bash脚本杀死种族条件

Question

I have a bash script which calls kill $PID , then waits for 2 seconds and if $PID still exists (by checking ps ), it calls kill -9 $PID as a fallback. AFAIK, this is a very standard script to kill a process, found in many sample scripts on the net.

The script works fine (I am not worried about PID reuse here), but today I managed to catch a rare race condition where the process is still being killed in the 2 seconds, and the kill -9 hangs . Here is the ps output:

root     17172  0.0  0.0   7920  1668 ?        S    Jul16   0:00 /bin/kill -9 16635
root     17173  0.0  0.0      0     0 ?        Z    Jul16   0:00 [kill] <defunct>

The kill -9 is hanging because it has hit a defunct process (ie from the previous kill), ie PID 16635 is the process already killed by the first kill.

If I do a kill -9 on the kill -9 process (PID 17172) , then it is all good.

Is there anyway I can prevent the "kill -9" from hanging in the first place? It is a fallback to ensure that the process is killed within 2 seconds. I do not want to add another background check to check the fallback kill -9 .

Update :

The best answer I could find is to use the timeout command on the kill -9 :

timeout -s 9 2 kill -9 $PID

This will ensure that after 2 seconds, if the kill -9 hangs, the timeout will issue a kill -9 on it.

Answer 1

This may be a better way to check if the process is still alive, and if you should send the kill -9

kill [pid]
sleep 5
kill -0 [pid]
if [ $? -eq  0 ] ; then
  echo 'process not terminated'
  kill -9 [pid]
else
  echo 'process terminated'
fi

This uses the sending of a signal through kill to make sure you have permission to send a kill signal to the process kill -0 . Which may be better then checking just if the PID is still alive. Then if if you do have access to the program terminate the program. The only other thing you can possible do is put in a longer sleep or something non trival like writing your own kill or checking the process state,etc.

Also you can do another sleep and then check for the race condition by seeing after like 20 seconds if the kill -9 is still alive and then killing that process. Which would at least alleviate the stalling of the script but not solve the race condition.