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