数据没有在 Tensorflow Worker 之间分发

Question

我编写了一个分布式 TensorFlow 程序，其中包含 1 ps 作业和 2 个 worker 作业。我期待数据批次在工人之间分发，但情况似乎并非如此。我看到只有一个工作人员(任务=0)正在做所有工作，而另一个工人处于空闲状态。您能否帮助我找到该程序的问题：

import math
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
# Flags for defining the tf.train.ClusterSpec
tf.app.flags.DEFINE_string("ps_hosts", "",
"Comma-separated list of hostname:port pairs")
tf.app.flags.DEFINE_string("worker_hosts", "",
"Comma-separated list of hostname:port pairs")
tf.app.flags.DEFINE_string("master_hosts", "oser502110:2222",
"Comma-separated list of hostname:port pairs")
# Flags for defining the tf.train.Server
tf.app.flags.DEFINE_string("job_name", "", "One of 'ps', 'worker'")
tf.app.flags.DEFINE_integer("task_index", 0, "Index of task within the job")
tf.app.flags.DEFINE_integer("hidden_units", 100,
"Number of units in the hidden layer of the NN")
tf.app.flags.DEFINE_string("data_dir", "/tmp/mnist-data",
"Directory for storing mnist data")
tf.app.flags.DEFINE_integer("batch_size", 100, "Training batch size")
tf.app.flags.DEFINE_string("worker_grpc_url", None,
"Worker GRPC URL")
FLAGS = tf.app.flags.FLAGS
IMAGE_PIXELS = 28
def main(_):
ps_hosts = FLAGS.ps_hosts.split(",")
worker_hosts = FLAGS.worker_hosts.split(",")
master_hosts = FLAGS.master_hosts.split(",")

cluster = tf.train.ClusterSpec({"ps": ps_hosts, "worker": worker_hosts})
# Create and start a server for the local task.
server = tf.train.Server(cluster,
job_name=FLAGS.job_name,
task_index=FLAGS.task_index)
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
is_chief = (FLAGS.task_index == 0)
if is_chief: tf.reset_default_graph()

# Assigns ops to the local worker by default.
with tf.device(tf.train.replica_device_setter(
worker_device="/job:worker/task:%d" % FLAGS.task_index,
cluster=cluster)):
# Variables of the hidden layer
hid_w = tf.Variable(
tf.truncated_normal([IMAGE_PIXELS * IMAGE_PIXELS, FLAGS.hidden_units],
stddev=1.0 / IMAGE_PIXELS), name="hid_w")
hid_b = tf.Variable(tf.zeros([FLAGS.hidden_units]), name="hid_b")
# Variables of the softmax layer
sm_w = tf.Variable(
tf.truncated_normal([FLAGS.hidden_units, 10],
stddev=1.0 / math.sqrt(FLAGS.hidden_units)),
name="sm_w")
sm_b = tf.Variable(tf.zeros([10]), name="sm_b")
x = tf.placeholder(tf.float32, [None, IMAGE_PIXELS * IMAGE_PIXELS])
y_ = tf.placeholder(tf.float32, [None, 10])
hid_lin = tf.nn.xw_plus_b(x, hid_w, hid_b)
hid = tf.nn.relu(hid_lin)
y = tf.nn.softmax(tf.nn.xw_plus_b(hid, sm_w, sm_b))
loss = -tf.reduce_sum(y_ * tf.log(tf.clip_by_value(y, 1e-10, 1.0)))
global_step = tf.Variable(0, trainable=False)
train_op = tf.train.AdagradOptimizer(0.01).minimize(
loss, global_step=global_step)
saver = tf.train.Saver()
#summary_op = tf.merge_all_summaries()
init_op = tf.initialize_all_variables()
# Create a "supervisor", which oversees the training process.
sv = tf.train.Supervisor(is_chief=is_chief,
logdir="/tmp/train_logs",
init_op=init_op,
recovery_wait_secs=1,
saver=saver,
global_step=global_step,
save_model_secs=600)
if is_chief:
print("Worker %d: Initializing session..." % FLAGS.task_index)
else:
print("Worker %d: Waiting for session to be initialized..." % FLAGS.task_index)
mnist = input_data.read_data_sets(FLAGS.data_dir, one_hot=True)
sess_config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=True,
device_filters=["/job:ps", "/job:worker/task:%d" % FLAGS.task_index])

# The supervisor takes care of session initialization, restoring from
# a checkpoint, and closing when done or an error occurs.
with sv.prepare_or_wait_for_session(server.target, config=sess_config) as sess:
print("Worker %d: Session initialization complete." % FLAGS.task_index)
# Loop until the supervisor shuts down or 1000000 steps have completed.
step = 0
while not sv.should_stop() and step < 1000000:
# Run a training step asynchronously.
# See `tf.train.SyncReplicasOptimizer` for additional details on how to
# perform *synchronous* training.
batch_xs, batch_ys = mnist.train.next_batch(FLAGS.batch_size)
print("FETCHING NEXT BATCH %d" % FLAGS.batch_size)
train_feed = {x: batch_xs, y_: batch_ys}
_, step = sess.run([train_op, global_step], feed_dict=train_feed)
if step % 100 == 0:
print("Done step %d" % step)
# Ask for all the services to stop.
sv.stop()
if __name__ == "__main__":
tf.app.run()

以下是任务=0的工作人员的日志：

2017-06-20 04：50：58.405431：I 张量流/核心/common_runtime/simple_placer.cc：841]Adagrad/value： (Const)/job：ps/replica：0/task：0/cpu：0 truncated_normal/stddev： (Const)：/job：worker/replica：0/task：0/gpu：0
2017-06-20 04：50：58.405456： I tensorflow/core/common_runtime/simple_placer.cc：841] truncated_normal/stddev： (Const)/job：worker/replica：0/task：0/gpu：0 truncated_normal/mean： (Const)：/job：worker/replica：0/task：0/gpu：0
2017-06-20 04：50：58.405481： I 张量流/核心/common_runtime/simple_placer.cc：841] truncated_normal/平均值： (常量)/作业：worker/副本：0/任务：0/gpu：0 truncated_normal/形状： (常量)：/作业：worker/replica：0/task：0/gpu：0
2017-06-20 04：50：58.405506： I 张量流/核心/common_runtime/simple_placer.cc：841] truncated_normal/形状： (常量)/作业：工作线程/副本：0/任务：0/gpu：0 工作线程 0： 会话初始化完成。
正在获取下一批 500
正在获取下一批 500 获取下一批 500 获取下一批 500 获取下一批 500 完成步骤 408800 ... ...

但是从工作人员 2 (任务 = 1) 开始，日志如下所示：

2017-06-20 04：51：07.288600：I tensorflow/core/common_runtime/simple_placer.cc：841] 零：(常量)/作业：worker/replica：0/任务：1/gpu：0 Adagrad/value：(常量)：/作业：ps/replica：0/任务：0/cpu：0
2017-06-20 04：51：07.288614：I tensorflow/core/common_runtime/simple_placer.cc：841] Adagrad/value： (Const)/job：ps/replica：0/task：0/cpu：0 truncated_normal/stddev： (Const)：/job：worker/replica：0/task：1/gpu：0
2017-06-20 04：51：07.288639： I tensorflow/core/common_runtime/simple_placer.cc：841] truncated_normal/stddev： (Const)/job：worker/replica：0/task：1/gpu：0 truncated_normal/mean： (常量)：/job：worker/replica：0/task：1/gpu：0
2017-06-20 04：51：07.288664： I tensorflow/core/common_runtime/simple_placer.cc：841] truncated_normal/mean： (常量)/作业：worker/replica：0/任务：1/gpu：0 truncated_normal/形状： (常量)：/作业：worker/replica：0/任务：1/gpu：0 2017-06-20 04：51：07.288689： I 张量流/核心/common_runtime/simple_placer.cc：841] truncated_normal/形状： (常量)/作业：worker/副本：0/任务：1/gpu：0

我期待两个工人都有类似的日志。请帮助我理解这一点。期待您的帮助。

Answer 1

您需要手动分离出每个工作人员的数据。

# Get the subset of data for this worker 
mnist = input_data.read_data_sets('/tmp/mnist_temp', one_hot=True)
num_old = mnist.train._num_examples
ids = list(range(task_index, mnist.train._num_examples, num_workers))
mnist.train._images = mnist.train._images[ids,]
mnist.train._labels = mnist.train._labels[ids,]
mnist.train._num_examples = mnist.train._images.shape[0]
print("subset of training examples ", mnist.train._num_examples,"/",num_old)