#!/bin/sh
#
# Description:  Control services under daemontools
#
# Author:       Philip Gwyn
# Support:      philip@awale.qc.ca
# License:      GNU General Public License (GPL)
# Copyright:    (C) 2007 Philp Gwyn / Awalé.
#
#
# OCF_RESKEY_service : the name of the service's sub-directory
# OCF_RESKEY_delay   : Time (in seconds) to let the process start
# OCF_RESKEY_process : name of a command that shows that the service is active
#
# set -x

# Directory that daemontools is installed in
DTDIR=""

HADIR=/etc/ha.d
. $HADIR/shellfuncs

if [ -d "$DTDIR/service" ] ; then
    true
# Normaly it goes in / (put "") but I HATE that, so i use /var/daemontools
elif [ -d "/var/daemontools/service" ] ; then 
    DTDIR="/var/daemontools"
else 
    ha_log "ERROR Can't find daemontools service directory"
    exit 1
fi

DTSERVICE="$DTDIR/service"
SVC="$DTDIR/command/svc"
SVSTAT="$DTDIR/command/svstat"


###############################################################
#### Simple usage statement
usage() {
    echo "Usage: $0 <service> [<delay>] {start|stop|status|restart}"
}

###############################################################
#### Find out if the service is running
is_up() {
    T=$($SVSTAT $SERVICEDIR)
    if [ $? -ne 0 ] ; then
        # failed
        return 0
    fi

    echo $T | grep "$SERVICEDIR: up " 2>/dev/null >/dev/null
    if [ $? -ne 0 ] ; then    
        # failed
        return 0
    fi

    return 1
}

#### Look in ps -e too find if there is a $PROCESS running.
#### This differs from is_up(), in that svc could claim a process is down
#### but not all the processes exited.  This happens if the process doesn't
#### handle TERM properly.  For instance if it doesn't propagate it to any
#### child processes.
#### If PROCESS isn't configed, use is_up()
is_active() {
    if [ "$PROCESS" ] ; then 
        PS=$(ps -e)
        echo $PS | grep -- $PROCESS 2>/dev/null >/dev/null
        if [ $? -ne 0 ] ; then
            # the process isn't running
            return 0
        fi
        return 1
    fi
    is_up
    return $?
}

##### Make sure a process is running
##### This is different from being active; is it responding to requests?
##### This is application-specific
is_running() {
    if [ "$VERIFY" = "" ] ; then
        # no verification to be done
        return 1
    fi
    type=${VERIFY/:[[:digit:]]*/}
    port=${VERIFY/[[:alnum:]]*:/}
    ha_log "INFO Verify type=$type port=$port"
    if [ "$type" = "http" ] ; then
        is_running_http $port
    elif [ "$type" = "IKC" ] ; then
        is_running_ikc $port
    else
        ha_log "ERROR Unknown monitor type $type"
        return 0
    fi
}

##### Test if a web server is running
is_running_http () {
    port=$1
    ret=$(echo -e "GET / HTTP/1.1\nHost: localhost\n\n" | \
            nc -w 10 localhost $port 2>&1 >/dev/null )
    if [ $? -ne 0 ] ; then
        ha_log "ERROR $ret"
        return 0
    fi
    return 1
}

is_running_ikc () {
    export IKC_port=$1
    perl << 'PERL'
use strict;
use warnings;
use POSIX qw( strftime );
use POE::Component::IKC::ClientLite;

sub ha_log {
    my $msg = join '', @_;
    print STDERR strftime $ENV{HA_DATEFMT}, localtime;
    print STDERR $msg;
    print STDERR "\n" unless $msg =~ /\n$/;
}
sub ha_die {
    ha_log @_;
    exit 1;
}

my $port=$ENV{IKC_port};
ha_die "ERROR we need IKC_port" unless $port;

my $poe = POE::Component::IKC::ClientLite->spawn(
                ip=>'127.0.0.1',
                port=>$port,
                timeout=>10,
                connect_timeout=>10
            );
ha_die "ERROR Unable to connect to 127.0.0.1:$port: ",
                POE::Component::IKC::ClientLite::error unless $poe;

my $pong = $poe->call( "IKC/ping" );
ha_log "INFO ikc://127.0.0.1:33100/IKC/ping responded $pong";
exit( $pong eq 'PONG' ? 0 : 1 );
PERL

    ret=$?
    unset IKC_port
    if [ $ret -ne 0 ] ; then
        return 0
    else
        return 1
    fi
}

###############################################################
##### Wait $DELAY seconds for is_active() to become $want :
#####    1 = process exists
#####    0 = process doesn't exist
##### is_active is polled every $DELAY seconds
wait_active() {
    want=$1
    is_active
    if [ $? -eq $want ] ; then
        return 1
    fi

    if [ "${DELAY:-0}" -gt 0 ] ; then
        waiting_for=$PROCESS
        if [ "$waiting_for" = "" ] ; then
            waiting_for=service
        fi

        ha_log "INFO $SERVICE: Giving $waiting_for $DELAY seconds"
        end=$(expr `date +%s` + $DELAY)
        is_active
        until [ $? -eq $want ] ; do
            ha_log "INFO $SERVICE: Waiting for $waiting_for"
            sleep 1
            [ $end -lt $(date +%s) ] && return 0
            is_active
        done
    fi
    return 1
}

###############################################################
kill_off() {
    sig=$1
    is_active
    if [ $? -eq 1 ] ; then
        ha_log "ERROR $SERVICE: Stopping with killall -$sig $PROCESS"
        killall -$sig $PROCESS
        return 1
    fi
    return 0
}


###############################################################
#### Check params
check() {
    if [ ! -d "$SERVICEDIR" -o ! -x "$SERVICEDIR/run" ] ; then
        ha_log "ERROR $SERVICE: Doesn't look like a daemontools service ($SERVICEDIR)"
        exit 6
    fi
}

validate_all() {
    if [ ! -d "$SERVICEDIR" -o ! -x "$SERVICEDIR/run" ] ; then
        exit 2
    fi
}


###############################################################
#### Start a service
start() {
    is_up
    if [ $? -eq 0 ] ; then
        touch $SERVICEDIR/down
        ha_log "INFO $SERVICE: Starting"
        ha_log "DEBUG $SERVICE: $SVC -o $SERVICEDIR"
        $SVC -o $SERVICEDIR
    fi

    wait_active 1

    is_up || is_active
    if [ $? -eq 0 ] ; then
        ha_log "ERROR $SERVICE: Startup failed!"
        return 7
    fi
}

###############################################################
#### Stop a service
stop() {
    is_up
    if [ $? -eq 1 ] ; then
        ha_log "INFO $SERVICE: Stoping"
        ha_log "DEBUG $SERVICE: $SVC -d $SERVICEDIR"
        $SVC -d $SERVICEDIR
        touch $SERVICEDIR/down
    fi

    wait_active 0
    if [ "$PROCESS" ] ; then 
        kill_off 15
        if [ $? -eq 1 ] ; then
            wait_active 0
            kill_off 9
            if [ $? -eq 1 ] ; then 
                wait_active 0
            fi        
        fi
    fi

    is_up || is_active
    if [ $? -eq 1 ] ; then
        ha_log "ERROR $SERVICE: Stop failed!"
        return 3
    fi

}

###############################################################
####
reload () {
    is_up
    if [ $? -eq 1 ] ; then
        is_active
        if [ $? -eq 1 ] ; then
            if [ "$PROCESS" ] ; then
                ha_log "INFO: $PROCESS: Sending SIGHUP"
                killall -HUP $PROCESS
            else
                ha_log "INFO: $SERVICE: Sending SIGHUP to service"
                $SVC -h $SERVICEDIR
            fi
            return 0
        fi
    fi
    return 7
}

###############################################################
#### Status of a service
status() {
    # ha_log "meta_interval=${OCF_RESKEY_CRM_meta_interval}"
    is_up
    if [ $? -eq 1 ] ; then
        is_active
        if [ $? -eq 1 ] ; then
            is_running
            if [ $? -eq 1 ] ; then
                # ha_log "INFO $SERVICE is running"
                echo "$SERVICE is running"
                return 0
            else
                echo "$SERVICE is active but not running"
                ha_log "INFO $SERVICE is active but not running"
            fi
        fi
        return 1
    fi
    # daemontools claims the service is down... but maybe it's still
    # exiting
    is_active
    if [ $? -eq 1 ] ; then
        # ha_log "INFO $SERVICE is down"
        echo "$SERVICE is down but process is active"
        ha_log "INFO $SERVICE is down but process is active"
        return 1
    fi
    echo "$SERVICE is stopped"
    ha_log "INFO $SERVICE is stopped"
    return 7
}

###############################################################
meta_data() {
    cat <<END
    <?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="daemontools">
<version>1.0</version>
<longdesc lang="en">
Manages daemontool services.
http://cr.yp.to/daemontools.html
</longdesc>
<shortdesc lang="en">Manages daemontool services.</shortdesc>

<parameters>

<parameter name="service" unique="1" required="1">
<shortdesc lang="en">Service name</shortdesc>
<longdesc lang="en">
Name of the service to manage.  Must be a directory under
$DTDIR and contain an executable run command.
</longdesc>
<content type="string" default="" />
</parameter>

<parameter name="delay" unique="1" required="0">
<shortdesc lang="en">Delay</shortdesc>
<longdesc lang="en">
Delay allowed for a service to start or stop.  In seconds.
</longdesc>
<content type="integer" default="10" />
</parameter>

<parameter name="process" unique="1" required="0">
<shortdesc lang="en">Process name</shortdesc>
<longdesc lang="en">
Name of a process to check to see if the service is active.
</longdesc>
<content type="string" default="" />
</parameter>

<parameter name="verify" unique="1" required="0">
<shortdesc lang="en">Application specific verification</shortdesc>
<longdesc lang="en">
Names the application-specific verification that must be done.  Just
because process is running doesn't mean it is working.
</longdesc>
<content type="string" default="" />
</parameter>

</parameters>

<actions>
<action name="start" timeout="90s" />
<action name="stop" timeout="100s" />
<action name="monitor" depth="10" timeout="20s" interval="5s" start-delay="1s" />
<action name="validate-all" timeout="30s" />
<action name="meta-data" timeout="5s" />
</actions>
</resource-agent>
END
}

###############################################################
#### Parse the params
SERVICE=$OCF_RESKEY_service
DELAY=${OCF_RESKEY_delay:-10}
PROCESS=${OCF_RESKEY_process}
VERIFY=${OCF_RESKEY_verify}
SERVICEDIR="$DTSERVICE/$SERVICE"

###############################################################
#### Run the command
# ha_log "INFO $SERVICEDIR ($1)"

_status=0
case "$1" in
    start)
        check
        start
        _status=$?
        ;;
    stop)
        check
        stop
        _status=$?
        ;;
    restart|reset)
        check
        stop
        start
        _status=$?
        ;;
    reload)
        check
        reload
        _status=$?
        ;;
    monitor|status)
        check
        status
        _status=$?
        ;;
    meta-data)
        meta_data
        ;;
    validate-all)
        validate_all
        ;;
    *)
        usage >&2
        _status=1
        ;;
esac

exit $_status