tests/run.sh

   1 #!/bin/bash -ex
   2 ##############################################################################
   3 # Copyright (c) 2016 NEC Corporation and others.
   4 #
   5 # All rights reserved. This program and the accompanying materials
   6 # are made available under the terms of the Apache License, Version 2.0
   7 # which accompanies this distribution, and is available at
   8 # http://www.apache.org/licenses/LICENSE-2.0
   9 ##############################################################################
  10
  11 IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-disk.img
  12 IMAGE_NAME=cirros
  13 IMAGE_FILE="${IMAGE_NAME}.img"
  14 IMAGE_FORMAT=qcow2
  15 VM_NAME=doctor_vm1
  16 VM_FLAVOR=m1.tiny
  17 ALARM_NAME=doctor_alarm1
  18 INSPECTOR_PORT=12345
  19 CONSUMER_PORT=12346
  20
  21 SUPPORTED_INSTALLER_TYPES="apex local"
  22 INSTALLER_TYPE=${INSTALLER_TYPE:-apex}
  23 INSTALLER_IP=${INSTALLER_IP:-none}
  24 COMPUTE_HOST=${COMPUTE_HOST:-overcloud-novacompute-0}
  25 COMPUTE_IP=${COMPUTE_IP:-none}
  26 COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  27 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  28
  29 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  30     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  31     exit 1
  32 fi
  33
  34 prepare_compute_ssh() {
  35     ssh_opts_cpu="$ssh_opts"
  36
  37     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  38         if [[ "$INSTALLER_IP" == "none" ]] ; then
  39             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  40             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  41         fi
  42
  43         if [[ "$COMPUTE_IP" == "none" ]] ; then
  44             COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  45                          "source stackrc; \
  46                           nova show $COMPUTE_HOST \
  47                           | awk '/ ctlplane network /{print \$5}'")
  48         fi
  49
  50         # get ssh key from installer node
  51         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  52         sudo chown $(whoami):$(whoami) instack_key
  53         chmod 400 instack_key
  54         ssh_opts_cpu+=" -i instack_key"
  55     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  56         if [[ "$COMPUTE_IP" == "none" ]] ; then
  57             COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  58             if [[ -z "$COMPUTE_IP" ]]; then
  59                 echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  60                 exit 1
  61             fi
  62         fi
  63
  64         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  65     fi
  66
  67     # verify connectivity to target compute host
  68     ping -c 1 "$COMPUTE_IP"
  69 }
  70
  71 download_image() {
  72     [ -e "$IMAGE_FILE" ] && return 0
  73     wget "$IMAGE_URL" -o "$IMAGE_FILE"
  74 }
  75
  76 register_image() {
  77     glance image-list | grep -q " $IMAGE_NAME " && return 0
  78     glance image-create --name "$IMAGE_NAME" \
  79                         --visibility public \
  80                         --disk-format "$IMAGE_FORMAT" \
  81                         --container-format bare \
  82                         --file "$IMAGE_FILE"
  83 }
  84
  85 boot_vm() {
  86     nova list | grep -q " $VM_NAME " && return 0
  87     nova boot --flavor "$VM_FLAVOR" \
  88               --image "$IMAGE_NAME" \
  89               "$VM_NAME"
  90     sleep 1
  91 }
  92
  93 create_alarm() {
  94     ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
  95     vm_id=$(nova list | grep " $VM_NAME " | awk '{print $2}')
  96     ceilometer alarm-event-create --name "$ALARM_NAME" \
  97         --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
  98         --description "VM failure" \
  99         --enabled True \
 100         --repeat-actions False \
 101         --severity "moderate" \
 102         --event-type compute.instance.update \
 103         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 104 }
 105
 106 start_monitor() {
 107     pgrep -f "python monitor.py" && return 0
 108     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 109         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 110 }
 111
 112 stop_monitor() {
 113     pgrep -f "python monitor.py" || return 0
 114     sudo kill $(pgrep -f "python monitor.py")
 115     cat monitor.log
 116 }
 117
 118 start_inspector() {
 119     pgrep -f "python inspector.py" && return 0
 120     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 121 }
 122
 123 stop_inspector() {
 124     pgrep -f "python inspector.py" || return 0
 125     kill $(pgrep -f "python inspector.py")
 126     cat inspector.log
 127 }
 128
 129 start_consumer() {
 130     pgrep -f "python consumer.py" && return 0
 131     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 132 }
 133
 134 stop_consumer() {
 135     pgrep -f "python consumer.py" || return 0
 136     kill $(pgrep -f "python consumer.py")
 137     cat consumer.log
 138 }
 139
 140 wait_for_vm_launch() {
 141     echo "waiting for vm launch..."
 142     while true
 143     do
 144         state=$(nova list | grep " $VM_NAME " | awk '{print $6}')
 145         [[ "$state" == "ACTIVE" ]] && return 0
 146         sleep 1
 147     done
 148 }
 149
 150 inject_failure() {
 151     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 152     cat > disable_network.sh << 'END_TXT'
 153 #!/bin/bash -x
 154 dev=$(sudo ip route | awk '/^default/{print $5}')
 155 sleep 1
 156 sudo ip link set $dev down
 157 sleep 180
 158 sudo ip link set $dev up
 159 sleep 1
 160 END_TXT
 161     chmod +x disable_network.sh
 162     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 163     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 164 }
 165
 166 calculate_notification_time() {
 167     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 168     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 169     echo "$notified $detected" | \
 170         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 171 }
 172
 173 cleanup() {
 174     set +e
 175     echo "cleanup..."
 176     stop_monitor
 177     stop_inspector
 178     stop_consumer
 179
 180     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 181     sleep 1
 182     nova delete "$VM_NAME"
 183     sleep 1
 184     alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 185     sleep 1
 186     [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 187     sleep 1
 188     image_id=$(glance image-list | grep " $IMAGE_NAME " | awk '{print $2}')
 189     sleep 1
 190     [ -n "$image_id" ] && glance image-delete "$image_id"
 191     #TODO: add host status check via nova admin api
 192     echo "waiting disabled compute host back to be enabled..."
 193     sleep 180
 194     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 195         "[ -e disable_network.log ] && cat disable_network.log"
 196 }
 197
 198
 199 echo "Note: doctor/tests/run.sh has been executed."
 200
 201 prepare_compute_ssh
 202
 203 trap cleanup EXIT
 204
 205 echo "preparing VM image..."
 206 download_image
 207 register_image
 208
 209 echo "starting doctor sample components..."
 210 start_monitor
 211 start_inspector
 212 start_consumer
 213
 214 echo "creating VM and alarm..."
 215 boot_vm
 216 create_alarm
 217 wait_for_vm_launch
 218
 219 sleep 60
 220 echo "injecting host failure..."
 221 inject_failure
 222 sleep 10
 223
 224 calculate_notification_time
 225
 226 echo "done"