Benjamin Renard commited on 2011-10-26 16:19:19
Showing 1 changed files, with 138 additions and 0 deletions.
| ... | ... |
@@ -0,0 +1,138 @@ |
| 1 |
+#!/bin/bash |
|
| 2 |
+# Author: Benjamin Renard |
|
| 3 |
+# Date: 2011-10-26 |
|
| 4 |
+# Description: Nagios check plugin to test Linux HA cluster |
|
| 5 |
+# status using cl_status |
|
| 6 |
+ |
|
| 7 |
+# Get program path |
|
| 8 |
+REVISION=1.0 |
|
| 9 |
+PROGNAME=`/usr/bin/basename $0` |
|
| 10 |
+PROGPATH=`echo $0 | /bin/sed -e 's,[\\/][^\\/][^\\/]*$,,'` |
|
| 11 |
+ |
|
| 12 |
+NODE_NAME=`uname -n` |
|
| 13 |
+CL_ST='/usr/bin/cl_status' |
|
| 14 |
+ |
|
| 15 |
+WARNING_NUMBER_PING_NODE=1 |
|
| 16 |
+ |
|
| 17 |
+if [ -f $PROGPATH/utils.sh ] |
|
| 18 |
+then |
|
| 19 |
+ .$PROGPATH/utils.sh |
|
| 20 |
+elif [ -f /usr/lib/nagios/plugins/utils.sh ] |
|
| 21 |
+then |
|
| 22 |
+ . /usr/lib/nagios/plugins/utils.sh |
|
| 23 |
+else |
|
| 24 |
+ echo "Can't find utils.sh" |
|
| 25 |
+ exit 3 |
|
| 26 |
+fi |
|
| 27 |
+ |
|
| 28 |
+function usage {
|
|
| 29 |
+ echo "Nagios plugin to test heartbeat status using cl_status |
|
| 30 |
+ |
|
| 31 |
+Usage: $0 [-h|--help] [-v|--version] |
|
| 32 |
+ --help -h Print this help information |
|
| 33 |
+ --version -v Print version of plugin |
|
| 34 |
+" |
|
| 35 |
+} |
|
| 36 |
+ |
|
| 37 |
+case "$1" in |
|
| 38 |
+ --help | -h) |
|
| 39 |
+ print_revision $PROGNAME $REVISION |
|
| 40 |
+ echo |
|
| 41 |
+ usage |
|
| 42 |
+ echo |
|
| 43 |
+ support |
|
| 44 |
+ exit $STATE_OK |
|
| 45 |
+ ;; |
|
| 46 |
+ --version | -v) |
|
| 47 |
+ print_revision $PROGNAME $REVISION |
|
| 48 |
+ exit $STATE_OK |
|
| 49 |
+ ;; |
|
| 50 |
+esac |
|
| 51 |
+ |
|
| 52 |
+$CL_ST hbstatus > /dev/null |
|
| 53 |
+res=$? |
|
| 54 |
+if [ $res -ne 0 ] |
|
| 55 |
+then |
|
| 56 |
+ echo "CRITICAL: Heartbeat is not running on this node" |
|
| 57 |
+ exit $CRITICAL |
|
| 58 |
+fi |
|
| 59 |
+ |
|
| 60 |
+NODES=`$CL_ST listnodes` |
|
| 61 |
+ |
|
| 62 |
+PERF_DATA="" |
|
| 63 |
+ |
|
| 64 |
+NODE_COUNT=0 |
|
| 65 |
+NODE_ACTIVE=0 |
|
| 66 |
+PING_COUNT=0 |
|
| 67 |
+PING_ACTIVE=0 |
|
| 68 |
+LINK_DOWN=0 |
|
| 69 |
+for node in $NODES |
|
| 70 |
+do |
|
| 71 |
+ if [ "$node" == "$NODE_NAME" ] |
|
| 72 |
+ then |
|
| 73 |
+ let NODE_COUNT=NODE_COUNT+1 |
|
| 74 |
+ let NODE_ACTIVE=NODE_ACTIVE+1 |
|
| 75 |
+ continue |
|
| 76 |
+ fi |
|
| 77 |
+ type=$( $CL_ST nodetype $node ) |
|
| 78 |
+ status=`$CL_ST nodestatus $node` |
|
| 79 |
+ res=$? |
|
| 80 |
+ case "$type" in |
|
| 81 |
+ normal) |
|
| 82 |
+ link=0 |
|
| 83 |
+ link_up=0 |
|
| 84 |
+ let NODE_COUNT=NODE_COUNT+1 |
|
| 85 |
+ if [ "$status" == "active" ] |
|
| 86 |
+ then |
|
| 87 |
+ let NODE_ACTIVE=NODE_ACTIVE+1 |
|
| 88 |
+ fi |
|
| 89 |
+ for link in $( $CL_ST listhblinks $node ) |
|
| 90 |
+ do |
|
| 91 |
+ let link_count=link_count+1 |
|
| 92 |
+ status=$( $CL_ST hblinkstatus $node $link ) |
|
| 93 |
+ if [ "$status" == "up" ] |
|
| 94 |
+ then |
|
| 95 |
+ let link_up=link_up+1 |
|
| 96 |
+ else |
|
| 97 |
+ let LINK_DOWN=LINK_DOWN+1 |
|
| 98 |
+ fi |
|
| 99 |
+ done |
|
| 100 |
+ let warn=link_count-1 |
|
| 101 |
+ PERF_DATA="$PERF_DATA linkup_$node:$link_up;$warn;0;0;$link_count" |
|
| 102 |
+ ;; |
|
| 103 |
+ ping) |
|
| 104 |
+ let PING_COUNT=PING_COUNT+1 |
|
| 105 |
+ if [ "$status" == "ping" ] |
|
| 106 |
+ then |
|
| 107 |
+ let PING_ACTIVE=PING_ACTIVE+1 |
|
| 108 |
+ fi |
|
| 109 |
+ ;; |
|
| 110 |
+ esac |
|
| 111 |
+done |
|
| 112 |
+ |
|
| 113 |
+PERF_DATA="$PERF_DATA ping_active:$PING_ACTIVE;$WARNING_NUMBER_PING_NODE;0;0;$PING_COUNT" |
|
| 114 |
+ |
|
| 115 |
+let warn=NODE_COUNT-1 |
|
| 116 |
+PERF_DATA="$PERF_DATA node_active:$NODE_ACTIVE;$warn;0;0;$NODE_COUNT" |
|
| 117 |
+ |
|
| 118 |
+msg="$NODE_ACTIVE/$NODE_COUNT heartbeat node(s) active(s)" |
|
| 119 |
+if [ $NODE_ACTIVE -eq 0 ] |
|
| 120 |
+then |
|
| 121 |
+ echo "CRITICAL : $msg |$PERF_DATA" |
|
| 122 |
+ exit $STATE_CRITICAL |
|
| 123 |
+elif [ $NODE_ACTIVE -ne $NODE_COUNT ] |
|
| 124 |
+then |
|
| 125 |
+ echo "WARNING : $msg |$PERF_DATA" |
|
| 126 |
+ exit $STATE_WARNING |
|
| 127 |
+elif [ $LINK_DOWN -ne 0 ] |
|
| 128 |
+then |
|
| 129 |
+ echo "WARNING : $msg but all link are not up |$PERF_DATA" |
|
| 130 |
+ exit $STATE_WARNING |
|
| 131 |
+elif [ $PING_ACTIVE -le $WARNING_NUMBER_PING_NODE ] |
|
| 132 |
+then |
|
| 133 |
+ echo "WARNING : $msg but only $PING_ACTIVE ping node(s) up |$PERF_DATA" |
|
| 134 |
+ exit $STATE_WARNING |
|
| 135 |
+else |
|
| 136 |
+ echo "OK : $msg |$PERF_DATA" |
|
| 137 |
+ exit $STATE_OK |
|
| 138 |
+fi |
|
| 0 | 139 |