I have a Bash script "script" which roughly looks like:
#!/bin/bash
cmd1 | cmd2 | cmd3
When I do a kill script (or more precisely when I do a 'stop script' in supervisord), not all cmd* are killed. How can I make sure they are terminated along with the script that spawned them?
Supervisord has stopasgroup and killasgroup options (false by default) which determine whether to propagate SIGTERM/SIGKILL signals to child processes.
[program:script]
command=script
stopasgroup=true
killasgroup=true
(These config variables are both documented at http://supervisord.org/configuration.html.)
Not sure about how using supervisord, but with pkill you can use the -P option to kill from a parent process down to all the children. Here's the process trees (starting from my running ssh daemon).
$ pstree -a -p 1792
sshd,1792
├─sshd,27150
│ └─sshd,27153
│ └─zsh,27154
│ └─test.sh,27325 ./test.sh
│ └─cat,27326
└─sshd,27182
└─sshd,27184
└─zsh,27185
└─pstree,27357 -a -p 1792
In one session I have a script test.sh with pid 27325, and in the other I'm running the command pstree -a -p 1792 (because sshd had pid 1792)
And after I run pkill -TERM -P 27325:
$ pstree -a -p 1792
sshd,1792
├─sshd,27150
│ └─sshd,27153
│ └─zsh,27154
└─sshd,27182
└─sshd,27184
└─zsh,27185
└─pstree,27387 -a -p 1792
This answer was essentially rephrased from the this other answer on stackoverflow: https://stackoverflow.com/a/392155/263969
Another solution is to trap SIGTERM and kill all children in the trap-code.
The only thing is that the shell runs trap-code for a received signal only after the currently running command has completed -- so in your case, this alone wouldn't help.
However, traps will be run in an "async" kind of way if the shell is in "wait".
#!/usr/bin/env bash
trap 'kill 0' TERM
( cmd1 | cmd2 | cmd3 ) & wait
The kill 0 sends SIGTERM to all processes in the current process group.
NOTE: I'm talking about Bash here.
