linux only - What's the best way to send a signal to all members of a process group?




descendants and (25)

Inspired by ysth’s comment

kill -- -PGID

instead of giving it a process number, give it the negation of the group number. As usual with almost any command, if you want a normal argument that starts with a - to not be interpreted as a switch, precede it with --

I want to kill a whole process tree. What is the best way to do this using any common scripting languages? I am looking for a simple solution.


The following has been tested on FreeBSD, Linux and MacOS X and only depends on pgrep and kill (the ps -o versions don't work under BSD). First argument is parent pid of which children have to be terminated. second argument is a boolean to determine whether the parent pid has to be terminated too.

KillChilds() {
        local pid="${1}"
        local self="${2:-false}"

        if children="$(pgrep -P "$pid")"; then
                for child in $children; do
                        KillChilds "$child" true
                done
        fi

        if [ "$self" == true ]; then
                kill -s SIGTERM "$pid" || (sleep 10 && kill -9 "$pid" &)
        fi
}

KillChilds $$ > /dev/null 2>&1

This will send SIGTERM to any child / grandchild process within a shell script and if SIGTERM doesn't succeed, it will wait 10 seconds and then send kill.


Earlier answer:

The following also works but will kill the shell itself on BSD.

KillSubTree() {
    local parent="${1}"
    for child in $(ps -o pid=$parent); do
            if [ $$ -ne $child ]; then (kill -s SIGTERM $child || (sleep 10 && kill -9 $child & )) > /dev/null 2>&1 ; fi
    done
}
# Example lanch from within script
KillSubTree $$ > /dev/null 2>&1

pkill -TERM -P 27888

This will kill all processes that have the parent process ID 27888.

Or more robust:

CPIDS=$(pgrep -P 27888); (sleep 33 && kill -KILL $CPIDS &); kill -TERM $CPIDS

which schedule killing 33 second later and politely ask processes to terminate.

See this answer for terminating all descendants.


It is probably better to kill the parent before the children; otherwise the parent may likely spawn new children again before he is killed himself. These will survive the killing.

My version of ps is different from that above; maybe too old, therefore the strange grepping...

To use a shell script instead of a shell function has many advantages...

However, it is basically zhigangs idea


#!/bin/bash
if test $# -lt 1 ; then
    echo >&2 "usage: kiltree pid (sig)"
fi ;

_pid=$1
_sig=${2:-TERM}
_children=$(ps j | grep "^[ ]*${_pid} " | cut -c 7-11) ;
echo >&2 kill -${_sig} ${_pid}
kill -${_sig} ${_pid}
for _child in ${_children}; do
    killtree ${_child} ${_sig}
done

This is my version of killing all the child processes using bash script. It does not use recursion and depends on pgrep command.

Use

killtree.sh PID SIGNAL

Contents of killtrees.sh

#!/bin/bash
PID=$1
if [ -z $PID ];
then
    echo "No pid specified"
fi

PPLIST=$PID
CHILD_LIST=`pgrep -P $PPLIST -d,`

while [ ! -z "$CHILD_LIST" ]
do
    PPLIST="$PPLIST,$CHILD_LIST"
    CHILD_LIST=`pgrep -P $CHILD_LIST -d,`
done

SIGNAL=$2

if [ -z $SIGNAL ]
then
    SIGNAL="TERM"
fi
#do substring from comma to space
kill -$SIGNAL ${PPLIST//,/ }

In sh the jobs command will list the background processes. In some cases it might be better to kill the newest process first, e.g. the older one created a shared socket. In those cases sort the PIDs in reverse order. Sometimes you want to wait moment for the jobs to write something on disk or stuff like that before they stop.

And don't kill if you don't have to!

for SIGNAL in TERM KILL; do
  for CHILD in $(jobs -s|sort -r); do
    kill -s $SIGNAL $CHILD
    sleep $MOMENT
  done
done

Killing child process in shell script:

Many time we need to kill child process which are hanged or block for some reason. eg. FTP connection issue.

There are two approaches,

1) To create separate new parent for each child which will monitor and kill child process once timeout reached.

Create test.sh as follows,

#!/bin/bash

declare -a CMDs=("AAA" "BBB" "CCC" "DDD")
for CMD in ${CMDs[*]}; do
    (sleep 10 & PID=$!; echo "Started $CMD => $PID"; sleep 5; echo "Killing $CMD => $PID"; kill $PID; echo "$CMD Completed.") &
done
exit;

and watch processes which are having name as 'test' in other terminal using following command.

watch -n1 'ps x -o "%p %r %c" | grep "test" '

Above script will create 4 new child processes and their parents. Each child process will run for 10sec. But once timeout of 5sec reach, thier respective parent processes will kill those childs. So child won't be able to complete execution(10sec). Play around those timings(switch 10 and 5) to see another behaviour. In that case child will finish execution in 5sec before it reaches timeout of 10sec.

2) Let the current parent monitor and kill child process once timeout reached. This won't create separate parent to monitor each child. Also you can manage all child processes properly within same parent.

Create test.sh as follows,

#!/bin/bash

declare -A CPIDs;
declare -a CMDs=("AAA" "BBB" "CCC" "DDD")

CMD_TIME=15;
for CMD in ${CMDs[*]}; do
    (echo "Started..$CMD"; sleep $CMD_TIME; echo "$CMD Done";) &
    CPIDs[$!]="$RN";
    sleep 1;
done

GPID=$(ps -o pgid= $$);
CNT_TIME_OUT=10;
CNT=0;
while (true); do
    declare -A TMP_CPIDs;

    for PID in "${!CPIDs[@]}"; do
        echo "Checking "${CPIDs[$PID]}"=>"$PID;

        if ps -p $PID > /dev/null ; then
          echo "-->"${CPIDs[$PID]}"=>"$PID" is running..";
          TMP_CPIDs[$PID]=${CPIDs[$PID]};
        else
          echo "-->"${CPIDs[$PID]}"=>"$PID" is completed.";
        fi
    done

    if [ ${#TMP_CPIDs[@]} == 0 ]; then
        echo "All commands completed.";
        break;
    else
        unset CPIDs;
        declare -A CPIDs;
        for PID in "${!TMP_CPIDs[@]}"; do
            CPIDs[$PID]=${TMP_CPIDs[$PID]};
        done
        unset TMP_CPIDs;

        if [ $CNT -gt $CNT_TIME_OUT ]; then
            echo ${CPIDs[@]}"PIDs not reponding. Timeout reached $CNT sec. killing all childern with GPID $GPID..";
            kill -- -$GPID;
        fi
    fi

    CNT=$((CNT+1));
    echo "waiting since $b secs..";
    sleep 1;
done

exit;

and watch processes which are having name as 'test' in other terminal using following command.

watch -n1 'ps x -o "%p %r %c" | grep "test" '

Above script will create 4 new child processes. We are storing pids of all child process and looping over them to check if they are finished their execution or still running. Child process will execution till CMD_TIME time. But if CNT_TIME_OUT timeout reach , All children will get killed by parent process. You can switch timing and play around with script to see behavior. One drawback of this approach is , it is using group id for killing all child tree. But parent process itself belong to same group so it will also get killed.

You may need to assign other group id to parent process if you don’t want parent to be killed.

More details can be found here,

Killing child process in shell script


rkill command from pslist package sends given signal (or SIGTERM by default) to specified process and all its descendants:

rkill [-SIG] pid/name...

Kill all the processes belonging to the same process tree using the Process Group ID (PGID)

  • kill -- -$PGID     Use default signal (TERM = 15)
  • kill -9 -$PGID     Use the signal KILL (9)

You can retrieve the PGID from any Process-ID (PID) of the same process tree

  • kill -- -$(ps -o pgid= $PID | grep -o '[0-9]*')   (signal TERM)
  • kill -9 -$(ps -o pgid= $PID | grep -o '[0-9]*')   (signal KILL)

Special thanks to tanager and Speakus for contributions on $PID remaining spaces and OSX compatibility.

Explanation

  • kill -9 -"$PGID" => Send signal 9 (KILL) to all child and grandchild...
  • PGID=$(ps opgid= "$PID") => Retrieve the Process-Group-ID from any Process-ID of the tree, not only the Process-Parent-ID. A variation of ps opgid= $PID is ps -o pgid --no-headers $PID where pgid can be replaced by pgrp.
    But:
    • ps inserts leading spaces when PID is less than five digits and right aligned as noticed by tanager. You can use:
      PGID=$(ps opgid= "$PID" | tr -d ' ')
    • ps from OSX always print the header, therefore Speakus proposes:
      PGID="$( ps -o pgid "$PID" | grep [0-9] | tr -d ' ' )"
  • grep -o [0-9]* prints successive digits only (does not print spaces or alphabetical headers).

Further command lines

PGID=$(ps -o pgid= $PID | grep -o [0-9]*)
kill -TERM -"$PGID"  # kill -15
kill -INT  -"$PGID"  # correspond to [CRTL+C] from keyboard
kill -QUIT -"$PGID"  # correspond to [CRTL+\] from keyboard
kill -CONT -"$PGID"  # restart a stopped process (above signals do not kill it)
sleep 2              # wait terminate process (more time if required)
kill -KILL -"$PGID"  # kill -9 if it does not intercept signals (or buggy)

Limitation

  • As noticed by davide and Hubert Kario, when kill is invoked by a process belonging to the same tree, kill risks to kill itself before terminating the whole tree killing.
  • Therefore, be sure to run the command using a process having a different Process-Group-ID.

Long story

> cat run-many-processes.sh
#!/bin/sh
echo "ProcessID=$$ begins ($0)"
./child.sh background &
./child.sh foreground
echo "ProcessID=$$ ends ($0)"

> cat child.sh
#!/bin/sh
echo "ProcessID=$$ begins ($0)"
./grandchild.sh background &
./grandchild.sh foreground
echo "ProcessID=$$ ends ($0)"

> cat grandchild.sh
#!/bin/sh
echo "ProcessID=$$ begins ($0)"
sleep 9999
echo "ProcessID=$$ ends ($0)"

Run the process tree in background using '&'

> ./run-many-processes.sh &    
ProcessID=28957 begins (./run-many-processes.sh)
ProcessID=28959 begins (./child.sh)
ProcessID=28958 begins (./child.sh)
ProcessID=28960 begins (./grandchild.sh)
ProcessID=28961 begins (./grandchild.sh)
ProcessID=28962 begins (./grandchild.sh)
ProcessID=28963 begins (./grandchild.sh)

> PID=$!                    # get the Parent Process ID
> PGID=$(ps opgid= "$PID")  # get the Process Group ID

> ps fj
 PPID   PID  PGID   SID TTY      TPGID STAT   UID   TIME COMMAND
28348 28349 28349 28349 pts/3    28969 Ss   33021   0:00 -bash
28349 28957 28957 28349 pts/3    28969 S    33021   0:00  \_ /bin/sh ./run-many-processes.sh
28957 28958 28957 28349 pts/3    28969 S    33021   0:00  |   \_ /bin/sh ./child.sh background
28958 28961 28957 28349 pts/3    28969 S    33021   0:00  |   |   \_ /bin/sh ./grandchild.sh background
28961 28965 28957 28349 pts/3    28969 S    33021   0:00  |   |   |   \_ sleep 9999
28958 28963 28957 28349 pts/3    28969 S    33021   0:00  |   |   \_ /bin/sh ./grandchild.sh foreground
28963 28967 28957 28349 pts/3    28969 S    33021   0:00  |   |       \_ sleep 9999
28957 28959 28957 28349 pts/3    28969 S    33021   0:00  |   \_ /bin/sh ./child.sh foreground
28959 28960 28957 28349 pts/3    28969 S    33021   0:00  |       \_ /bin/sh ./grandchild.sh background
28960 28964 28957 28349 pts/3    28969 S    33021   0:00  |       |   \_ sleep 9999
28959 28962 28957 28349 pts/3    28969 S    33021   0:00  |       \_ /bin/sh ./grandchild.sh foreground
28962 28966 28957 28349 pts/3    28969 S    33021   0:00  |           \_ sleep 9999
28349 28969 28969 28349 pts/3    28969 R+   33021   0:00  \_ ps fj

The command pkill -P $PID does not kill the grandchild:

> pkill -P "$PID"
./run-many-processes.sh: line 4: 28958 Terminated              ./child.sh background
./run-many-processes.sh: line 4: 28959 Terminated              ./child.sh foreground
ProcessID=28957 ends (./run-many-processes.sh)
[1]+  Done                    ./run-many-processes.sh

> ps fj
 PPID   PID  PGID   SID TTY      TPGID STAT   UID   TIME COMMAND
28348 28349 28349 28349 pts/3    28987 Ss   33021   0:00 -bash
28349 28987 28987 28349 pts/3    28987 R+   33021   0:00  \_ ps fj
    1 28963 28957 28349 pts/3    28987 S    33021   0:00 /bin/sh ./grandchild.sh foreground
28963 28967 28957 28349 pts/3    28987 S    33021   0:00  \_ sleep 9999
    1 28962 28957 28349 pts/3    28987 S    33021   0:00 /bin/sh ./grandchild.sh foreground
28962 28966 28957 28349 pts/3    28987 S    33021   0:00  \_ sleep 9999
    1 28961 28957 28349 pts/3    28987 S    33021   0:00 /bin/sh ./grandchild.sh background
28961 28965 28957 28349 pts/3    28987 S    33021   0:00  \_ sleep 9999
    1 28960 28957 28349 pts/3    28987 S    33021   0:00 /bin/sh ./grandchild.sh background
28960 28964 28957 28349 pts/3    28987 S    33021   0:00  \_ sleep 9999

The command kill -- -$PGID kills all processes including the grandchild.

> kill --    -"$PGID"  # default signal is TERM (kill -15)
> kill -CONT -"$PGID"  # awake stopped processes
> kill -KILL -"$PGID"  # kill -9 to be sure

> ps fj
 PPID   PID  PGID   SID TTY      TPGID STAT   UID   TIME COMMAND
28348 28349 28349 28349 pts/3    29039 Ss   33021   0:00 -bash
28349 29039 29039 28349 pts/3    29039 R+   33021   0:00  \_ ps fj

Conclusion

I notice in this example PID and PGID are equal (28957).
This is why I originally thought kill -- -$PID was enough. But in the case the process is spawn within a Makefile the Process ID is different from the Group ID.

I think kill -- -$(ps -o pgid= $PID | grep -o [0-9]*) is the best simple trick to kill a whole process tree when called from a different Group ID (another process tree).


if you know pass the pid of the parent process, here's a shell script that should work:

for child in $(ps -o pid,ppid -ax | \
   awk "{ if ( \$2 == $pid ) { print \$1 }}")
do
  echo "Killing child process $child because ppid = $pid"
  kill $child
done

I use a little bit modified version of a method described here: https://.com/a/5311362/563175

So it looks like that:

kill `pstree -p 24901 | sed 's/(/\n(/g' | grep '(' | sed 's/(\(.*\)).*/\1/' | tr "\n" " "`

where 24901 is parent's PID.

It looks pretty ugly but does it's job perfectly.


To kill a process tree recursively, use killtree():

#!/bin/bash

killtree() {
    local _pid=$1
    local _sig=${2:--TERM}
    kill -stop ${_pid} # needed to stop quickly forking parent from producing children between child killing and parent killing
    for _child in $(ps -o pid --no-headers --ppid ${_pid}); do
        killtree ${_child} ${_sig}
    done
    kill -${_sig} ${_pid}
}

if [ $# -eq 0 -o $# -gt 2 ]; then
    echo "Usage: $(basename $0) <pid> [signal]"
    exit 1
fi

killtree [email protected]

if you have pstree and perl on your system, you can try this:

perl -e 'kill 9, (`pstree -p PID` =~ m/\((\d+)\)/sg)'

You don't say if the tree you want to kill is a single process group. (This is often the case if the tree is the result of forking from a server start or a shell command line.) You can discover process groups using GNU ps as follows:

 ps x -o  "%p %r %y %x %c "

If it is a process group you want to kill, just use the kill(1) command but instead of giving it a process number, give it the negation of the group number. For example to kill every process in group 5112, use kill -TERM -- -5112.


It's super easy to do this with python using psutil. Just install psutil with pip and then you have a full suite of process manipulation tools:

def killChildren(pid):
    parent = psutil.Process(pid)
    for child in parent.get_children(True):
        if child.is_running():
            child.terminate()

Modified version of zhigang's answer:

#!/usr/bin/env bash
set -eu

killtree() {
    local pid
    for pid; do
        kill -stop $pid
        local cpid
        for cpid in $(pgrep -P $pid); do
            killtree $cpid
        done
        kill $pid
        kill -cont $pid
        wait $pid 2>/dev/null || true
   done
}

cpids() {
    local pid=$1 options=${2:-} space=${3:-}
    local cpid
    for cpid in $(pgrep -P $pid); do
        echo "$space$cpid"
        if [[ "${options/a/}" != "$options" ]]; then
            cpids $cpid "$options" "$space  "
        fi
    done
}

while true; do sleep 1; done &
cpid=$!
for i in $(seq 1 2); do
    cpids $$ a
    sleep 1
done
killtree $cpid
echo ---
cpids $$ a

ps -o pid= --ppid $PPID | xargs kill -9 

The following shell function is similar to many of the other answers, but it works both on Linux and BSD (OS X, etc) without external dependencies like pgrep:

killtree() {
    local parent=$1 child
    for child in $(ps -o ppid= -o pid= | awk "\$1==$parent {print \$2}"); do
        killtree $child
    done
    kill $parent
}

Based on zhigang's answer, this avoids self-killing:

init_killtree() {
    local pid=$1 child

    for child in $(pgrep -P $pid); do
        init_killtree $child
    done
    [ $pid -ne $$ ] && kill -kill $pid
}

This script also work:

#/bin/sh while true do echo "Enter parent process id [type quit for exit]" read ppid if [ $ppid -eq "quit" -o $ppid -eq "QUIT" ];then exit 0 fi for i in `ps -ef| awk '$3 == '$ppid' { print $2 }'` do echo killing $i kill $i done done


If you want to kill a process by name:

killall -9 -g someprocessname

or

pgrep someprocessname | xargs pkill -9 -g

Thanks for your wisdom, folks. My script was leaving some child processes on exit and the negation tip made things easier. I wrote this function to be used in other scripts if necessary:

# kill my group's subprocesses:          killGroup
# kill also myself:                      killGroup -x
# kill another group's subprocesses:     killGroup N  
# kill that group all:                   killGroup -x N
# N: PID of the main process (= process group ID).

function killGroup () {
    local prid mainpid
    case $1 in
        -x) [ -n "$2" ] && kill -9 -$2 || kill -9 -$$ ;;
        "") mainpid=$$ ;;
         *) mainpid=$1 ;;
    esac
    prid=$(ps ax -o pid,pgid | grep $mainpid)
    prid=${prid//$mainpid/}
    kill -9 $prid 2>/dev/null
    return
}

Cheers.


brad's answer is what I'd recommend too, except that you can do away with awk altogether if you use the --ppid option to ps.

for child in $(ps -o pid -ax --ppid $PPID) do ....... done

Here is a variation of @zhigang's answer which does without AWK, relying only on Bash's native parsing possibilities:

function killtree {
  kill -STOP "$1"
  ps -e -o pid= -o ppid= | while read -r pid ppid
                           do
                             [[ $ppid = $1 ]] || continue
                             killtree "$pid"  || true # Skip over failures
                           done
  kill -CONT "$1"          
  kill -TERM "$1"
}

It seems to work fine on both Macs and Linux. In situations where you can't rely on being able to manage process groups -- like when writing scripts for testing a piece of software which must be built in multiple environments -- this tree-walking technique is definitely helpful.


Register your handler with signal.signal like this:

#!/usr/bin/env python
import signal
import sys
def signal_handler(sig, frame):
        print('You pressed Ctrl+C!')
        sys.exit(0)
signal.signal(signal.SIGINT, signal_handler)
print('Press Ctrl+C')
signal.pause()

Code adapted from here.

More documentation on signal can be found here.





linux shell process signals