My solution to zombie process - Hope this helps

The following is the solution that I've used to kill zombie SQLTURBO processes on a RISC/6000 (AIX 3.2) & Informix Online (5.?). The shell script was executed periodically by a cron process as root. As I mentioned before in a previous e-mail process, part of the script is run as root and the other as informix. The reasoning is root is needed to kill the processes (root has privileges to kill anybody's process) bu informix is needed to examine the database and invoke informix related commands. I can't remember the exact syntax for the cron process but the command to invoke it was: root's crontab ---------------- * 0 6,10,11,12,13,15,16,17,19 * * 1,2,3,4,5 su - informix -c "zkill.frm" >> zombies.log For our users, terminals were dropped after 2 hours of idlesness. Most "zombies" occurred between 11am - 1:30pm (lunch time & people forgot to logoff) and 3:00pm - 6:00 (meetings & leaving for the day), Monday-Friday. So, I had cron run the process during those times as well as first thing in the morning before most people logged on the day. For those unfamiliar with cron, the 5 fields are: Minute 1-59 Hour 0-23 Day of month 1-31 Month 1-12 Day of week 0-6 where 0=Sunday where a comma is used between multiple values, hyphen is used for a range and an asterisk to indicate all possible values. Incidentally, I had a cron process that printed zombies.log weekly and nulled it out to begin over again. Also, I use a userid of "dba_owner" which represents the owner of all the tables. I hope this isn't too crypic and proves to be useful to someone. It probably needs some tweaking as I wasn't able to find the actual crontab syntax. Bournce shell script: zkill.infrm ------------------------------------- date # # Set input field separator (IFS) to a newline IFS=' ' # # Set default values for variables # INFORMIXDIR="/home/informix"ZOMBIES="NO" ZOMBIE_IDS="" # # Read all SQL processes, determine if there are any zombies which are denoted by # the dash ("-") in the process for SQLPROC in `ps -eF user,pid,tty,comm`; do IFS= ' ' # Set IFS to space & TAB set $SQLPROC ZOMBIE_USER=$1 # Get userid PID=$2 PTS=$3 COMM=$4 if [ "$COMM" = "sqlturbo" ] && [ "$PTS" = "-" ] \\ && [ "$ZOMBIE_USER" != "dba_owner" ] ; then ZOMBIES="YES" # Build the list of zombie processes ZOMBIE_IDS="${ZOMBIE_IDS} ${PID} ${ZOMBIE_USER} ${COMM}" fi done # If there are zombies, begin elimination processes if [ "$ZOMBIES" = "YES" ] ; then # Will be echoed to zombie.log echo "These zombies exist for informix..." $ZOMBIE_IDS IFS=' ' # Sorry, but I don't remember when tbconfig(1&2) were set to. We had 2 instances # of our database. for TBCONFIG in tbconfig tbconfig2; do export TBCONFIG IFS=' ' # set IFS to newline # # Check the interface status for each zombie with the Online engine to prevent # cancelling a process that maybe in a critical section of the database # for TBLINE in `$INFORMIXDIR/bin/tbstat -u`; do IFS=' ' # Set IFS to space set $TBLINE for PID in $ZOMBIE_IDS; do # Verify the Process ID & ensure that no flags are set # by Informix stating a critical part of the db if [ "$3" = $PID -a "$2" = "-------" ] ; then #- Kill the connection to the Engine $INFORMIXDIR/bin/tbmode - z $PID 2> /dev/null # Find & Kill fglgo process PROC=`ps - fp $PID | grep $PID` set $PROC PPID=$3 # su back to root (-c to skip prompting for password) # & kill the process su - root -c "kill -9 $PPID" # Log nameto file specified in crontab echo "Killed by informix: " $1 $PPID # Send auto-generated mail to remind user to logoff # before leaving terminal idle. mail $1 < /u/dba_owner/bin/zombie.mail break; fi done done else echo "No sqlturbo zombies found by informix....." fi # # End of zkill.infrm zombie.mail -------------------- Your terminal was idle for more than 2 hours. As a result, the system automatically drops your connection. Please remember to logoff when you leave your terminal. This message is automatically generated so please do not respond via e-mail.