check_pg_streaming_replication/check_pg_streaming_replication

#!/bin/bash
#
# Nagios plugin to check Postgresql streamin replication state
# 
# Could be use on Master or on standby node
#
# Requirement : 
#
#     On master node : Slaves must be able to connect with user PG_USER
#                      to database postgres as trust
#
#     On standby node : PG_USER must be able to connect localy as trust
#
# Author : Benjamin Renard <brenard@easter-eggs.com>
# Date : Wed, 14 Mar 2012 14:45:55 +0000
# Source : http://git.zionetrix.net/check_pg_streaming_replication
#

PG_USER=postgres
PSQL_BIN=/usr/bin/psql
PG_MAIN=/var/lib/postgresql/9.1/main
if [ -f /etc/debian_version ]
then
	AUTO_PG_MAIN=$( ls -1d /var/lib/postgresql/9*/main 2> /dev/null|sort -n|tail -n 1 )
	[ -n "$AUTO_PG_MAIN" -a -d "$AUTO_PG_MAIN" ] && PG_MAIN=$AUTO_PG_MAIN
elif [ -f /etc/redhat-release ]
then
	AUTO_PG_MAIN=$( ls -1d /var/lib/pgsql/9*/data 2> /dev/null|sort -n|tail -n 1 )
	[ -n "$AUTO_PG_MAIN" -a -d "$AUTO_PG_MAIN" ] && PG_MAIN=$AUTO_PG_MAIN
fi
RECOVERY_CONF_FILENAME=recovery.conf
RECOVERY_CONF=""
PG_DEFAULT_PORT=5432
PG_DB=""

DEBUG=0

function usage () {
	cat << EOF
Usage : $0 [-d] [-h] [options]
	-u pg_user		Specify Postgres user (Default : $PG_USER)
	-b psql_bin		Specify psql binary path (Default : $PSQL_BIN)
	-m pg_main		Specify Postgres main directory path
				(By default, try to auto-detect it, on your system it :
                                $PG_MAIN)
	-r recovery_conf	Specify Postgres recovery configuration file path
				(Default : $PG_MAIN/$RECOVERY_CONF_FILENAME)
	-p pg_port		Specify default Postgres master TCP port (Default : $PG_DEFAULT_PORT)
	-D dbname		Specify DB name on Postgres master/slave to connect on (Default : PG_USER)
	-d			Debug mode
	-h 			Show this message
EOF
	exit 0
}

while getopts "hu:b:m:r:p:D:d" OPTION
do
	case $OPTION in
		u)
			PG_USER=$OPTARG
		;;
		b)
			PSQL_BIN=$OPTARG
		;;
		m)
			PG_MAIN=$OPTARG
		;;
		r)
			RECOVERY_CONF=$OPTARG
		;;
		p)
			PG_DEFAULT_PORT=$OPTARG
		;;
		D)
			PG_DB=$OPTARG
		;;
		d)
			DEBUG=1
		;;
		h)
			usage
		;;
		\?)
			echo -n "Unkown option"
			usage
	esac
done

# Check PG_USER
[ -z "$PG_USER" ] && echo "UNKNOWN : Postgres user not specify" && exit 3
id "$PG_USER" > /dev/null 2>&1
[ $? -ne 0 ] && echo "UNKNOWN : Invalid Postgres user ($PG_USER)" && exit 3

# Check PSQL_BIN
[ ! -x "$PSQL_BIN" ] && echo "UNKNOWN : Invalid psql bin path ($PSQL_BIN)" && exit 3

# Check PG_MAIN
[ ! -d "$PG_MAIN/" ] && echo "UNKNOWN : Invalid Postgres main directory path ($PG_MAIN)" && exit 3

# Check RECOVERY_CONF
[ -z "$RECOVERY_CONF" ] && RECOVERY_CONF="$PG_MAIN/$RECOVERY_CONF_FILENAME"

# Check PG_DEFAULT_PORT
[ $( echo "$PG_DEFAULT_PORT"|grep -c -E '^[0-9]*$' ) -ne 1 ] && "UNKNOWN : Postgres default master TCP port must be an integer." && exit 3

# If PG_DB is not provided with -D parameter, use PG_USER as default value
[ -z "$PB_DB" ] && PG_DB="$PG_USER"

function psql_get () {
	echo "$1"|su - $PG_USER -c "$PSQL_BIN -d "$PG_DB" -t -P format=unaligned"
}

function debug() {
	if [ $DEBUG -eq 1 ]
	then
		echo "[DEBUG] $1"
	fi
}

debug "Running options :
PG_USER = $PG_USER
PSQL_BIN = $PSQL_BIN
PG_MAIN = $PG_MAIN
RECOVERY_CONF = $RECOVERY_CONF
PG_DEFAULT_PORT = $PG_DEFAULT_PORT"

# Postgres is running ?
if [ $DEBUG -eq 0 ]
then
	psql_get '\q' 2> /dev/null
else
	psql_get '\q'
fi
if [ $? -ne 0 ]
then
	echo "CRITICAL : Postgres is not running !"
	exit 2
fi
debug "Postgres is running"

RECOVERY_MODE=0
[ $( psql_get 'SELECT pg_is_in_recovery();' ) == "t" ] && RECOVERY_MODE=1

if [ -f $RECOVERY_CONF ]
then
	debug "File recovery.conf found. Hot-standby mode."
	
	# Check recovery mode
	if [ $RECOVERY_MODE -ne 1 ]
	then
		echo "CRITICAL : Not in recovery mode while recovery.conf file found !"
		exit 2
	fi
	debug "Postgres is in recovery mode"

	LAST_XLOG_RECEIVE=$( psql_get "SELECT pg_last_xlog_receive_location()" )
	debug "Last xlog file receive : $LAST_XLOG_RECEIVE"
	LAST_XLOG_REPLAY=$( psql_get "SELECT pg_last_xlog_replay_location()" )
	debug "Last xlog file replay : $LAST_XLOG_REPLAY"


	# Get master connection informations from recovery.conf file
	MASTER_CONN_INFOS=$( egrep '^ *primary_conninfo' $RECOVERY_CONF|sed "s/^ *primary_conninfo *= *[\"\']\([^\"\']*\)[\"\'].*$/\1/" )
	if [ ! -n "$MASTER_CONN_INFOS" ]
	then
		echo "UNKNOWN : Can't retreive master connection informations form recovery.conf file"
		exit 3
	fi
	debug "Master connection informations : $MASTER_CONN_INFOS"

	M_HOST=$( echo "$MASTER_CONN_INFOS"|sed 's/^.*host= *\([^ ]*\) *.*$/\1/' )
	if [ ! -n "$M_HOST" ]
	then
		echo "UNKNOWN : Can't retreive master host from recovery.conf file"
		exit 3
	fi
	debug "Master host : $M_HOST"

	M_PORT=$( echo "$MASTER_CONN_INFOS"|sed 's/^.*port= *\([^ ]*\) *.*$/\1/' )
	if [ ! -n "$M_PORT" ]
	then
		debug "Master port not specify, use default : $PG_DEFAULT_PORT"
		M_PORT=$PG_DEFAULT_PORT
	else
		debug "Master port : $M_PORT"
	fi

	M_USER=$( echo "$MASTER_CONN_INFOS"|sed 's/^.*user= *\([^ ]*\) *.*$/\1/' )
	if [ ! -n "$M_USER" ]
	then
		debug "Master user not specify, use default : $PG_USER"
		M_USER=$PG_USER
	else
		debug "Master user : $M_USER"
	fi
	
	# Get current xlog file from master
	M_CUR_XLOG="$( echo 'SELECT pg_current_xlog_location()'|su - $PG_USER -c "$PSQL_BIN -U $M_USER -h $M_HOST -p $M_PORT -d $PG_DB -t -P format=unaligned" )"
	if [ ! -n "$M_CUR_XLOG" ]
	then
		echo "UNKNOWN : Can't retreive current xlog from master server"
		exit 3
	fi
	debug "Master current xlog : $M_CUR_XLOG"

	# Master current xlog is the last receive xlog ?
	if [ "$M_CUR_XLOG" != "$LAST_XLOG_RECEIVE" ]
	then
		echo "CRITICAL : Master current xlog is not the last receive xlog"
		exit 2
	fi
	debug "Master current xlog is the last receive xlog"

	# The last receive xlog is the last replay file ?
	if [ "$LAST_XLOG_RECEIVE" != "$LAST_XLOG_REPLAY" ]
	then
		echo "WARNING : last receive xlog file is not the last replay file"
		exit 1
	fi
	debug "Last receive xlog file is the last replay file"

	echo "OK : Hot-standby server is uptodate"
	exit 0
else
	debug "File recovery.conf not found. Master mode."

	# Check recovery mode
	if [ $RECOVERY_MODE -eq 1 ]
	then
		echo "CRITICAL : In recovery mode while recovery.conf file not found !"
		exit 2
	fi
	debug "Postgres is not in recovery mode"

	# Check standby client
	STANDBY_CLIENTS=$( psql_get "SELECT client_addr, sync_state FROM pg_stat_replication;" )
	if [ ! -n "$STANDBY_CLIENTS" ]
	then
		echo "WARNING : no stand-by client connected"
		exit 1
	fi
	debug "Stand-by client(s) : $( echo  -n $STANDBY_CLIENTS|sed 's/\n/ , /g' )"

	STANDBY_CLIENTS_TXT=""
	STANDBY_CLIENTS_COUNT=0
	for line in $STANDBY_CLIENTS
	do
		let STANDBY_CLIENTS_COUNT=STANDBY_CLIENTS_COUNT+1

		IP=$( echo $line|cut -d '|' -f 1 )
		MODE=$( echo $line|cut -d '|' -f 2 )
		STANDBY_CLIENTS_TXT="$STANDBY_CLIENTS_TXT $IP (mode=$MODE)"
	done

	echo "OK : $STANDBY_CLIENTS_COUNT stand-by client(s) connected - $STANDBY_CLIENTS_TXT"
	exit 0
fi