Shell script to kill orphaned Hadoop Task processes

kill_hadoop_orphans.sh

#!/bin/bash

# Kill tasks holding on to deleted userlogs. These are most likely abandoned jobs.

function get_bad_pids {
  for i in `ps -ef | grep java | awk '{print $2;}'`; do 
    cnt=`/usr/sbin/lsof -p $i | grep deleted | grep /var/log/hadoop-0.20/userlogs/attempt | wc -l`; 
    if [ $cnt -gt 0 ]; then
      PIDS=$i:$PIDS; 
    fi 
  done
}

PIDS=""
get_bad_pids
IFS=':'
for pid in $PIDS; do 
  kill -9  $pid;
done