Monitor a process, then restart it if its down....

[jmckinzie]

Hello,

I have a particlar process that I want to monitor continously and that I want to restart if it stops.

Any ideas?

Thanks,

[Pete Randall]

Perhaps a cron or at job that runs every minute, checks for the presence of said process and re-initiates it if absent?


Pete

[Steven E. Protter]

Shalom,

Lets use sendmail as an example.

Script stubb


procname=sendmail
UNIX95=1

while true
do
sprocs=$(ps -C $procname)
if [ $sprocs -eq 0 ]
then
/sbin/init.d/sendmail start
fi
sleep 1800
done


This script stub(incomplete script) sees if sendmail is running. It does not use grep, which Bill Hassell says is a no-no.

If the number of sendmail processes is 0 it restarts the daemon.

What it needs is code if it can't restart the daemon. Obviously by changing it, you can monitor any daemon.

Note, if you want to monitor oracle, its unfortuneatly quite possible for there to be a process and the database be hoplessly down. You need sql code to accurately monitor oracle.

SEP

[Rodney Hills]

If this is a process that runs all the time, you could put an entry in "/etc/inittab". If you set the second parameter to "respawn", then init will re-launch the process if it terminates.

HTH

-- Rod Hills

[Victor Fridyev]

Hi,

You can put the process into /etc/inittab with respawn option. RTFM for inittab
HTH

[TwoProc · Accepted solution]

Well we do it much like Steve's post. But I actually start that monitoring script from cron every 15 minutes, then have that script check and see if a copy of itself is already running, if it is, it exits. Otherwise, it proceeds to go into a loop and monitor the process that supposed to be running.

This way, if the program dies, the monitor catches it. If the monitor dies, it is relaunched via cron, and if both die, then the monitor launches, and in turn launches the monitored process.

If I may expand on/change Steve's example; here's the script to have run from cron. Be aware that to make it run from cron, you may have put put in absolute paths for every command in the file (e.g. put in "/usr/bin/ps" instead of "ps").

#!/bin/ksh

export CYCLE_TIME=1800
export PROCNAME=sendmail

# Chk to see if monitor is already running
export MYPROG=`basename $0`
let PROGCOUNT=0
export PROGCOUNT

UNIX95=1
PROGCOUNT=`ps -C ${MYPROG}| wc -l`
UNIX95=0
if [ $PROGCOUNT -gt 2 ]; then
exit 0
fi

# Monitor sendmail
while true
do

UNIX95=1
PROGCOUNT=`ps -C ${PROCNAME}|wc -l`
UNIX95=0
if [ $PROGCOUNT -lt 2 ]; then
/sbin/init.d/sendmail start
fi

sleep $CYCLE_TIME

done

[jmckinzie]

#!/bin/ksh

export CYCLE_TIME=1800
export PROCNAME=sendmail

# Chk to see if monitor is already running
export MYPROG=`basename $0`
let PROGCOUNT=0
export PROGCOUNT

UNIX95=1
PROGCOUNT=`ps -C ${MYPROG}| wc -l`
UNIX95=0
if [ $PROGCOUNT -gt 2 ]; then
exit 0
fi

# Monitor sendmail
while true
do

UNIX95=1
PROGCOUNT=`ps -C ${PROCNAME}|wc -l`
UNIX95=0
if [ $PROGCOUNT -lt 2 ]; then
/sbin/init.d/sendmail start
fi

sleep $CYCLE_TIME

done