secondo/Algebras/Hadoop/clusterManagement/ps-cluster-uninstall

#!/bin/bash

bin=`dirname "$0"`
bin=`cd "$bin"; pwd`
source $bin/ps-functions

# This script removes an installed parallel Secondo away from a computer cluster.
# Following environment variables must exist on nodes of the cluster.
# * PARALLEL_SECONDO_MASTER
# * PARALLEL_SECONDO_SLAVES
# * PARALLEL_SECONDO
# * PARALLEL_SECONDO_DATASERVER_NAME

# Following documents should be removed.
# * remove all data server pathes
# * remove .parasecrc file and the "source .parasecrc" line in .bashrc file from all slaves
# * remvoe ps_SDK folder in $SECONDO_BUILD_DIR
# * clean the SecondoConfig.ini file




# 0. Process the arguments.

WARNINFO="Warning !! "
ERRORINFO="ERROR !! "

PSDKLOC=$SECONDO_BUILD_DIR/bin/ps_SDK
SECCONF=$SECONDO_BUILD_DIR/bin/SecondoConfig.ini
DSLIST=$PSDKLOC/DSLIST.ui
IPLIST=$PSDKLOC/IPLIST.ui
RCFILENAME=".parasecrc"
### ----------------------------------------------------------------------------------------
# 1. Check the existence of all required environment variables
if [ "$PARALLEL_SECONDO_MASTER" = "" ]; then
  echo "${ERRORINFO}No Parallel Secondo installed before."
  echo "Uninstallation for Parallel Secondo Fails."
  exit -1
fi

if [ ! -f $PARALLEL_SECONDO_MASTER -o ! -f $PARALLEL_SECONDO_SLAVES ]; then
  echo "${ERRORINFO}Master or Slave list doesn't exist."
  echo "Uninstallation for Parallel Secondo Fails."
  exit -1
fi

if [ ! -d $HADOOP_HOME ]; then
  echo "${ERRORINFO}Hadoop_Home is not available."
  echo "Uninstallation for Parallel Secondo Fails."
  exit -1
fi

if [ ! -d $PSDKLOC ]; then
  echo "${WARNINFO}The installation SDK directory is deleted."
  mkdir $PSDKLOC
  if [ $? -ne 0 ]; then
    # In case $SECONDO_BUILD_DIR doesn't exist.
    echo "${ERRORINFO}Creation of SDK directory fails."
    echo "Uninstallation for Parallel Secondo Fails."
    exit -1
  fi
fi

LIP=$(get_localIP)
MIP=$(cat $PARALLEL_SECONDO_MASTER | cut -d ':' -f1)
if [ "$LIP" != "$MIP" ]; then
    echo "${ERRORINFO}The current node is not the master."
    echo "Uninstallation for Parallel Secondo Fails."
    exit -1
fi

# 2. Check there is no Hadoop or Secondo monitors running
JPNUM=$(jps -l | grep -v Jps | grep hadoop | wc -l)
if [ $JPNUM -gt 0 ]; then
  echo "${WARNINFO}The Hadoop is running now, attempt to turn it off ... ... "
  $HADOOP_HOME/bin/stop-all.sh
  WTIME=5
  HPON=true
  while [ $WTIME -gt 0 ]; do
    JPNUM=$(jps -l | grep -v Jps | grep hadoop | wc -l)
    if [ $JPNUM -eq 0 ]; then
      HPON=false
      break
    fi
    let WTIME--
  done
  
  if $HPON ; then
    echo "${ERRORINFO}The Hadoop cannot be turned off automatically."
    echo "Uninstallation for Parallel Secondo Fails."
    exit -1
  fi
else
  echo "No Hadoop deamons are running on this cluster."
fi

IPList=($(get_slaveIPs -m))
$bin/ps-cluster-queryMonitorStatus 2>&1 1>/dev/null
NSNUM=$?   				#The number of nodes without running Secondo Monitors
SMNUM=$(( ${#IPList[*]} - $NSNUM ))	#The number of nodes running Secondo Monitors
if [ $SMNUM -gt 0 ]; then
  echo "${WARNINFO}$SMNUM Secondo monitors are running now, attempt to turn them off ... ... "
  $bin/ps-stop-AllMonitors
  if [ $? -ne 0 ]; then
    echo "${ERRORINFO}Secondo Monitors cannot be turned off automatically."
    echo "Uninstallation for Parallel Secondo Fails."
    exit -1
  fi
else
  echo "No Secondo monitors are running on this cluster."
fi

# 3. remove all data server pathes
cat /dev/null > $DSLIST
for DST in $(cat $PARALLEL_SECONDO_MASTER $PARALLEL_SECONDO_SLAVES | awk '!x[$0]++'); do
  IPADDR=$(echo $DST | cut -d':' -f1)
  DSPATH=$(echo $DST | cut -d':' -f2)
  DSPATH=${DSPATH%/*}
  echo "$IPADDR:$DSPATH" >> $DSLIST
done

function removeDataServer
{
  IPADDR=$1
  DSPATH=$2
  echo "Remove Data Server: $IPADDR:$DSPATH" >&2
  ssh -o LogLevel=quiet $IPADDR "rm -rf $DSPATH" 2>/dev/null &
}

aJobs=()
DBSERVERS=($(cat $DSLIST | tr '\n' ' '))
DSNUM=${#DBSERVERS[*]}
for((DSIDX=0;DSIDX<$DSNUM;))do
  DServ=${DBSERVERS[$DSIDX]}
  IPADDR=$(echo $DServ | cut -d':' -f1)
  DSPATH=$(echo $DServ | cut -d':' -f2)
  
  for((token=1;token<=$PS_PIPE_Width;token++));do
    if [ $DSIDX -gt $DSNUM ]; then
      break
    fi
    
    if [ ! "${aJobs[$token]}" ] || ! kill -0 ${aJobs[$token]} 2>/dev/null; then
      removeDataServer $IPADDR $DSPATH &
      aJobs[$token]=$!
      let DSIDX++
      break
    fi
  done
done

# 4. remove .parasecrc file and the "source .parasecrc" line in .bashrc file from all slaves
cat $DSLIST | cut -d':' -f 1 | awk '!x[$0]++' > $IPLIST
ANODES=($(cat $IPLIST | tr '\n' ' '))
for nodeIP in ${ANODES[*]};do
  ssh -o LogLevel=quiet $nodeIP "sed $(PSEDOPT) \"s/^source [\\\$A-Z_./]*$RCFILENAME//\" \$HOME/.bashrc "
  ssh -o LogLevel=quiet $nodeIP "if [ -f $RCFILENAME ]; then rm ${RCFILENAME}; fi"
  echo "Remove environment setting on node $nodeIP"
done

# 5. remvoe ps_SDK folder in $SECONDO_BUILD_DIR
rm -rf $PSDKLOC

# 6. clean the SecondoConfig.ini file
# Remove all Parallel Secondo Configurations from $SECCONF
# Find the start line 
declare -i ST=$(grep -n "\[ParallelSecondo\]" $SECCONF| cut -d':' -f1)
if [ $ST -gt 0 ]; then
  declare -i ED=$(sed -n "$(( $ST + 1 )),$ p" $SECCONF | grep -n "^[[]" | head -1 | cut -d':' -f1)
  if [ $ED -eq 0 ]; then
    ED=$(wc -l $SECCONF | cut -d' ' -f1)
  else
    ED=$(( $ST + $ED - 1 ))
  fi
  sed $(PSEDOPT) "$(( $ST + 1 )),${ED}s/^\([^#]\)*//" $SECCONF
  # Delete the backup file in MacOSX
  if [ "$SECONDO_PLATFORM" = "mac_osx" ]; then
    rm "${SECCONF}.bak"
  fi
fi





echo -e "\n\n\n************************************************"
echo "Congratulations! The parallel Secondo has been correctly removed from your cluster."
echo -e "\n\n"
echo "************************************************"

exit 0