location instance of server. When this task is
# executed, 'depends' of computation would've been transferred. data_file
# could've been sent with the computation 'depends'; however, to illustrate
# how files can be sent separately (to distribute data fragments among
# servers), files are transferred to servers in this example
print(' Sending %s to %s' % (data_file, location))
if (yield pycos.Pycos().send_file(
location, data_file, timeout=5, overwrite=True)) < 0:
print('Could not send data file "%s" to %s' % (data_file, location))
raise StopIteration(-1)
# 'setup_server' is executed on remote server at 'location' with argument
# data_file
yield computation.enable_server(location, data_file)
raise StopIteration(0)
def client_proc(computation, njobs, task=None):
# schedule computation with the scheduler; scheduler accepts one computation
# at a time, so if scheduler is shared, the computation is queued until it
# is done with already scheduled computations
if (yield computation.schedule()):
raise Exception('Could not schedule computation')
# pair EC2 node with this client with:
yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 9706))
# if multiple nodes are used, 'broadcast' option can be used to pair with
# all nodes with just one statement as:
# yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 9706), broadcast=True)
# execute n jobs (tasks) and get their results. Note that number of
# jobs created can be more than number of server processes available; the
# scheduler will use as many processes as necessary/available, running one
# job at a server process
args = [(i, random.uniform(3, 10)) for i in range(njobs)]
results = yield computation.run_results(compute, args)
for result in results:
print('job %s result: %s' % (result[0], result[1]))
yield computation.close()
def node_available(avail_info, data_file, task=None):
import os
# 'node_available' is executed locally (at client) when a node is
# available. 'location' is Location instance of node. When this task is
# executed, 'depends' of computation would've been transferred.
# data_file could've been sent with the computation 'depends'; however, to
# illustrate how files can be sent separately (e.g., to transfer different
# files to different nodes), file is transferred with 'node_available'.
print(' Sending %s to %s' % (data_file, avail_info.location.addr))
sent = yield pycos.Pycos().send_file(avail_info.location,
data_file,
overwrite=True,
timeout=5)
if (sent < 0):
print('Could not send data file "%s" to %s' %
(data_file, avail_info.location))
raise StopIteration(-1)
# value of task (last value yield'ed or value of 'raise StopIteration') will
# be passed to node_setup as argument(s).
yield computation.enable_node(avail_info.location.addr, data_file)
# pair EC2 node with this client with:
yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 9706))
# if multiple nodes are used, 'broadcast' option can be used to pair with
# all nodes with just one statement as:
# yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 9706), broadcast=True)
# execute n jobs (tasks) and get their results. Note that number of
# jobs created can be more than number of server processes available; the
# scheduler will use as many processes as necessary/available, running one
# job at a server process
args = [(i, random.uniform(3, 10)) for i in range(njobs)]
results = yield computation.run_results(compute, args)
for result in results:
print('job %s result: %s' % (result[0], result[1]))
yield computation.close()
if __name__ == '__main__':
import sys, random
# pycos.logger.setLevel(pycos.Logger.DEBUG)
pycos.Pycos(node='127.0.0.1', tcp_port=9705, udp_port=9705)
njobs = 10 if len(sys.argv) == 1 else int(sys.argv[1])
# if scheduler is not already running (on a node as a program),
# start private scheduler:
Scheduler()
# use 'compute' for computation jobs
computation = Computation([compute])
pycos.Task(client_proc, computation, njobs)
def rti_1(a, b=1, task=None):
pycos.logger.debug('running %s/%s with %s, %s', task.name, id(task), a, b)
msg = yield task.receive()
if b % 2 == 0:
yield task.sleep(b)
pycos.logger.debug('%s/%s done', task.name, id(task))
# (remote) monitor (if any) gets this exception (to be
# interpreted as normal termination)
raise StopIteration(msg)
else:
# (remote) monitor (if any) gets this exception, too
raise Exception('invalid invocation: %s' % b)
pycos.logger.setLevel(pycos.Logger.DEBUG)
# 'secret' is set so only peers that use same secret can communicate
scheduler = pycos.Pycos(name='server', secret='test')
# register rti_1 so remote clients can request execution
rti1 = pycos.RTI(rti_1)
rti1.register()
if sys.version_info.major > 2:
read_input = input
else:
read_input = raw_input
while True:
try:
line = read_input().strip().lower()
if line in ('quit', 'exit'):
break
except:
break
else:
pycos.logger.warning('ignoring invalid message')
def rti_test(task=None):
# if server is on remote network, automatic discovery won't work,
# so add it explicitly
# yield scheduler.peer('192.168.21.5')
# get reference to RTI at server
rti1 = yield pycos.RTI.locate('rti_1')
print('RTI is at %s' % rti1.location)
# 5 (remote) tasks are created with rti1
n = 5
# set monitor (monitor_proc task) for tasks created for this RTI
yield rti1.monitor(pycos.Task(monitor_proc, n))
for i in range(n):
rtask = yield rti1('test%s' % i, b=i)
pycos.logger.debug('RTI %s created' % rtask)
# If necessary, each rtask can also be set (different) 'monitor'
rtask.send('msg:%s' % i)
yield task.sleep(random.uniform(0, 1))
pycos.logger.setLevel(pycos.Logger.DEBUG)
# use 'test' secret so peers that use same secret are recognized
scheduler = pycos.Pycos(name='client', secret='test')
pycos.Task(rti_test)
# pair EC2 node with this client with:
yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 51347))
# if multiple nodes are used, 'broadcast' option can be used to pair with
# all nodes with just one statement as:
# yield pycos.Pycos().peer(pycos.Location('54.204.242.185', 51347), broadcast=True)
# execute n jobs (tasks) and get their results. Note that number of
# jobs created can be more than number of server processes available; the
# scheduler will use as many processes as necessary/available, running one
# job at a server process
args = [random.uniform(3, 10) for _ in range(njobs)]
results = yield computation.run_results(compute, args)
for result in results:
print('result: %s' % result)
yield computation.close()
if __name__ == '__main__':
import sys, random
# pycos.logger.setLevel(pycos.Logger.DEBUG)
pycos.Pycos(node='127.0.0.1', tcp_port=4567)
njobs = 10 if len(sys.argv) == 1 else int(sys.argv[1])
# if scheduler is not already running (on a node as a program),
# start private scheduler:
Scheduler()
# send 'compute' generator function
computation = Computation([compute])
pycos.Task(client_proc, computation, njobs)
