#!/bin/ksh
#
#
# (c) 2002 Sun Microsystems, Inc. Use is subject to license terms.  

set +u

ckpt_dir=$3

if [ ! -f $ckpt_dir/ckpt.log ]; then
   touch $ckpt_dir/ckpt.log
   chmod 666 $ckpt_dir/ckpt.log
fi

# create temp directory for holding checkpoint info

tmpdir=$ckpt_dir/ckpt.$1
mkdir -p $tmpdir
cd $tmpdir

# create log file

#F=$tmpdir/checkpoint.log
F=~/$REQNAME.co$1
touch $F

print -------------------------------------------------------------  >> $F 2>&1
print `basename $0` called at `date`      >> $F 2>&1
print called by: `id`			 >> $F 2>&1
print with args: $*			 >> $F 2>&1

print `date +"%D %T"` Job $1 "(pid=$2)" restarting >> $ckpt_dir/ckpt.log

currcpr=`cat currcpr`
if [ "$currcpr" != "2" ]; then
   currcpr=1
fi
print Restart command: cpr -r cpr_$1.$currcpr >> $F 2>&1
cpr -r cpr_$1.$currcpr >> $F 2>&1

# Now be careful: The restart command is the parent process of the restarted
# job. SGE is the parent process of the restart command.
# If the job was killed (probably due to a migration request), we need to
# tell our parent that by killing ourselves. SGE will also detect an 
# exit status > 128 analogous to a KILL

exit_status=$?
print Exit status of restart command: $exit_status >> $F 2>&1

# poll for job completion (based on job script pid)

if [ $exit_status -gt 0 ]; then
   exit_status=100
else
   while ps -p $2
   do
      sleep 2
   done
fi

# This doesn't work under Irix 6.2, since the variable $$ is not
# correctly set
#if [ $exit_status -gt 128 ]; then
#   signal=`expr $exit_status - 128`
#   print Killing ourself: kill -$signal $$        >> $F 2>&1   
#   /usr/bin/kill -$signal $pid                   >> $F 2>&1
#fi

# If killing ourselves didn't help or the exit_status was  < 128 exit 
# with the exit status of our child

print `date +"%D %T"` Job $1 "(pid=$2) exiting, status=$exit_status" >> $ckpt_dir/ckpt.log

print Exiting with exit status: $exit_status      >> $F 2>&1
exit $exit_status
