diff --git a/README.md b/README.md
index 8ab6dd3b..9ec2aab8 100644
--- a/README.md
+++ b/README.md
@@ -163,6 +163,7 @@ The command line arguments are as follows:
 | shell_env          | no        | --shell_env LD_LIBRARY_PATH=/usr/local/lib64/     | Specifies key-value pairs for environment variables which will be set in your python worker/ps processes.                                                                                                         |
 | conf_file          | no        | --conf_file tony-local.xml                        | Location of a TonY configuration file, also support remote path, like `--conf_file hdfs://nameservice01/user/tony/tony-remote.xml`                                                                                |
 | conf               | no        | --conf tony.application.security.enabled=false    | Override configurations from your configuration file via command line
+| sidecar_tensorboard_log_dir | no | --sidecar_tensorboard_log_dir /hdfs/path/tensorboard_log_dir | HDFS path to tensorboard log dir, it will enable sidecar tensorboard managed by TonY. More detailed example refers to tony-examples/mnist_tensorflow module                                          |
 
 ## TonY configurations
 
@@ -211,3 +212,7 @@ For more information about TonY, check out the following:
 2. How do I configure arbitrary TensorFlow job types?
 
     Please see the [wiki](https://github.com/linkedin/TonY/wiki/TonY-Configurations#task-configuration) on TensorFlow task configuration for details.
+
+3. My tensorflow's partial workers hang when chief finished. Or evaluator hang when chief and workers finished.
+   
+   Please see the [PR#521](https://github.com/tony-framework/TonY/pull/621) on Tensorflow configuration to solve it.
\ No newline at end of file
diff --git a/tony-examples/mnist-tensorflow/README.md b/tony-examples/mnist-tensorflow/README.md
index c86ea2c5..bb22f71b 100644
--- a/tony-examples/mnist-tensorflow/README.md
+++ b/tony-examples/mnist-tensorflow/README.md
@@ -113,8 +113,12 @@ We have tested this example with 3 Workers (4 GB RAM + 1 vCPU) using MultiWorke
 ### Tensorboard Usage
 TonY supports two modes(custom and sidecar) to start tensorboard.
 1. [Custom] Allow users to start tensorboard in code, more details can be found in mnist_distributed.py example.
-2. [Sidecar] Using the built-in tensorboard, it will start extra executor to running tensorboard by TonY. Only one thing to do is specify the log dir in tony xml, like as follows
+2. [Sidecar] Using the built-in sidecar tensorboard, the extra tensorboard task executor will be managed by TonY. 
+   The failure of sidecar tensorboard will not affect the entire training job. 
+   Only one thing for user to do is to specify the log dir in tony xml or in tony cli, like as follows. 
+   Tips: the conf priority in tony cli is prior to in tony xml. 
 
+tony.xml
 ```
 <configuration>
   ....
@@ -123,4 +127,13 @@ TonY supports two modes(custom and sidecar) to start tensorboard.
     <value>/tmp/xxxxxxx</value>
   </property>
 </configuration>
-```
\ No newline at end of file
+```
+tony cli
+
+    $ java -cp "`hadoop classpath --glob`:MyJob/*:MyJob/" \
+            com.linkedin.tony.cli.ClusterSubmitter \
+            -executes models/mnist_distributed.py \
+            -task_params '--input_dir /path/to/hdfs/input --output_dir /path/to/hdfs/output' \
+            -src_dir src \
+            -python_binary_path /home/user_name/python_virtual_env/bin/python
+            -sidecar_tensorboard_log_dir /path/to/hdfs/tensorboard_log_dir