heartbeat/resources/OCF/WAS6

#!/bin/sh
# WAS6 
#
# Description:  Manages a Websphere Application Server as an HA resource
#
#
# Author:       Ru Xiang Min
# Support:      linux-ha@lists.linux-ha.org
# License:      GNU General Public License (GPL)
# Copyright:    (C) 2006 International Business Machines China, Ltd., Inc.
#
#
# An example usage in /etc/ha.d/haresources:
#       node1  10.0.0.170 WAS::/opt/IBM/WebSphere/AppServer/profiles/default/config/cells/Node01Cell/nodes/Node01/serverindex.xml
#
# See usage() function below for more details...
#
#         OCF parameters are as below:
#               OCF_RESKEY_profile
#                 (WAS profile name, used for the single server edition of WAS6)

#######################################################################
# Initialization:

. ${OCF_ROOT}/resource.d/heartbeat/.ocf-shellfuncs

#######################################################################

WAS_DIR=/opt/IBM/WebSphere/AppServer
if
  [ ! -d $WAS_DIR ]
then
  WAS_DIR=/usr/IBM/WebSphere/AppServer
fi
STARTTIME=300   #       5 minutes
DEFAULT_WASPORTS="9080"
#
#
WAS_BIN=$WAS_DIR/bin
DEFAULT=default
#
#       Print usage message
#
usage() {
  methods=`WAS_methods | grep -v methods`
  methods=`echo $methods | tr ' ' '|'`
  cat <<-!
        usage: $0 ($methods)

        For the single server edition of WAS6, you have to set the following
        enviroment virable:
                OCF_RESKEY_profile
                        (WAS profile name)


        $0 manages a Websphere Application Server 6(WAS6) as an HA resource

        The 'start' operation starts WAS6.
        The 'stop' operation stops WAS6.
        The 'status' operation reports whether WAS6 is running
        The 'monitor' operation reports whether the WAS6 seems to be working
                (httpd also needs to be working for this case)
        The 'validate-all' operation reports whether the OCF instance parameter (OCF_RESKEY_profileName ) is valid
        The 'methods' operation reports on the methods $0 supports

        This is known to work with the Single Server edition of Websphere.

        The default profile name for the single server edition is:
        $DEFAULT

        The start and stop operations must be run as root.

        The status operation will report a pid of "-" for the
        WAS root process using unless it is run as root.

        If you don't have xmllint on your system, parsing of WAS
        configuration files is very primitive.

        We run servlet/snoop on the seventh transport port listed in
        the config file for the "monitor" operation.

	!
}

meta_data() {
        cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="WAS6">
<version>1.0</version>

<longdesc lang="en">
Resource script for WAS6. It manages a Websphere Application Server (WAS6) as
an HA resource.
</longdesc>
<shortdesc lang="en">WAS6 resource agent</shortdesc>

<parameters>
<parameter name="profile" unique="0" required="1">
<longdesc lang="en">
The WAS profile name.
</longdesc>
<shortdesc lang="en">profile name</shortdesc>
<content type="string" default="$DEFAULT" />
</parameter>
</parameters>

<actions>
<action name="start" timeout="300" />
<action name="stop" timeout="300" />
<action name="status" depth="0" timeout="30" interval="10" start-delay="10" />
<action name="monitor" depth="0" timeout="30" interval="10" start-delay="10" />
<action name="validate-all" timeout="5" />
<action name="meta-data" timeout="5" />
<action name="methods" timeout="5" />
</actions>
</resource-agent>
END
}

#
#       Reformat the XML document in a sort of canonical form
#       if we can.  If we don't have xmllint, we just cat it out
#       and hope for the best ;-)
#
xmlcat() {
  if
    [ "X$XMLcat" = X ]
  then
    XMLcat=`which xmllint 2>/dev/null`
    if
      [ "X${XMLcat}" = X  -o ! -x "${XMLcat}" ]
    then
      XMLcat=cat
    else
      XMLcat="$XMLcat --recover --format"
    fi
  fi
  for j in "$@"
  do
    ${XMLcat} "$j"
  done
}

#
#This is a bit skanky, but it works anyway...
#
# It's not really skanky if we can find xmllint on the system, because it
# reformats tags so they are all on one line, which is all we we need...
#
#
# Get the numbers of the ports WAS should be listening on...
#
# If we don't have xmllint around, then the applicationserver and the
# port= specification have to be on the same line in the XML config file.
#
GetWASPorts() {
  case $1 in
    [0-9]*)     echo "$1" | tr ',' '\012';;
    *)
        xmlcat ${WAS_DIR}/profiles/${WAS_PROFILE_NAME}/config/cells/${WAS_CELL}/nodes/${WAS_NODE}/serverindex.xml | 
        grep port=                              |
        sed -e 's%.*port= *"* *%%'              \
                -e 's%[^0-9][^0-9]*.*$%%'
        # Delete up to port=, throw away optional quote and optional
        #       white space.
        # Throw away everything after the first non-digit.
        # This should leave us the port number all by itself...
  esac
}

#
#       We assume that the seventh port listed in the serverindex.xml
#       is the one we should run servlet/snoop on.
#
GetWASSnoopPort() {
        GetWASPorts "$@" | sed -n '7p' 
}

#
#       Return information on the processname/id for the WAS ports
#
#       pid/java        is the expected output.  Several lines, one per port...
#
#
WASPortInfo() {
  pat=""
  once=yes
  PortCount=0
  for j in $*
  do
    case $pat in
      "")       pat="$j";;
      *)        pat="$pat|$j";;
    esac
    PortCount=`expr $PortCount + 1`
  done
  netstat -ltnp  2>/dev/null| egrep -i "($pat) .*LISTEN" | sed 's%.*LISTEN *%%'
}

#
#       Return the number of WAS ports which are open
#
CheckWASPortsInUse() {
  count=`WASPortInfo "$@" | wc -l`
  echo $count
}

#
#       Return the pid(s) of the processes that have WAS ports open
#
WASPIDs() {
  WASPortInfo "$@" | sort -u | cut -f1 -d/
}

#
#       The version of ps that returns all processes and their (long) args
#       It's only used by WAS_procs, which isn't used for anything ;-)
#
ps_long() {
  ps axww
}


#
#       The total set of WAS processes (single server only)
#
WAS_procs() {
  ps_long | grep -i "config=$1"  | grep -i java | cut -d' ' -f1
}



#
# methods: What methods/operations do we support?
#
WAS_methods() {
  cat <<-!
        start
        stop
        status
        methods
        validate-all
        meta-data
        usage
	!
  if
    have_binary $WGET
  then
    echo "	monitor"
  fi
}

#
#       Return WAS status (silently)
#
WAS_status() {
  WASPorts=`GetWASPorts $1`
  PortsInUse=`CheckWASPortsInUse $WASPorts`
  case $PortsInUse in
    0)  false;;
    *)  true;;
  esac
}

#
#       Report on WAS status to stdout...
#
WAS_report_status() {
  WASPorts=`GetWASPorts $1`
  PortCount=`echo $WASPorts | wc -w`
  PortCount=`echo $PortCount`
  PortsInUse=`CheckWASPortsInUse $WASPorts`
  case $PortsInUse in
    0)  ocf_log debug "WAS: server $1 is stopped."; return $OCF_NOT_RUNNING;;
    *)
        pids=`WASPIDs $WASPorts`
        if
          [ $PortsInUse -ge $PortCount ]
        then
          ocf_log debug "WAS: server $1 is running (pid" $pids "et al)."
        else
          ocf_log debug "WAS: server $1 is running (pid $pids et al) but not listening on all ports."
        fi
        return $OCF_SUCCESS;;
  esac
}

#
#       Monitor WAS - does it really seem to be working?
#
#       For this we invoke the snoop applet via wget.
#
#       This is actually faster than WAS_status above...
#
WAS_monitor() {
  tmpfile=`maketempfile`
  SnoopPort=`GetWASSnoopPort $1`
  output=`$WGET -nv -O$tmpfile  http://localhost:$SnoopPort/snoop 2>&1`
  rc=$?
  if
    [ $rc -eq 0 ]
  then
    if
      grep -i 'user-agent.*Wget' $tmpfile >/dev/null
    then
      : OK
    else
      ocf_log "err" "WAS: $1: no user-agent from snoop application"
      rc=$OCF_ERR_GENERIC
    fi
  else
    ocf_log "err" "WAS: $1: wget failure: $output"
    rc=$OCF_ERR_GENERIC
  fi
  rm -fr $tmpfile
  return $rc
}

#
#       Start WAS instance
#
WAS_start() {
# Launch Arguments:
#     -nowait
#     -quiet
#     -logfile <filename>
#     -replacelog
#     -trace
#     -script [<script filename >] [-background]
#     -timeout <seconds>
#     -statusport <portnumber>
#     -profileName <profile>
#     -help
  if
    [ -x $WAS_BIN/startServer.sh ]
  then
    cmd="$WAS_BIN/startServer.sh server1  -profileName $1"
  fi

  if
    ocf_run $cmd
  then
    if
      WAS_wait_4_start $STARTTIME "$@"
    then
      #true
      return $OCF_SUCCESS
    else
      ocf_log "err" "WAS server $1 did not start correctly"
      return $OCF_ERR_GENERIC
    fi
  else
    #false
    if 
       WAS_wait_4_start $STARTTIME "$@"
    then
       #true
       return $OCF_SUCCESS
    else
       ocf_log "err" "WAS server $1 did not start correctly"
       return $OCF_ERR_GENERIC
    fi
  fi
}

#
#       Wait for WAS to actually start up.
#
#       It seems to take between 30 and 60 seconds for it to
#       start up on a trivial WAS instance.
#
WAS_wait_4_start() {
  max=$1
  retries=0
  shift
  while
    [ $retries -lt $max ]
  do
    if
      WAS_status "$@"
    then
      return $OCF_SUCCESS
    else
      sleep 1
    fi
    retries=`expr $retries + 1`
  done
  WAS_status "$@"
}


#
#       Shut down WAS
#
WAS_stop() {
  # They don't return good return codes...
  # And, they seem to allow anyone to stop WAS (!)
  if
    [ -x $WAS_BIN/stopServer.sh ]
  then
    ocf_run $WAS_BIN/stopServer.sh server1 -profileName $1
  else
    WASPorts=`GetWASPorts $1`
    kill `WASPIDs $WASPorts`
  fi
  if
    WAS_status $1
  then
    ocf_log "err" "WAS: $1 did not stop correctly"
    #false
    return $OCF_ERR_GENERIC
  else
    #true
    return $OCF_SUCCESS
  fi
}

#
#       Check if the port is valid
#
CheckPort() {
  ocf_is_decimal "$1" && [ $1 -gt 0 ]
}

WAS_validate_all() {
  if [ -x $WAS_BIN/startServer.sh ]; then
  # $arg should be profile name 
        if [ ! -f ${WAS_DIR}/profiles/${arg}/config/cells/${WAS_CELL}/nodes/${WAS_NODE}/serverindex.xml ]; then
            ocf_log err "profile [$arg] does not exist"
            exit $OCF_ERR_ARGS
        fi

  # $arg should specify a valid port number at the very least
        local WASPorts=`GetWASPorts $arg`
        if [ -z "$WASPorts" ]; then
            ocf_log err "No port number specified in configuration file of profile [$arg]"
            exit $OCF_ERR_CONFIGURED
        fi

        local port
        local have_valid_port=false
        for port in "$WASPorts"; do
            if CheckPort $port; then
                have_valid_port=true
                break
            fi
        done
        if [ "false" = "$have_valid_port" ]; then
            ocf_log err "No valid port number specified in configuration file of profile [$arg]"
            exit $OCF_ERR_CONFIGURED
        fi

  elif [ -x $WAS_BIN/startupServer.sh ]; then
  # $arg should be port number
        if CheckPort "$arg"; then
            ocf_log err "Port number is required but [$arg] is not valid port number"
            exit $OCF_ERR_ARGS
        fi
  else
  # Do not know hot to validate_all
        ocf_log warn "Do not know how to validate-all, assuming validation OK"
        return $OCF_SUCCESS
  fi
}
#
#       'main' starts here...
#

if
  ( [ $# -ne 1 ] )
then
  usage
  exit $OCF_ERR_ARGS
fi

# These operations don't require OCF instance parameters to be set
case "$1" in

  meta-data)    meta_data
                exit $OCF_SUCCESS;;

  usage)        usage
                exit $OCF_SUCCESS;;

  methods)      WAS_methods
                exit $?;;
  *);;
esac


#
#       Supply default configuration parameter(s)
#

if
   [ -z $OCF_RESKEY_profile ] 
then
    arg=$DEFAULT
else
    arg=$OCF_RESKEY_profile
fi

if
   [ ! -d ${WAS_DIR}/profiles/$arg ]
then
    ocf_log "err" "WAS profile $arg does not exist!"
    usage
    exit $OCF_ERR_ARGS
fi

WAS_PROFILE_NAME=$arg
if [ "${WAS_PROFILE_NAME:=}" != "" ]; then
    WAS_PROFILE_FSDB_SCRIPT=${WAS_DIR}/properties/fsdb/${WAS_PROFILE_NAME}.sh
fi

if [ "${WAS_PROFILE_FSDB_SCRIPT:=}" != "" ] && [ -f ${WAS_PROFILE_FSDB_SCRIPT} ]; then
    . ${WAS_PROFILE_FSDB_SCRIPT}
fi

if [ "${WAS_USER_SCRIPT:=}" != "" ]; then
    . ${WAS_USER_SCRIPT}
fi

# What kind of method was invoked?
case "$1" in

  start)        WAS_start $arg
                exit $?;;

  stop)         WAS_stop $arg
                exit $?;;

  status)       WAS_report_status $arg
                exit $?;;

  monitor)      WAS_monitor $arg
                exit $?;;

  validate-all) WAS_validate_all $arg
                exit $?;;

  *)            usage
                exit $OCF_ERR_UNIMPLEMENTED;;
esac