Import tensorflow AS TF the FLAGS = tf.app.flags.FLAGS tf.app.flags.DEFINE_string ( " job_name " , " " , " type ps or start a service worker " ) tf.app.flags.DEFINE_integer ( " task_index " , 0, " specify ps worker or to which server among Task: 0, Task:. 1 " ) DEF main (the argv): # define a global count op, to list the number of training steps in using the hook global_step = TF. contrib.framework.get_or_create_global_step () # specific cluster described object targeting rules worker ps, ps, or more than one worker, a first station: / job: worker / task: 0, a second: / job: worker / task: 1 , ps, too tf.train.ClusterSpec = Cluster ({ " PS " : [ " 192.168.0.4:2222 " ,], " worker " : [ " 192.168.109.128:2323 " ,]}) # create different service ps worker, job_name designated ps is still worker, task_index, specify which servers start server = tf.train.Server (Cluster, job_name = FLAGS.job_name, task_index = FLAGS.task_index) # do different things based on different servers, ps save the parameters, worker designated computing devices running model IF FLAGS.job_name == ' PS ' : # parameter parameter server only accepts server.join () the else : worker_device= " / Job: worker / Task: 0 / CPU: 0 / " # Specify the device to run with tf.device (tf.train.replica_device_setter (worker_device = worker_device, Cluster = Cluster)): # demonstrates a matrix multiplication operation x = tf.Variable ([[. 1, 2,. 3,. 4 ]]) W = tf.Variable ([[2], [. 4], [. 5], [. 7 ]]) MAT = tf.matmul (X, W) # create a distributed session with tf.train.MonitoredTrainingSession ( master = " GRPC: //192.168.0.1: 2222 " , # specify whether it is the main Work is_chief = (FLAGS.task_index == 0), # Determines whether the book is a front worker config = tf.ConfigProto (log_device_placement = True), # printing apparatus information Hooks = [tf.train.StopAtStepHook (last_step = 1000)] # Specifies the number of training steps, the specified number of steps needed to define a global count the OP ) AS mon_sess: the while not mon_sess.should_stop (): # should_stops is abnormal stop mon_sess.run (MAT) iF __name__ == " __main__ " : tf.app.run ()