Make the check more adjustable to allow some delay between xlog files sent by master, received and replayed

README.md

@@ -11,8 +11,13 @@ This script :
     - retreive from Postgres the last _xlog_ file receive and the _xlog_ file replay
     - check if Postgres recovery configuration file is NOT present (_CRITICAL_ raise if present)
     - retreive master connection informations from Postgres recovery configuration file (_UNKNOWN_ raise on error). Default Postgres master TCP port will be used if port is not specify.
-    - retreive current _xlog_ file from Postgres master server by making a connection on master server (_UNKNOWN_ raise on error).
+    - retreive the current state and sync state of the host from Postgres master server by making a connection on master server (_UNKNOWN_ raise on error).
-    - check if the last receive _xlog_ file is the last replay _xlog_ file (_WARNING_ raise if not)
+    - check if the current state of the host is "streaming" (_CRITICAL_ raise if not)
+    - check if the current sync state of the host is "sync" (_CRITICAL_ raise if not)
+    - if the check of the current XLOG file of the master host is enabled :
+      - retreive current _xlog_ file from Postgres master server by making a connection on master server (_UNKNOWN_ raise on error).
+      - check if the current master _xlog_ file is the last received _xlog_ file (_CRITICAL_ raise if not)
+    - check if the last received _xlog_ file is the last replay _xlog_ file : if not, check the current delay with the last replayed transaction against _replay_warn_delay_ and _replay_crit_delay_ thresholds and raise corresponding error if they are exceeded
     - Return _OK_ state
   - if Postgres is not in recovery mode :
     - check if Postgres recovery configuration file is present (_CRITICAL_ raise if present)
@@ -33,17 +38,23 @@ Requirements
 Usage
 -----
 
-    Usage : ./check_pg_streaming_replication [-h] [-d] [options]
+    Usage : check_pg_streaming_replication [-d] [-h] [options]
-        -u pg_user Specify Postgres user (Default : postgres)
+        -u pg_user              Specify Postgres user (Default : postgres)
-        -b psql_bin Specify psql binary path (Default : /usr/bin/psql)
+        -b psql_bin             Specify psql binary path (Default : /usr/bin/psql)
-        -m pg_main Specify Postgres main directory path
+        -m pg_main              Specify Postgres main directory path
-        (Default : /var/lib/postgresql/9.1/main)
+                                (By default, try to auto-detect it, on your system it :
-        -r recovery_conf Specify Postgres recovery configuration file path
+                                /var/lib/postgresql/9.6/main)
-        (Default : /var/lib/postgresql/9.1/main/recovery.conf)
+        -r recovery_conf        Specify Postgres recovery configuration file path
-        -p pg_port Specify default Postgres master TCP port (Default : 5432)
+                                (Default : [PG_MAIN]/recovery.conf)
-        -D dbname Specify DB name on Postgres hosts to connect on (Default : postgres)
+        -U pg_master_user       Specify Postgres user to use on master (Default : user from recovery.conf file)
-        -d Debug mode
+        -p pg_port              Specify default Postgres master TCP port (Default : 5432)
-        -h Show this message
+        -D dbname               Specify DB name on Postgres master/slave to connect on (Default : PG_USER)
+        -C 1/0                  Enable or disable check if the current XLOG file of the master host is the same
+                                of the last replay XLOG file (Default : 1)
+        -w replay_warn_delay    Specify the replay warning delay in second (Default : 3)
+        -c replay_crit_delay    Specify the replay critical delay in second (Default : 5)
+        -d                      Debug mode
+        -h                      Show this message
 
 Copyright
 ---------

check_pg_streaming_replication

@@ -35,7 +35,11 @@ fi
 RECOVERY_CONF_FILENAME=recovery.conf
 RECOVERY_CONF=""
 PG_DEFAULT_PORT=5432
+PG_DEFAULT_APP_NAME=$( hostname )
 PG_DB=""
+CHECK_CUR_MASTER_XLOG=1
+REPLAY_WARNING_DELAY=3
+REPLAY_CRITICAL_DELAY=5
 
 DEBUG=0
 
@@ -52,13 +56,17 @@ Usage : $0 [-d] [-h] [options]
 	-U pg_master_user	Specify Postgres user to use on master (Default : user from recovery.conf file)
 	-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)
+	-C 1/0			Enable or disable check if the current XLOG file of the master host is the same
+				of the last replay XLOG file (Default : $CHECK_CUR_MASTER_XLOG)
+	-w replay_warn_delay	Specify the replay warning delay in second (Default : $REPLAY_WARNING_DELAY)
+	-c replay_crit_delay	Specify the replay critical delay in second (Default : $REPLAY_CRITICAL_DELAY)
 	-d			Debug mode
 	-h 			Show this message
 EOF
 	exit 0
 }
 
-while getopts "hu:b:m:r:U:p:D:d" OPTION
+while getopts "hu:b:m:r:U:p:D:C:w:c:d" OPTION
 do
 	case $OPTION in
 		u)
@@ -82,6 +90,15 @@ do
 		D)
 			PG_DB=$OPTARG
 		;;
+		C)
+			CHECK_CUR_MASTER_XLOG=$OPTARG
+		;;
+		w)
+			REPLAY_WARNING_DELAY=$OPTARG
+		;;
+		c)
+			REPLAY_CRITICAL_DELAY=$OPTARG
+		;;
 		d)
 			DEBUG=1
 		;;
@@ -134,11 +151,17 @@ function debug() {
 }
 
 debug "Running options :
+PG_DB = $PG_DB
 PG_USER = $PG_USER
 PSQL_BIN = $PSQL_BIN
 PG_MAIN = $PG_MAIN
 RECOVERY_CONF = $RECOVERY_CONF
-PG_DEFAULT_PORT = $PG_DEFAULT_PORT"
+PG_DEFAULT_PORT = $PG_DEFAULT_PORT
+PG_DEFAULT_APP_NAME = $PG_DEFAULT_APP_NAME
+CHECK_CUR_MASTER_XLOG = $CHECK_CUR_MASTER_XLOG
+REPLAY_WARNING_DELAY = $REPLAY_WARNING_DELAY
+REPLAY_CRITICAL_DELAY = $REPLAY_CRITICAL_DELAY
+"
 
 # Postgres is running ?
 if [ $DEBUG -eq 0 ]
@@ -215,29 +238,79 @@ then
 			debug "Master user : $M_USER"
 		fi
 	fi
-	
+
-	# Get current xlog file from master
+	M_APP_NAME=$( echo "$MASTER_CONN_INFOS"| grep 'application_name=' | sed "s/^.*application_name=[ \'\"]*\([^ \'\"]\+\)[ \'\"]*.*$/\1/" )
-	M_CUR_XLOG="$( psql_master_get 'SELECT pg_current_xlog_location()' )"
+	if [ ! -n "$M_APP_NAME" ]
-	if [ ! -n "$M_CUR_XLOG" ]
 	then
-		echo "UNKNOWN : Can't retreive current xlog from master server"
+		debug "Master application name not specified, use default : $PG_DEFAULT_APP_NAME"
+		M_APP_NAME=$PG_DEFAULT_APP_NAME
+	else
+		debug "Master application name : $M_APP_NAME"
+	fi
+
+	# Get current state/sync_state from master
+	M_CUR_STATE_SYNC_STATE="$( psql_master_get "SELECT state,sync_state FROM pg_stat_replication WHERE application_name='$M_APP_NAME';" )"
+	if [ ! -n "$M_CUR_STATE_SYNC_STATE" ]
+	then
+		echo "UNKNOWN : Can't retreive current state and sync state from master server"
 		exit 3
 	fi
-	debug "Master current xlog : $M_CUR_XLOG"
+	debug "Master current state / sync_state : $M_CUR_STATE_SYNC_STATE"
 
-	# Master current xlog is the last receive xlog ?
+	M_CUR_STATE=$( echo "$M_CUR_STATE_SYNC_STATE"|cut -d'|' -f1 )
-	if [ "$M_CUR_XLOG" != "$LAST_XLOG_RECEIVE" ]
+	debug "Master current state : $M_CUR_STATE"
+	if [ "$M_CUR_STATE" != "streaming" ]
 	then
-		echo "CRITICAL : Master current xlog is not the last receive xlog"
+		echo "CRITICAL : this host is not in streaming state according to master host (current state = '$M_CUR_STATE')"
 		exit 2
 	fi
-	debug "Master current xlog is the last receive xlog"
+
+	M_CUR_SYNC_STATE=$( echo "$M_CUR_STATE_SYNC_STATE"|cut -d'|' -f2 )
+	debug "Master current sync state : $M_CUR_SYNC_STATE"
+	if [ "$M_CUR_SYNC_STATE" != "sync" ]
+	then
+		echo "CRITICAL : this host is not synchronized according to master host (current sync state = '$M_CUR_SYNC_STATE')"
+		exit 2
+	fi
+
+	# Check current master XLOG file vs last replay XLOG file
+	if [ "$CHECK_CUR_MASTER_XLOG" == "1" ]
+	then
+		# Get current xlog file from master
+		M_CUR_XLOG="$( psql_master_get 'SELECT pg_current_xlog_location()' )"
+		if [ ! -n "$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"
+	fi
 
 	# 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"
+		debug "/!\ The last receive xlog is NOT the last replay file ('$M_CUR_XLOG' / '$LAST_XLOG_RECEIVE')"
-		exit 1
+		REPLAY_DELAY="$( psql_get 'SELECT EXTRACT(EPOCH FROM now() - pg_last_xact_replay_timestamp());' )"
+		debug "Replay delay is $REPLAY_DELAY second(s)"
+		if [ $( echo "$REPLAY_DELAY >= $REPLAY_CRITICAL_DELAY"|bc -l ) -gt 0 ]
+		then
+			echo "CRITICAL : last receive xlog file is not the last replay file ('$LAST_XLOG_RECEIVE' / '$LAST_XLOG_REPLAY') and replay delay is $REPLAY_DELAY second(s)"
+			exit 2
+		fi
+		if [ $( echo "$REPLAY_DELAY >= $REPLAY_WARNING_DELAY"|bc -l ) -gt 0 ]
+		then
+			echo "WARNING : last receive xlog file is not the last replay file ('$LAST_XLOG_RECEIVE' / '$LAST_XLOG_REPLAY') and replay delay is $REPLAY_DELAY second(s)"
+			exit 1
+		fi
+		debug "Replay delay is not worrying"
 	fi
 	debug "Last receive xlog file is the last replay file"