utilities/ovs-monitor

   1 #!/bin/sh
   2
   3 # Copyright (C) 2008, 2009 Nicira Networks, Inc.
   4 #
   5 # Licensed under the Apache License, Version 2.0 (the "License");
   6 # you may not use this file except in compliance with the License.
   7 # You may obtain a copy of the License at:
   8 #
   9 #     http://www.apache.org/licenses/LICENSE-2.0
  10 #
  11 # Unless required by applicable law or agreed to in writing, software
  12 # distributed under the License is distributed on an "AS IS" BASIS,
  13 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  14 # See the License for the specific language governing permissions and
  15 # limitations under the License.
  16
  17 PATH=/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin
  18
  19 OPENFLOWD_PID=/var/run/openvswitch/ovs-openflowd.pid
  20 OPENFLOWD_SOCK=/var/run/openvswitch/ovs-openflowd.mgmt
  21 LOG_FILE=/var/log/openvswitch/ovs-monitor.log
  22 INTERVAL=1
  23 FAIL_THRESH=3
  24
  25 usage() {
  26     echo usage: $0 options
  27     echo
  28     echo "OPTIONS:"
  29     echo "  -h   Show this message"
  30     echo "  -p   PID file for ovs-openflowd (default: $OPENFLOWD_PID)"
  31     echo "  -s   Unix socket for ovs-openflowd (default: $OPENFLOWD_SOCK)"
  32     echo "  -l   File to log messages (default: $LOG_FILE)"
  33     echo "  -i   Interval to send probes in seconds (default: $INTERVAL)"
  34     echo "  -c   Number of failed probes before reboot (default: $FAIL_THRESH)"
  35 }
  36
  37 log() {
  38     echo `date +"%b %d %X"`:$1
  39     echo `date +"%b %d %X"`:$1 >> $LOG_FILE
  40 }
  41
  42
  43 while getopts "hp:s:l:i:c:" OPTION; do
  44     case $OPTION in
  45         h)
  46             usage
  47             exit 1
  48             ;;
  49
  50         p)
  51             OPENFLOWD_PID=$OPTARG
  52             ;;
  53
  54         s)
  55             OPENFLOWD_SOCK=$OPTARG
  56             ;;
  57
  58         l)
  59             LOG_FILE=$OPTARG
  60             ;;
  61
  62         i)
  63             INTERVAL=$OPTARG
  64             ;;
  65
  66         c)
  67             FAIL_THRESH=$OPTARG
  68             ;;
  69
  70         *)
  71             echo "Unknown option: ${OPTION}"
  72     esac
  73 done
  74
  75
  76 if [ ! -f $OPENFLOWD_PID ]; then
  77     log "No ovs-openflowd pid file: ${OPENFLOWD_PID}"
  78     echo "No ovs-openflowd pid file: ${OPENFLOWD_PID}"
  79 fi
  80
  81 if [ ! -S $OPENFLOWD_SOCK ]; then
  82     log "No ovs-openflowd sock file: ${OPENFLOWD_SOCK}"
  83     echo "No ovs-openflowd sock file: ${OPENFLOWD_SOCK}"
  84 fi
  85
  86 if [ ! -d `dirname $LOG_FILE` ]; then
  87     mkdir -p `dirname $LOG_FILE`
  88 fi
  89
  90 let DP_DOWN=0
  91 let OPENFLOWD_DOWN=0
  92 log "===== Starting Monitor ===="
  93 while `/bin/true`; do
  94     # Only check for liveness if ovs-openflowd's PID file exists.  The PID
  95     # file is removed when ovs-openflowd is brought down gracefully.
  96     if [ -f $OPENFLOWD_PID ]; then
  97         pid=`cat $OPENFLOWD_PID`
  98         if [ -d /proc/$pid ]; then
  99             # Check if the ovs-openflowd and datapath still can communicate
 100             if [ -S $OPENFLOWD_SOCK ]; then
 101                 ovs-ofctl probe -t 2 unix:$OPENFLOWD_SOCK
 102                 if [ $? -ne 0 ]; then
 103                     log "datapath probe failed"
 104                     let DP_DOWN++
 105                 else
 106                     let DP_DOWN=0
 107                 fi
 108             fi
 109             let OPENFLOWD_DOWN=0
 110         else
 111             log "ovs-openflowd probe failed"
 112             let OPENFLOWD_DOWN++
 113         fi
 114     fi
 115
 116     if [ $OPENFLOWD_DOWN -ge $FAIL_THRESH ]; then
 117         log "Failed to probe ovs-openflowd after ${OPENFLOWD_DOWN} tries...rebooting!"
 118         reboot
 119     fi
 120
 121     if [ $DP_DOWN -ge $FAIL_THRESH ]; then
 122         log "Failed to probe datapath after ${DP_DOWN} tries...rebooting!"
 123         reboot
 124     fi
 125
 126     sleep $INTERVAL
 127 done