Initial commit
Benjamin Renard commited on 2012-03-14 15:04:14
Showing 2 changed files, with 161 additions and 0 deletions.
- .gitignore 0000000..b25c15b
- check_pg_streaming_replication 0000000..baaa1cd
| ... | ... |
@@ -0,0 +1 @@ |
| 1 |
+*~ |
| ... | ... |
@@ -0,0 +1,160 @@ |
| 1 |
+#!/bin/bash |
|
| 2 |
+# |
|
| 3 |
+# Nagios plugin to check Postgresql streamin replication state |
|
| 4 |
+# |
|
| 5 |
+# Could be use on Master or on standby node |
|
| 6 |
+# |
|
| 7 |
+# Requirement : |
|
| 8 |
+# |
|
| 9 |
+# On master node : Slaves must be able to connect with user PG_USER |
|
| 10 |
+# to database postgres as trust |
|
| 11 |
+# |
|
| 12 |
+# On standby node : PG_USER must be able to connect localy as trust |
|
| 13 |
+# |
|
| 14 |
+# Author : Benjamin Renard <brenard@easter-eggs.com> |
|
| 15 |
+# Date : Wed, 14 Mar 2012 14:45:55 +0000 |
|
| 16 |
+# Source : http://git.zionetrix.net/check_pg_streaming_replication |
|
| 17 |
+# |
|
| 18 |
+ |
|
| 19 |
+PG_USER=postgres |
|
| 20 |
+PSQL_BIN=/usr/bin/psql |
|
| 21 |
+PG_MAIN=/var/lib/postgresql/9.1/main |
|
| 22 |
+RECOVERY_CONF=$PG_MAIN/recovery.conf |
|
| 23 |
+PG_DEFAULT_PORT=5432 |
|
| 24 |
+ |
|
| 25 |
+DEBUG=0 |
|
| 26 |
+[ "$1" == "-d" ] && DEBUG=1 |
|
| 27 |
+ |
|
| 28 |
+function psql_get () {
|
|
| 29 |
+ echo "$1"|su - $PG_USER -c "$PSQL_BIN -t -P format=unaligned" |
|
| 30 |
+} |
|
| 31 |
+ |
|
| 32 |
+function debug() {
|
|
| 33 |
+ if [ $DEBUG -eq 1 ] |
|
| 34 |
+ then |
|
| 35 |
+ echo "[DEBUG] $1" |
|
| 36 |
+ fi |
|
| 37 |
+} |
|
| 38 |
+ |
|
| 39 |
+# Postgres is running ? |
|
| 40 |
+if [ $DEBUG -eq 0 ] |
|
| 41 |
+then |
|
| 42 |
+ psql_get '\q' 2> /dev/null |
|
| 43 |
+else |
|
| 44 |
+ psql_get '\q' |
|
| 45 |
+fi |
|
| 46 |
+if [ $? -ne 0 ] |
|
| 47 |
+then |
|
| 48 |
+ echo "CRITICAL : Postgres is not running !" |
|
| 49 |
+ exit 2 |
|
| 50 |
+fi |
|
| 51 |
+debug "Postgres is running" |
|
| 52 |
+ |
|
| 53 |
+RECOVERY_MODE=0 |
|
| 54 |
+[ $( psql_get 'SELECT pg_is_in_recovery();' ) == "t" ] && RECOVERY_MODE=1 |
|
| 55 |
+ |
|
| 56 |
+if [ -f $RECOVERY_CONF ] |
|
| 57 |
+then |
|
| 58 |
+ debug "File recovery.conf found. Hot-standby mode." |
|
| 59 |
+ |
|
| 60 |
+ # Check recovery mode |
|
| 61 |
+ if [ $RECOVERY_MODE -ne 1 ] |
|
| 62 |
+ then |
|
| 63 |
+ echo "CRITICAL : Not in recovery mode while recovery.conf file found !" |
|
| 64 |
+ exit 2 |
|
| 65 |
+ fi |
|
| 66 |
+ debug "Postgres is in recovery mode" |
|
| 67 |
+ |
|
| 68 |
+ LAST_XLOG_RECEIVE=$( psql_get "SELECT pg_last_xlog_receive_location()" ) |
|
| 69 |
+ debug "Last xlog file receive : $LAST_XLOG_RECEIVE" |
|
| 70 |
+ LAST_XLOG_REPLAY=$( psql_get "SELECT pg_last_xlog_replay_location()" ) |
|
| 71 |
+ debug "Last xlog file replay : $LAST_XLOG_REPLAY" |
|
| 72 |
+ |
|
| 73 |
+ |
|
| 74 |
+ # Get master connection informations from recovery.conf file |
|
| 75 |
+ MASTER_CONN_INFOS=$( egrep '^ *primary_conninfo' $RECOVERY_CONF|sed "s/^ *primary_conninfo *= *[\"\']\([^\"\']*\)[\"\'].*$/\1/" ) |
|
| 76 |
+ if [ ! -n "$MASTER_CONN_INFOS" ] |
|
| 77 |
+ then |
|
| 78 |
+ echo "UNKNOWN : Can't retreive master connection informations form recovery.conf file" |
|
| 79 |
+ exit 3 |
|
| 80 |
+ fi |
|
| 81 |
+ debug "Master connection informations : $MASTER_CONN_INFOS" |
|
| 82 |
+ |
|
| 83 |
+ M_HOST=$( echo "$MASTER_CONN_INFOS"|sed 's/^.*host= *\([^ ]*\) *.*$/\1/' ) |
|
| 84 |
+ if [ ! -n "$M_HOST" ] |
|
| 85 |
+ then |
|
| 86 |
+ echo "UNKNOWN : Can't retreive master host from recovery.conf file" |
|
| 87 |
+ exit 3 |
|
| 88 |
+ fi |
|
| 89 |
+ debug "Master host : $M_HOST" |
|
| 90 |
+ |
|
| 91 |
+ M_PORT=$( echo "$MASTER_CONN_INFOS"|sed 's/^.*port= *\([^ ]*\) *.*$/\1/' ) |
|
| 92 |
+ if [ ! -n "$M_PORT" ] |
|
| 93 |
+ then |
|
| 94 |
+ debug "Master port not specify, use default : $PG_DEFAULT_PORT" |
|
| 95 |
+ M_PORT=$PG_DEFAULT_PORT |
|
| 96 |
+ else |
|
| 97 |
+ debug "Master port : $M_PORT" |
|
| 98 |
+ fi |
|
| 99 |
+ |
|
| 100 |
+ # Get current xlog file from master |
|
| 101 |
+ M_CUR_XLOG="$( echo 'SELECT pg_current_xlog_location()'|su - $PG_USER -c "$PSQL_BIN -h $M_HOST -p $M_PORT -t -P format=unaligned" )" |
|
| 102 |
+ if [ ! -n "$M_CUR_XLOG" ] |
|
| 103 |
+ then |
|
| 104 |
+ echo "UNKNOWN : Can't retreive current xlog from master server" |
|
| 105 |
+ exit 3 |
|
| 106 |
+ fi |
|
| 107 |
+ debug "Master current xlog : $M_CUR_XLOG" |
|
| 108 |
+ |
|
| 109 |
+ # Master current xlog is the last receive xlog ? |
|
| 110 |
+ if [ "$M_CUR_XLOG" != "$LAST_XLOG_RECEIVE" ] |
|
| 111 |
+ then |
|
| 112 |
+ echo "CRITICAL : Master current xlog is not the last receive xlog" |
|
| 113 |
+ exit 2 |
|
| 114 |
+ fi |
|
| 115 |
+ debug "Master current xlog is the last receive xlog" |
|
| 116 |
+ |
|
| 117 |
+ # The last receive xlog is the last replay file ? |
|
| 118 |
+ if [ "$LAST_XLOG_RECEIVE" != "$LAST_XLOG_REPLAY" ] |
|
| 119 |
+ then |
|
| 120 |
+ echo "WARNING : last receive xlog file is not the last replay file" |
|
| 121 |
+ exit 1 |
|
| 122 |
+ fi |
|
| 123 |
+ debug "Last receive xlog file is the last replay file" |
|
| 124 |
+ |
|
| 125 |
+ echo "OK : Hot-standby server is uptodate" |
|
| 126 |
+ exit 0 |
|
| 127 |
+else |
|
| 128 |
+ debug "File recovery.conf not found. Master mode." |
|
| 129 |
+ |
|
| 130 |
+ # Check recovery mode |
|
| 131 |
+ if [ $RECOVERY_MODE -eq 1 ] |
|
| 132 |
+ then |
|
| 133 |
+ echo "CRITICAL : In recovery mode while recovery.conf file not found !" |
|
| 134 |
+ exit 2 |
|
| 135 |
+ fi |
|
| 136 |
+ debug "Postgres is not in recovery mode" |
|
| 137 |
+ |
|
| 138 |
+ # Check standby client |
|
| 139 |
+ STANDBY_CLIENTS=$( psql_get "SELECT client_addr, sync_state FROM pg_stat_replication;" ) |
|
| 140 |
+ if [ ! -n "$STANDBY_CLIENTS" ] |
|
| 141 |
+ then |
|
| 142 |
+ echo "WARNING : no stand-by client connected" |
|
| 143 |
+ exit 1 |
|
| 144 |
+ fi |
|
| 145 |
+ debug "Stand-by client(s) : $( echo -n $STANDBY_CLIENTS|sed 's/\n/ , /g' )" |
|
| 146 |
+ |
|
| 147 |
+ STANDBY_CLIENTS_TXT="" |
|
| 148 |
+ STANDBY_CLIENTS_COUNT=0 |
|
| 149 |
+ for line in $STANDBY_CLIENTS |
|
| 150 |
+ do |
|
| 151 |
+ let STANDBY_CLIENTS_COUNT=STANDBY_CLIENTS_COUNT+1 |
|
| 152 |
+ |
|
| 153 |
+ IP=$( echo $line|cut -d '|' -f 1 ) |
|
| 154 |
+ MODE=$( echo $line|cut -d '|' -f 2 ) |
|
| 155 |
+ STANDBY_CLIENTS_TXT="$STANDBY_CLIENTS_TXT $IP (mode=$MODE)" |
|
| 156 |
+ done |
|
| 157 |
+ |
|
| 158 |
+ echo "OK : $STANDBY_CLIENTS_COUNT stand-by client(s) connected - $STANDBY_CLIENTS_TXT" |
|
| 159 |
+ exit 0 |
|
| 160 |
+fi |
|
| 0 | 161 |