#! /bin/sh

# Copyright 2004-2017 The MathWorks, Inc.

#========================= realpath.sh (start) ============================
#-----------------------------------------------------------------------------
# Usage: realpath <filename>
# Returns the actual path in the file system of a file. It follows links. 
# It returns an empty path if an error occurs.
# Return status: 0 if successful.
# If return status is 0, the function echoes out the real path to the file.
# Return status 1 Exceeded the maximum number of links to follow.
# Return status 2 Some other error occurred.
realpath() {
    filename_rpath=$1
    SUCCESS_STATUS_rpath=0
    MAX_LINKS_EXCEEDED_rpath=1
    OTHER_ERROR_rpath=2
    #
    # Now filename_rpath is either a file or a link to a file.
    #
    cpath_rpath=`pwd`

    #
    # Follow up to 8 links before giving up. Same as BSD 4.3
    # We cd into the directory where the file is located, and do a /bin/pwd 
    # to get the name of the CWD.  If the file is a symbolic link, we update 
    # the basename of the file and cd into the directory that the link points 
    # to and repeat the process.
    # Once we arrive in a directory where we do not have a soft-link, we are 
    # done.
    n_rpath=1
    maxlinks_rpath=8
    while [ $n_rpath -le $maxlinks_rpath ]
    do
        #
        # Get directory part of $filename_rpath correctly!
        #
        newdir_rpath=`dirname "$filename_rpath"`
        # dirname shouldn't return empty instead of ".", but let's be paranoid.
        if [ -z "${newdir_rpath}" ]; then
            newdir_rpath=".";
        fi
        (cd "$newdir_rpath") > /dev/null 2>&1
        if [ $? -ne 0 ]; then
            # This should not happen.  The file is in a non-existing directory.
            cd "$cpath_rpath"
            return $OTHER_ERROR_rpath
        fi
        cd "$newdir_rpath"
        #
        # Need the function pwd - not the shell built-in one.  The command 
        # /bin/pwd resolves all symbolic links, but the shell built-in one 
        # does not.
        #
        newdir_rpath=`/bin/pwd`
        # Stip the directories off the filename_rpath.
        newbase_rpath=`basename "$filename_rpath"`

        lscmd=`ls -l "$newbase_rpath" 2>/dev/null`
        if [ ! "$lscmd" ]; then
            # This should not happen, the file does not exist.
            cd "$cpath_rpath"
            return $OTHER_ERROR_rpath
        fi
        #
        # Check for link.  The link target is everything after ' -> ' in 
        # the output of ls.
        #
        if [ `expr "$lscmd" : '.*->.*'` -ne 0 ]; then
            filename_rpath=`echo "$lscmd" | sed 's/.*-> //'`
        else
            #
            # We are done.  We found a file and not a symbolic link.
            # newdir_rpath contains the directory name, newbase_rpath contains 
            # the file name.
            cd "$cpath_rpath"
            echo "$newdir_rpath/$newbase_rpath"
            return $SUCCESS_STATUS_rpath
        fi
        n_rpath=`expr $n_rpath + 1`
    done
    # We exceeded the maximum number of links to follow.
    cd "$cpath_rpath"
    return $MAX_LINKS_EXCEEDED_rpath
}
#========================= realpath.sh (end) ==============================

#========================= pathsetup.sh (start) ============================
#-----------------------------------------------------------------------------
# Usage: warnIfNotInBin <full path> <scriptname>
warnIfNotInBin() {
    # Search for toolbox/distcomp/bin in $1.
    if [ `expr "$1" : ".*toolbox/distcomp/bin$"` -eq 0 ]; then
        echo "Warning: $2 should be run only from toolbox/distcomp/bin,"
        echo "or using a symbolic link to toolbox/distcomp/bin/$2."
        echo ""
    fi
}

# THIS MUST BE RUN FIRST BEFORE SETTING UP THE APPLICATION VARIABLES
# Get the fully qualified path to the script
SCRIPT="$0"
REALPATH=`realpath "$SCRIPT"`
# Get the path to distcomp bin BASE from this by removing the name of the shell
# script.
BASE=`echo $REALPATH | sed -e 's;\/[^\/]*$;;g'`
warnIfNotInBin "$BASE" startjobmanager
# Make sure we are in the correct directory to run setbase.sh
cd "$BASE"
# Set base directory variables
. util/setbase.sh
#-----------------------------------------------------------------------------
#========================= pathsetup.sh (end) ==============================

usage()
{
echo
echo "startjobmanager:    Start a job manager process and the associated job manager"
echo "                    lookup process under the mdce service, which maintains them"
echo "                    after that.  The job manager handles the storage of jobs and"
echo "                    the distribution of tasks contained in jobs to MATLAB workers"
echo "                    that are registered with it."
echo "                    The mdce service must already be running on the specified"
echo "                    computer."
echo
echo "Usage:  startjobmanager [ -name job_manager_name ]"
echo "                        [ -remotehost hostname ]"
echo "                        [ -clean | -cleanPreserveJobs ]"
echo "                        [ -baseport port_number ]"
echo "                        [ -v ]"
echo "                        [ -help ]"
echo
echo "-name               Specify the name of the job manager. This identifies the"
echo "                    job manager to MATLAB worker sessions and MATLAB clients."
echo "                    The default is the value of DEFAULT_JOB_MANAGER_NAME in the"
echo "                    mdce_def file."
echo
echo "-remotehost         Specify the name of the host where you want to start the"
echo "                    job manager and the job manager lookup process. If omitted,"
echo "                    they are started on the local host."
echo
echo "-clean              Delete all checkpoint information stored on disk from previous"
echo "                    instances of this job manager before starting. This will"
echo "                    clean the job manager so that it will initialize with no jobs"
echo "                    or tasks."
echo
echo "-cleanPreserveJobs  Delete checkpoint information stored on disk from previous"
echo "                    instances of this job manager before starting. Any existing "
echo "                    jobs and tasks will be preserved. "
echo
echo "-baseport           Specify the base port that the mdce service on the remote host"
echo "                    is using. You only need to specify this if the value of"
echo "                    BASE_PORT in the local mdce_def file does not match the base"
echo "                    port being used by the mdce service on the remote host."
echo
echo "-waitForWorkers     Specify the required minimum number of workers to register"
echo "                    with the jobmanager before starting to process jobs. As long"
echo "                    as the number of workers registered is smaller than this number,"
echo "                    the job manager remains in the paused state."
echo
echo "-v                  Be verbose. Display the progress of the command execution."
echo
echo "-help               Print this help information."
echo
echo "Examples:           1) Start the job manager MyJobManager on the local host."
echo
echo "                    startjobmanager -name MyJobManager"
echo
echo "                    2) Start the job manager MyJobManager on the host JMHost."
echo
echo "                    startjobmanager -name MyJobManager -remotehost JMHost"
echo
echo "                    3) Start the job manager MyJobManager on the local host and"
echo "                       wait for at least 4 workers to register before processing"
echo "                       the queue."
echo
echo "                    startjobmanager -name MyJobManager -waitForWorkers 4"
echo
echo "See also:           mdce, nodestatus, stopjobmanager, startworker and stopworker."
echo
}

#-----------------------------------------------------------------------------

SERVICE_CLEAN="false"
PRESERVE_JOBS="false"
REMOTE_COMMAND_VERBOSITY="false"
CERTIFICATE_FILE=""
READ_DATABASE_DIRECTORY=""
WAIT_ON_NUM_WORKERS="0"
JM_SERVICE_CONFIG_FILE="start-jobmanager.config"
while [ -n "$1" ] ; do
    case $1 in 
    -name)
        SERVICE_NAME="$2"
        shift
        ;;
    -clean)
        SERVICE_CLEAN="true"
        PRESERVE_JOBS="false"
        ;;
    -cleanPreserveJobs)
        SERVICE_CLEAN="true"
        PRESERVE_JOBS="true"
        ;;
    -remotehost)
        REMOTE_HOSTNAME=$2
        shift
        ;;
    -baseport)
        READ_BASE_PORT=$2
        shift
        ;;
    -waitForWorkers)
        WAIT_ON_NUM_WORKERS=$2
        shift
        ;;
    -multicast)
        READ_JOB_MANAGER_HOST="USE_MULTICAST"
        ;;
    -allowResizing)
        MDCS_ALLOW_RESIZING="true"
        ;;
    -v)
        REMOTE_COMMAND_VERBOSITY="true"
        ;;
    -certificate)
        CERTIFICATE_FILE=$2
        shift
        ;;
    -help|-h)
        usage
        exit 1
        ;;

# The rest of the flags are undocumented.
    -GDSendpoint)
        GDS_END_POINT=$2
        shift
        ;;
    -GDSlogintoken)
        GDS_LOGIN_TOKEN=$2
        shift
        ;;
    -GDSqueueid)
        GDS_QUEUE_ID=$2
        shift
        ;;
    -useGDSstorage)
        USE_GDS_STORAGE="true"
        ;;
    -useMSMPI)
        USE_MSMPI="true"
        ;;
    -databaseDirectory)
        READ_DATABASE_DIRECTORY="$2"
        shift
        ;;
    -scriptlogfile)
        READ_LOG_FILE="$2"
        shift
        ;;
    -startResizeService)
        START_RESIZE_SERVICE="true"
        ;;
    -cloudCenterToken)
        CLOUD_CENTER_TOKEN=$2
        shift
        ;;
    -clusterID)
        CLOUD_CENTER_CLUSTER_ID=$2
        shift
        ;;
    -jmServiceConfigFile)
        JM_SERVICE_CONFIG_FILE=$2
        shift
        ;;
    *)
        echo "Error: unrecognized option: $1"
        usage
        exit 1
        ;;
    esac
    shift
done    

# Check the job manager name for invalid characters
. "$UTILBASE/checkInputs"
validateJobManagerName $SERVICE_NAME

# Set the general MDCE environment
. "$UTILBASE/setmdceenv"
sourceMdceDef
defineJRECMD

REMOTE_HOSTNAME=${REMOTE_HOSTNAME:-$HOSTNAME}
BASE_PORT=${READ_BASE_PORT:-$BASE_PORT}
JOB_MANAGER_HOST=${READ_JOB_MANAGER_HOST:-JOB_MANAGER_LOOKUP_NOT_SPECIFIED}
DATABASE_DIRECTORY=${READ_DATABASE_DIRECTORY:-$DATABASE_DIRECTORY}
USE_GDS_STORAGE=${USE_GDS_STORAGE:-"false"}
USE_MSMPI=${USE_MSMPI:-"false"}
MDCS_ALLOW_RESIZING=${MDCS_ALLOW_RESIZING:-"false"}
SCRIPT_LOG_FILE=${READ_LOG_FILE:-""}
START_RESIZE_SERVICE=${START_RESIZE_SERVICE:-"false"}
if [ "$USE_GDS_STORAGE" = "true" ] ; then
    if [ -z "$GDS_END_POINT" -o -z "$GDS_LOGIN_TOKEN" -o -z "$GDS_QUEUE_ID" ] ; then
        echo "Error: -useGDSstorage requires -GDSendpoint, -GDSlogintoken and -GDSqueueid to be specified"
        exit 1
    fi
else
    if [ -n "$GDS_END_POINT" -o -n "$GDS_LOGIN_TOKEN" -o -n "$GDS_QUEUE_ID" ] ; then
        echo "Error: -useGDSstorage must be specified in order to use -GDSendpoint, -GDSlogintoken and -GDSqueueid"
        exit 1
    fi
fi

if [ "$START_RESIZE_SERVICE" = "true" ] ; then
    if [ -z "$CLOUD_CENTER_TOKEN" -o -z "$CLOUD_CENTER_CLUSTER_ID" ] ; then
        echo "Error: -startResizeService requires -cloudCenterToken and -clusterID to be specified"
        exit 1
    fi
else
    if [ -n "$CLOUD_CENTER_TOKEN" -o -n "$CLOUD_CENTER_CLUSTER_ID" ] ; then
        echo "Error: -startResizeService must be specified in order to use -cloudCenterToken and -clusterID"
        exit 1
    fi
fi

# Start the lookup service if not already running.
$JRECMD \
    ${COMMAND_LINE_JRE_MEMORY} \
    ${COMMAND_LINE_JRE_GC} \
    -classpath  "$REMOTE_COMMAND_CLASSPATH" \
    -Djava.library.path="$NATIVE_LIBRARY_PATH" \
    -Djava.security.policy="$CONFIGBASE/jsk-all.policy" \
    -Dcom.mathworks.toolbox.distcomp.matlabroot=$MATBASE \
    -Dcom.mathworks.toolbox.distcomp.remote_command_type="lookup" \
    -Dcom.mathworks.toolbox.distcomp.remote_hostname=$REMOTE_HOSTNAME \
    -Dcom.mathworks.toolbox.distcomp.base_port=$BASE_PORT \
    -Dcom.mathworks.toolbox.distcomp.remote_command_verbosity=$REMOTE_COMMAND_VERBOSITY \
    -Dcom.mathworks.toolbox.distcomp.remote_command_action="start" \
    -Dcom.mathworks.toolbox.distcomp.servicename="$SERVICE_NAME" \
    -Dcom.mathworks.toolbox.distcomp.service_config_file="start-jini.config" \
    -Dcom.mathworks.toolbox.distcomp.clean_checkpoint_info=$SERVICE_CLEAN \
    -Dcom.mathworks.toolbox.distcomp.script_log_file="$SCRIPT_LOG_FILE" \
    com.mathworks.toolbox.distcomp.control.RunCommandSender \
    "$CONFIGBASE/control-startstop.config"
# Exit immediately in case of failure.
if [ $? -ne 0 ] ; then
    exit $?
fi

# Start the job manager itself.
$JRECMD \
    ${COMMAND_LINE_JRE_MEMORY} \
    ${COMMAND_LINE_JRE_GC} \
    -classpath  "$REMOTE_COMMAND_CLASSPATH" \
    -Djava.library.path="$NATIVE_LIBRARY_PATH" \
    -Djava.security.policy="$CONFIGBASE/jsk-all.policy" \
    -Dcom.mathworks.toolbox.distcomp.matlabroot=$MATBASE \
    -Dcom.mathworks.toolbox.distcomp.servicename="$SERVICE_NAME" \
    -Dcom.mathworks.toolbox.distcomp.remote_hostname=$REMOTE_HOSTNAME \
    -Dcom.mathworks.toolbox.distcomp.base_port=$BASE_PORT \
    -Dcom.mathworks.toolbox.distcomp.remote_command_verbosity=$REMOTE_COMMAND_VERBOSITY \
    -Dcom.mathworks.toolbox.distcomp.remote_command_action="start" \
    -Dcom.mathworks.toolbox.distcomp.service_config_file=$JM_SERVICE_CONFIG_FILE \
    -Dcom.mathworks.toolbox.distcomp.clean_checkpoint_info=$SERVICE_CLEAN \
    -Dcom.mathworks.toolbox.distcomp.preserve_job_database=$PRESERVE_JOBS \
    -Dcom.mathworks.toolbox.distcomp.use_GDS_storage=$USE_GDS_STORAGE \
    -Dcom.mathworks.toolbox.distcomp.GDS_login_token=$GDS_LOGIN_TOKEN \
    -Dcom.mathworks.toolbox.distcomp.GDS_end_point=$GDS_END_POINT \
    -Dcom.mathworks.toolbox.distcomp.GDS_queue_id=$GDS_QUEUE_ID \
    -Dcom.mathworks.toolbox.distcomp.use_MSMPI=$USE_MSMPI \
    -Dcom.mathworks.toolbox.distcomp.allowResizing=$MDCS_ALLOW_RESIZING \
    -Dcom.mathworks.toolbox.distcomp.database_directory_override="$DATABASE_DIRECTORY" \
    -Dcom.mathworks.toolbox.distcomp.lookup_hosts="$JOB_MANAGER_HOST" \
    -Dcom.mathworks.toolbox.distcomp.control.certificate_file="$CERTIFICATE_FILE" \
    -Dcom.mathworks.toolbox.distcomp.script_log_file="$SCRIPT_LOG_FILE" \
    -Dcom.mathworks.toolbox.distcomp.wait_on_num_workers=$WAIT_ON_NUM_WORKERS \
    -Dcom.mathworks.toolbox.distcomp.mjs.start_resize_service=$START_RESIZE_SERVICE \
    -Dcom.mathworks.toolbox.distcomp.mjs.cloud_center_token=$CLOUD_CENTER_TOKEN \
    -Dcom.mathworks.toolbox.distcomp.mjs.cluster_id=$CLOUD_CENTER_CLUSTER_ID \
    -Ddistcomp.rmi.server.hostname=$HOSTNAME \
    com.mathworks.toolbox.distcomp.control.RunCommandSender \
    "$CONFIGBASE/control-startstop-jobmanager.config"
JM_EXIT_STATUS=$?

# Handle job manager startup failure by stopping the lookup service if there are
# no job managers running.
if [ $JM_EXIT_STATUS -ne 0 ] ; then
    $JRECMD \
        ${COMMAND_LINE_JRE_MEMORY} \
        ${COMMAND_LINE_JRE_GC} \
        -classpath  "$REMOTE_COMMAND_CLASSPATH" \
        -Djava.library.path="$NATIVE_LIBRARY_PATH" \
        -Djava.security.policy="$CONFIGBASE/jsk-all.policy" \
        -Dcom.mathworks.toolbox.distcomp.matlabroot=$MATBASE \
        -Dcom.mathworks.toolbox.distcomp.remote_command_type="lookup" \
        -Dcom.mathworks.toolbox.distcomp.remote_hostname=$REMOTE_HOSTNAME \
        -Dcom.mathworks.toolbox.distcomp.base_port=$BASE_PORT \
        -Dcom.mathworks.toolbox.distcomp.remote_command_verbosity=$REMOTE_COMMAND_VERBOSITY \
        -Dcom.mathworks.toolbox.distcomp.remote_command_action="stop" \
        -Dcom.mathworks.toolbox.distcomp.servicename="$SERVICE_NAME" \
        -Dcom.mathworks.toolbox.distcomp.service_config_file="start-jini.config" \
        -Dcom.mathworks.toolbox.distcomp.clean_checkpoint_info=$SERVICE_CLEAN \
        -Dcom.mathworks.toolbox.distcomp.script_log_file="$SCRIPT_LOG_FILE" \
        com.mathworks.toolbox.distcomp.control.RunCommandSender \
        "$CONFIGBASE/control-startstop.config"
fi
exit $JM_EXIT_STATUS