TensorBoardDebugWrapperSession遇RPC名称解析失败，是代码问题还是Bug？

先看一下你的环境版本：

• Python 2.7.16
• TensorBoard: 1.13.1
• TensorFlow: 1.13.1

你用来复现问题的代码如下：

import tensorflow as tf
from tensorflow.python import debug as tf_debug
graph = tf.Graph()
with graph.as_default():
    inp = tf.placeholder(tf.float32,shape=(1))
    inp2 = tf.placeholder(tf.float32,shape=(1))
    var = tf.math.multiply(inp,inp2)
sess = tf.Session(graph=graph)
with sess.as_default():
    sess = tf_debug.TensorBoardDebugWrapperSession(sess,"grpc://localhost:6064")
    with tf.summary.FileWriter('/home/penalvad/stattus4/stattus4-audio-models/notebooks/',graph=graph,session=sess) as writer:
        tf.summary.initialize(graph=self.graph,session=sess)
        sess.run(var,feed_dict={inp:10.0,inp2:2.0})

遇到的报错信息：

_Rendezvous: <_Rendezvous of RPC that terminated with: status = StatusCode.UNAVAILABLE details = "Name resolution failure" debug_error_string = "{"created":"@1564495830.026710193","description":"Failed to create subchannel","file":"src/core/ext/filters/client_channel/client_channel.cc","file_line":2721,"referenced_errors":[{"created":"@1564495830.026708437","description":"Name resolution failure","file":"src/core/ext/filters/client_channel/client_channel.cc","file_line":3026,"grpc_status":14}]}"

结论 #

这不是TensorFlow的Bug，而是操作配置上的问题。报错里的"Name resolution failure"说明gRPC无法正确解析你指定的连接地址，结合你后续的更新说明来看：

• 之前大概率存在gRPC协议前缀错误（比如漏写grpc://）或者端口不匹配的情况
• 同时需要确保TensorBoard Debug Server确实在localhost:6064这个端口正常运行，端口未启动或不匹配也会触发连接失败

修正地址格式与端口后问题解决，也验证了这一点——只要调整好连接参数，确保Debug Server在对应端口运行，就能正常工作。

内容的提问来源于stack exchange，提问作者Syncaa