pkill which waits

Question

Is there a pkill-like tool which does this:

1. send signal to all matching processes
2. wait N seconds until the processes have terminated
3. if all processes have terminated, nice: exit
4. if some processes have not terminated send SIGKILL

For me it is important that the tool waits until the processes have really terminated.

I know that it is quite easy to do this in my favourite scripting language, but in this case it would be very nice if I use a tool which already exists.

This needs to run on SuSE-Linux and Ubuntu.

Answer 1

There is no “standard” command which provides the behaviour you’re after. However, on Debian and derivatives, you can use start-stop-daemon’s --stop action with the --retry option:

start-stop-daemon --stop --oknodo --retry 15 -n daemontokill

will send SIGTERM to all processes named daemontokill, wait up to 15s for them to stop, then send SIGKILL to all remaining processes (from the initial selection), and wait another 15s for them to die. It will exit with status 0 if there was nothing to kill or all the processes stopped, 2 if some processes were still around after the second timeout.

There are a number of options to match processes in various ways, see the documentation (linked above) for details. You can also provide a more detailed schedule with varying timeouts.

start-stop-daemon is part of the dpkg package so it’s always available on Debian systems (and derivatives). Some non-.deb distributions make the package available too; for example, openSUSE Leap 42 has it. It’s quite straightforward to build on other platforms:

git clone https://salsa.debian.org/dpkg-team/dpkg.git 
cd dpkg
autoreconf -fi && ./configure && make

You’ll need autoconf, automake, libtool, gettext. Once the build is finished you’ll find start-stop-daemon in the utils directory.

Answer 2

You can use the timeout command with killall -w

e.g:

# timeout 3 killall -vws2 $PROCESS_NAME || killall -vws9 $PROCESS_NAME

timeout will "timeout" after 3 secends and will exit with error code, which then you catch with || and then run killall -s9 to force the kill.

Answer 3

You can create your own script that will fulfil your case. Following this post useless use of kill -9 and source reallykill

#!/bin/sh
#
# $Id: reallykill,v 1.2 2004/10/11 09:36:48 jmates Exp $
#
# The author disclaims all copyrights and releases this script into the
# public domain.
#
# Kills the specified processes by starting with the TERM signal and
# only if necessary working up to the violent KILL signal. For more
# information, see:
#
# http://sial.org/howto/shell/kill-9/

if [ -z "$1" ]; then
  echo "usage: `basename $0` pid [pid ..]" >&2
  exit 100
fi

cycle_kill () {
  PID=$1
  RETVAL=0

  for signal in "TERM" "INT" "HUP" "KILL"; do
    kill -$signal $PID
    RETVAL=$?
    [ $RETVAL -eq 0 ] && break
    echo "warning: kill failed: pid=$PID, signal=$signal" >&2
    sleep 1
  done

  return $RETVAL
}

for pid in "$@"; do
  cycle_kill $pid
done

So you can just download script to your bin directory:

 curl http://web.archive.org/web/20080213192428if_/http://sial.org:80/howto/shell/kill-9/reallykill > $HOME/bin/killwait && chmod +x $HOME/bin/killwait

I would use this that way:

ping -q localhost &
killwait `pidof ping`

Maybe you would like to change order of signals with INT as first, so you would have output with summary of ping command. Provided command does not wait if KILL signal succeed but it should work as process will be killed by system.