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     if [ ! -r instack_key ]; then
  51         sudo chown $(whoami):$(whoami) instack_key
  52     fi
  53     chmod 400 instack_key
  54     ssh_opts_cpu="$ssh_opts -i instack_key -l heat-admin"
  55 }
  56
  57 download_image() {
  58     [ -e "$IMAGE_FILE" ] && return 0
  59     wget "$IMAGE_URL" -o "$IMAGE_FILE"
  60 }
  61
  62 register_image() {
  63     glance image-list | grep -q " $IMAGE_NAME " && return 0
  64     glance image-create --name "$IMAGE_NAME" \
  65                         --visibility public \
  66                         --disk-format "$IMAGE_FORMAT" \
  67                         --container-format bare \
  68                         --file "$IMAGE_FILE"
  69 }
  70
  71 boot_vm() {
  72     nova list | grep -q " $VM_NAME " && return 0
  73     nova boot --flavor "$VM_FLAVOR" \
  74               --image "$IMAGE_NAME" \
  75               "$VM_NAME"
  76     sleep 1
  77 }
  78
  79 create_alarm() {
  80     ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
  81     vm_id=$(nova list | grep " $VM_NAME " | awk '{print $2}')
  82     ceilometer alarm-event-create --name "$ALARM_NAME" \
  83         --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
  84         --description "VM failure" \
  85         --enabled True \
  86         --repeat-actions False \
  87         --severity "moderate" \
  88         --event-type compute.instance.update \
  89         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
  90 }
  91
  92 start_monitor() {
  93     pgrep -f "python monitor.py" && return 0
  94     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
  95         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
  96 }
  97
  98 stop_monitor() {
  99     pgrep -f "python monitor.py" || return 0
 100     sudo kill $(pgrep -f "python monitor.py")
 101     cat monitor.log
 102 }
 103
 104 start_inspector() {
 105     pgrep -f "python inspector.py" && return 0
 106     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 107 }
 108
 109 stop_inspector() {
 110     pgrep -f "python inspector.py" || return 0
 111     kill $(pgrep -f "python inspector.py")
 112     cat inspector.log
 113 }
 114
 115 start_consumer() {
 116     pgrep -f "python consumer.py" && return 0
 117     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 118 }
 119
 120 stop_consumer() {
 121     pgrep -f "python consumer.py" || return 0
 122     kill $(pgrep -f "python consumer.py")
 123     cat consumer.log
 124 }
 125
 126 wait_for_vm_launch() {
 127     echo "waiting for vm launch..."
 128     while true
 129     do
 130         state=$(nova list | grep " $VM_NAME " | awk '{print $6}')
 131         [[ "$state" == "ACTIVE" ]] && return 0
 132         sleep 1
 133     done
 134 }
 135
 136 inject_failure() {
 137     echo "disabling network of comupte host [$COMPUTE_HOST] for 3 mins..."
 138     cat > disable_network.sh << 'END_TXT'
 139 #!/bin/bash -x
 140 dev=$(/usr/sbin/ip route | awk '/^default/{print $5}')
 141 sleep 1
 142 echo sudo ip link set $dev down
 143 sleep 120
 144 echo sudo ip link set $dev up
 145 sleep 1
 146 END_TXT
 147     chmod +x disable_network.sh
 148     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_IP:"
 149     ssh $ssh_opts_cpu "$COMPUTE_IP:" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 150 }
 151
 152 calculate_notification_time() {
 153     detect=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 154     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 155     duration=$(echo "$notified $detect" | awk '{print $1 - $2 }')
 156     echo "$notified $detect" | \
 157         awk '{d = $1 - $2; if (d < 1 ) print d " OK"; else print d " NG"}'
 158 }
 159
 160 cleanup() {
 161     set +e
 162     echo "cleanup..."
 163     stop_monitor
 164     stop_inspector
 165     stop_consumer
 166     ssh $ssh_opts_cpu $COMPUTE_IP \
 167         "[ -e disable_network.log ] && cat disable_network.log"
 168
 169     nova service-force-down --unset "$COMPUTE_HOST" nova-compute
 170     sleep 1
 171     nova delete "$VM_NAME"
 172     sleep 1
 173     alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 174     sleep 1
 175     [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 176     sleep 1
 177     image_id=$(glance image-list | grep " $IMAGE_NAME " | awk '{print $2}')
 178     sleep 1
 179     [ -n "$image_id" ] && glance image-delete "$image_id"
 180     #TODO: add host status check via nova admin api
 181     echo "waiting disabled compute host back to be enabled..."
 182     sleep 180
 183 }
 184
 185
 186 echo "Note: doctor/tests/run.sh has been executed."
 187
 188 prepare_compute_ssh
 189
 190 trap cleanup ERR
 191
 192 echo "preparing VM image..."
 193 download_image
 194 register_image
 195
 196 echo "starting doctor sample components..."
 197 start_monitor
 198 start_inspector
 199 start_consumer
 200
 201 echo "creating VM and alarm..."
 202 boot_vm
 203 create_alarm
 204 wait_for_vm_launch
 205
 206 sleep 60
 207 echo "injecting host failure..."
 208 inject_failure
 209 sleep 10
 210
 211 calculate_notification_time
 212
 213 cleanup
 214
 215 echo "done"