#!/bin/sh
#
# Check for zombie jobs (not terminated).
# Also scan logs for ERROR messages
#

# check_for_zombie_jobs storage=STORAGE [client=localhost-fd]

if [ $# = 2 ] ; then
    client="$2"
else 
    client="client"
fi

bin/bconsole -c bin/bconsole.conf <<END_OF_DATA 2>&1 >/dev/null
@output tmp/dir.out
status dir
@output tmp/fd.out
status $client
@output tmp/sd.out
status $1
@output
quit
END_OF_DATA

export zstat
grep "No Jobs running." tmp/dir.out 2>&1 >/dev/null
if [ $? != 0 ] ; then
  echo " "
  echo "  !!!! Zombie Jobs in Director !!!!"
  echo "  !!!! Zombie Jobs in Director !!!!" >>test.out
  echo " "
  zstat=1
  exit 1
fi
grep "No Jobs running." tmp/fd.out 2>&1 >/dev/null
if [ $? != 0 ] ; then
  echo " "
  echo "  !!!! Zombie Jobs in File daemon !!!!"
  echo "  !!!! Zombie Jobs in File daemon !!!!" >>test.out
  echo " "
  zstat=1
  exit 1
fi
grep "No Jobs running." tmp/sd.out 2>&1 >/dev/null
if [ $? != 0 ] ; then
  echo " "
  echo "  !!!! Zombie Jobs in Storage daemon !!!!"
  echo "  !!!! Zombie Jobs in Storage daemon !!!!" >>test.out
  echo " "
  zstat=1
  exit 1
fi
grep "ERROR" tmp/log*.out 2>&1 >/dev/null
if [ $? = 0 ] ; then
  echo " "
  echo "  !!!! ERROR in log output !!!!"
  echo "  !!!! ERROR in log output !!!!" >>test.out
  grep "ERROR" tmp/log*.out
  echo " "
  zstat=1
  exit 1
fi
grep "Fatal Error" tmp/log*.out 2>&1 >/dev/null
if [ $? = 0 ] ; then
  echo " "
  echo "  !!!! Fatal Error in log output !!!!"
  echo "  !!!! Fatal Error in log output !!!!" >>test.out
  grep "Fatal Error" tmp/log*.out
  echo " "
  zstat=1
  exit 1
fi

grep "glibc detected" tmp/log*.out 2>&1 >/dev/null
if [ $? = 0 ] ; then
  echo " "
  echo "  !!!! glibc detected Error in log output !!!!"
  echo "  !!!! glibc detected Error in log output !!!!" >>test.out
  grep "glibc detected" tmp/log*.out
  echo " "
  zstat=1
  exit 1
fi