Kill program if no output to STDOUT for x seconds?

A Bash script runs a program in the background and redirects stdout of that program to a file. How can I kill the program if it hasn’t written to stdout for the last x seconds?

prog > out.txt &
wait $PID

I imagine to pipe the output through some other program that checks for continuing output. unbuffer could be necessary to disables output buffering.

unbuffer prog | ... > out.txt &
Asked By: bivox


It could be something like:


while read line -t 10
   echo $line

if [ $? -gt 128 ]
  echo "Timeout occurred"
  killall -SIGKILL procces
Answered By: White Owl

You can use /usr/bin/inotifywait inside your script, without the -m option, so monitoring stops at the first write event, and with the -t 10 option so the process exits with code 2 if no write-modification event was registered within the last 10s:

#!/usr/bin/env bash

my_prog > out.txt &

while [ "$my_watch_status" != "2" ]; do
    /usr/bin/inotifywait -qq -t 10 -e modify out.txt 2>/dev/null & 
    wait $!

kill -9 "$my_prog_pid"
exit <my exit_code>
  • Option -qq ("very quiet") and redirection of standard error to /dev/null are meant to suppress any inotifywait diagnostic information on standard error and event information on standard output.
  • Every time out.txt is written to (within a 10s time window), a new 10s monitoring time window is started.
  • kill -9, aka SIGKILL, is used for not knowing better about possible requirements for you to exit my_prog in an orderly fashion. If a graceful exit is preferable, a simple kill "$my_prog_pid" can be used.
  • The script kills my_prog and exits, optionally with your choice of exit code, only when no write-modification of out.txt has occurred within the prescribed time window.
  • Note that, during execution, the time window is not a perfect 10s. The actual time-window will be a little more than 10s (think "ms" for one file) because it takes a little time to establish monitoring. This is unsubstantial in your case, because you monitor only one file, but it could become quite substantial if you were to monitor a large tree (think of a directory structure with thousands of files) in a recursive manner. If the 10s time window is application-critical, you might want to explore other methods than the one proposed here, to not have to reestablish monitoring every time-window span or, as a worst-case scenario, everytime a write event occurs within that time span.
Answered By: Cbhihe
Categories: Answers Tags: ,
Answers are sorted by their score. The answer accepted by the question owner as the best is marked with
at the top-right corner.