简介

Tensorflow API提供了Cluster、Server以及Supervisor来支持模型的分布式训练。

关于Tensorflow的分布式训练介绍可以参考Distributed Tensorflow。简单的概括说明如下：

• Tensorflow分布式Cluster由多个Task组成，每个Task对应一个tf.train.Server实例，作为Cluster的一个单独节点；
• 多个相同作用的Task可以被划分为一个job，例如ps job作为参数服务器只保存Tensorflow model的参数，而worker job则作为计算节点只执行计算密集型的Graph计算。
• Cluster中的Task会相对进行通信，以便进行状态同步、参数更新等操作。

Tensorflow分布式集群的所有节点执行的代码是相同的。分布式任务代码具有固定的模式：

# 第1步：命令行参数解析，获取集群的信息ps_hosts和worker_hosts，以及当前节点的角色信息job_name和task_index

# 第2步：创建当前task结点的Server
cluster = tf.train.ClusterSpec({"ps": ps_hosts, "worker": worker_hosts})
server = tf.train.Server(cluster, job_name=FLAGS.job_name, task_index=FLAGS.task_index)

# 第3步：如果当前节点是ps，则调用server.join()无休止等待；如果是worker，则执行第4步。
if FLAGS.job_name == "ps":
    server.join()

# 第4步：则构建要训练的模型
# build tensorflow graph model

# 第5步：创建tf.train.Supervisor来管理模型的训练过程
# Create a "supervisor", which oversees the training process.
sv = tf.train.Supervisor(is_chief=(FLAGS.task_index == 0), logdir="/tmp/train_logs")
# The supervisor takes care of session initialization and restoring from a checkpoint.
sess = sv.prepare_or_wait_for_session(server.target)
# Loop until the supervisor shuts down
while not sv.should_stop()
     # train model

Tensorflow分布式训练代码框架

根据上面说到的Tensorflow分布式训练代码固定模式，如果要编写一个分布式的Tensorlfow代码，其框架如下所示。

import tensorflow as tf

# 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")

# 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")

FLAGS = tf.app.flags.FLAGS


def main(_):
  ps_hosts = FLAGS.ps_hosts.split(",")
  worker_hosts = FLAGS.worker_hosts(",")

  # Create a cluster from the parameter server and worker hosts.
  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":
    # 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)):

      # Build model...
      loss = ...
      global_step = tf.Variable(0)

      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=(FLAGS.task_index == 0),
                             logdir="/tmp/train_logs",
                             init_op=init_op,
                             summary_op=summary_op,
                             saver=saver,
                             global_step=global_step,
                             save_model_secs=600)

    # The supervisor takes care of session initialization and restoring from
    # a checkpoint.
    sess = sv.prepare_or_wait_for_session(server.target)

    # Start queue runners for the input pipelines (if any).
    sv.start_queue_runners(sess)

    # 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.
      _, step = sess.run([train_op, global_step])


if __name__ == "__main__":
  tf.app.run()

对于所有Tensorflow分布式代码，可变的只有两点：

1. 构建tensorflow graph模型代码；
2. 每一步执行训练的代码

分布式MNIST任务

我们通过修改tensorflow/tensorflow提供的mnist_softmax.py来构造分布式的MNIST样例来进行验证。修改后的代码请参考mnist_dist.py。

我们同样通过tensorlfow的Docker image来启动一个容器来进行验证。

$ docker run -d -v /path/to/your/code:/tensorflow/mnist --name tensorflow tensorflow/tensorflow

启动tensorflow之后，启动4个Terminal，然后通过下面命令进入tensorflow容器，切换到/tensorflow/mnist目录下

$ docker exec -ti tensorflow /bin/bash
$ cd /tensorflow/mnist

然后在四个Terminal中分别执行下面一个命令来启动Tensorflow cluster的一个task节点，

# Start ps 0
python mnist_dist.py --ps_hosts=localhost:2221,localhost:2222 --worker_hosts=localhost:2223,localhost:2224 --job_name=ps --task_index=0

# Start ps 1
python mnist_dist.py --ps_hosts=localhost:2221,localhost:2222 --worker_hosts=localhost:2223,localhost:2224 --job_name=ps --task_index=1

# Start worker 0
python mnist_dist.py --ps_hosts=localhost:2221,localhost:2222 --worker_hosts=localhost:2223,localhost:2224 --job_name=worker --task_index=0

# Start worker 1
python mnist_dist.py --ps_hosts=localhost:2221,localhost:2222 --worker_hosts=localhost:2223,localhost:2224 --job_name=worker --task_index=1

具体效果自己验证哈。