#!/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
RECOVERY_CONF=$PG_MAIN/recovery.conf
PG_DEFAULT_PORT=5432

DEBUG=0
[ "$1" == "-d" ] && DEBUG=1

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

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

# 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
	
	# Get current xlog file from master
	M_CUR_XLOG="$( echo 'SELECT pg_current_xlog_location()'|su - $PG_USER -c "$PSQL_BIN -h $M_HOST -p $M_PORT -t -P format=unaligned" )"
	if [ ! -n "$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