With screen and nohup running, there is no need to worry about deep learning code anymore | exiting the terminal will not affect the operation of server program code

Run deep learning code on the server ,xshell Disconnect the code and hang up . How to run programs on the server   Even closing the terminal will not affect the code ？ 

screen 

I use it screen. Recommend this method , I feel better than nohup Easy to use 10000 times . 

1.  Connect to server

In the figure below, I connected 88 99  This server .

2.  newly build session

Input... At the terminal

screen -S  name

Then it will close your current window , Pop up a new window , Maybe the tab name is screen（ I am using xshell, Different software may differ ）.  You are in the new pop-up window

bash  Code .sh

Just run your program .

You can close it directly xshell Wait for the terminal to disconnect , To do something else .

The following figure shows that although the tab name has changed , But it does 88 99  The server . 

3.  Hang up  

If you don't need to disconnect , I also want to do other things with the server

, There are two ways ： 

• Use
  Ctrl+ A + D
  , It is to throw the currently running content into the background , You can continue to use the server in the current window . It will show as follows , notice detached Then it will return to your normal terminal . 

• More simply, you can open another server window .

Double click directly on the left session manager . Now? 2 and 3 Both are my current servers .

4.  Back to running code session

How to find the code I run after reconnecting to the server ？

 

• If you don't remember before screen establish session What's your name , Use
  screen -ls
  view list .

For example, the following figure shows that I have a mydemo By me detach 了 . And then look at the next step .

• If you remember what you created before session  What's its name , Your direct
  screen -r  name
  , You can return to the operation interface . 

The following figure shows that my program is still running normally .

5  eliminate session 

• If you run the code , This session useless . Quit current session, then kill That's all right. .  Look at the picture below , At this time, the reply list shows that I have mydemo This still exists . Kill this process directly in the yellow square OK 了 . If the program is killed, it will die  ╭(●｀∀´●)╯. 

• If it is an abnormal death, it will show dead, Use
  screen -wipe
  Clear it OK 了 . 

As you can see below , The reply list shows mydemo It's dead , Clear it from the yellow box , Look again, the list is empty .

nohup 

Use nohup It is to directly let the code run in the background , And output the operation log . 

1.  Run code  

Enter the corresponding file directory , Input

nohup  Files to run  >  Log files  2>&1 & 

  For example, I want to run  /home/sian/test  Under the  demo.sh file ：

（ If you encounter errors, please pull to the end of this article ）

 

Explain the function of the above sentence ： 

• nohup demo.sh
   

function demo.sh file

If you only use this statement, you will be prompted ： 

nohup: ignoring input and appending output to ‘nohup.out’

It tells you to put the output in nohup.out In the text . you

ls

to glance at , A nohup.out, The running results are stored in this file .

• nohup demo.sh > train.log
   

function demo.sh, And store the operation log in train.log in .

You will be prompted after running ： 

nohup: ignoring input and redirecting stderr to stdout

Is to change the original default file stderr and stdout Relocate to the new file you specified .

Here's the picture , The above nohup.out It will become our designated train.log. 

•  nohup demo.sh > train.log 2>&1 
   

among  

2>&1

The role of is “redirecting stderr to stdout ”, That is, redirect standard errors to standard output , Then redirect the standard output to  train.log  In file .

added

2>&1

Then there won't be “redirecting stderr to stdout” This prompt .

If you don't add that

&

, Just write

2 > 1

Two files will be generated , One is 1, One is train.log,1 There will be warnings and errors in the output .

• 2 representative stderr(standard error, Standard error output )

• 1 representative stdout (standard output, standard output ) 

• nohup demo.sh > train.log 2>&1 &

This last

&

It means to execute in the background .

for instance ：

• If you don't add this
  &
  , The implementation is as follows , The code will stay in this interface until it runs out , Your terminal window cannot be used anymore .

• Add `&` after , Direct execution of the program ,** And throw the program in the background **, It does not affect the use of the current terminal window , You can also exit the server connection directly .

2.  Check the log  

Generally, the running log will be generated under your running directory , Find the file and use it directly `tail -f train.log` Just open the file . 

3.  Stop running  

If you want to stop running , You need to use

ps -aux | grep " Run file name " 

（ perhaps

ps -def | grep " Run file name " 

） find  nohup  To run a script  PID, Then use

kill -9 PID

To delete . 

4.  What to do if you report an error  

Show that the executive file cannot be found  

nohup: failed to run command ‘demo.py’: No such file or directory 

First step , Try to modify the instruction

 

Directly on my computer

nohup  file name  ...

Will always report errors that the file cannot be found .  I need it

nohup bash xxx.sh

perhaps

nohup python xxx.py 

, You need to add the prefix of normal execution to avoid error , You can have a try . 

The second step is to check whether you have entered the wrong directory .

 

for instance ：

You want to run  /home/sian/test  Under the  demo.sh, You must enter test Below the file , Or specify the path when executing .

The figure below 1, I went into test Folder execution , Write the file name directly during execution :ok:, The generated execution log file is also test Under the table of contents .

The figure below 2, I am here sian Execute... In folder , After execution, you need to fill in the address of the document , The generated log will be in sian Under the document . 

If you can confirm the address ,

The third step is to check the authority of the execution file , If the file does not have execution permission , Just open the execution permission . 

Take a chestnut ：

• use
  ll
  View permissions , See that the file permissions shown in the figure below do not
  -x
  This one , It proved to be caused by lack of authority .

• use
  chmod a+x 
  Add execution permission to the file .

You can see that the following figure has been displayed after adding the execution permission

-x

了 . Go back and execute the document, and the problem will be solved .

If the above two steps are ok , It's still hard to make , Then Baidu ！