tests/run.sh

   1 #!/bin/bash -e
   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 [[ "${CI_DEBUG:-true}" == "true" ]] && set -x
  12
  13 IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-disk.img
  14 IMAGE_NAME=cirros
  15 IMAGE_FILE="${IMAGE_NAME}.img"
  16 IMAGE_FORMAT=qcow2
  17 VM_NAME=doctor_vm1
  18 VM_FLAVOR=m1.tiny
  19 ALARM_NAME=doctor_alarm1
  20 INSPECTOR_PORT=12345
  21 CONSUMER_PORT=12346
  22 DOCTOR_USER=doctor
  23 DOCTOR_PW=doctor
  24 DOCTOR_PROJECT=doctor
  25 #TODO: change back to `_member_` when JIRA DOCTOR-55 is done
  26 DOCTOR_ROLE=admin
  27
  28 SUPPORTED_INSTALLER_TYPES="apex local"
  29 INSTALLER_TYPE=${INSTALLER_TYPE:-apex}
  30 INSTALLER_IP=${INSTALLER_IP:-none}
  31
  32 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  33 as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
  34                 --os-tenant-name $DOCTOR_PROJECT"
  35
  36 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  37     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  38     exit 1
  39 fi
  40
  41 get_compute_host_info() {
  42     # get computer host info which VM boot in
  43     COMPUTE_HOST=$(openstack $as_doctor_user server show $VM_NAME |
  44                    grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }' |
  45                    awk -F '.' '{print $1}')
  46     if [[ -z "$COMPUTE_HOST" ]] ; then
  47         echo "ERROR: failed to get compute hostname"
  48         exit 1
  49     fi
  50     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  51         COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  52         if [[ "$INSTALLER_IP" == "none" ]] ; then
  53             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  54             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  55         fi
  56         COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  57              "source stackrc; \
  58              nova show $COMPUTE_HOST \
  59              | awk '/ ctlplane network /{print \$5}'")
  60     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  61         COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
  62         COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  63         if [[ -z "$COMPUTE_IP" ]]; then
  64             echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  65             exit 1
  66         fi
  67     fi
  68
  69     # verify connectivity to target compute host
  70     ping -c 1 "$COMPUTE_IP"
  71     if [[ $? -ne 0 ]] ; then
  72         echo "ERROR: can not ping to computer host"
  73         exit 1
  74     fi
  75 }
  76
  77 prepare_compute_ssh() {
  78     ssh_opts_cpu="$ssh_opts"
  79
  80     # get ssh key from installer node
  81     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  82         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  83         sudo chown $(whoami):$(whoami) instack_key
  84         chmod 400 instack_key
  85         ssh_opts_cpu+=" -i instack_key"
  86     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  87         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  88     fi
  89
  90     # verify ssh to target compute host
  91     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
  92     if [[ $? -ne 0 ]] ; then
  93         echo "ERROR: can not ssh to computer host"
  94         exit 1
  95     fi
  96 }
  97
  98 get_consumer_ip() {
  99     #get network of the compute_ip address
 100     network_pre=${COMPUTE_IP%.*}
 101     network=${network_pre}.0/24
 102
 103     #if there is a dedicated route, use it
 104     dev=$(ip route |grep $network | awk '{print $3}' |head -n 1)
 105
 106     #if there is no route, use default route
 107     [[ -z $dev ]] && dev=$(ip route |grep ^default | awk '{print $3}' |head -n 1)
 108
 109     CONSUMER_IP=$(ip addr show $dev |grep inet | grep $network_pre | awk '{print $2}' | cut -d'/' -f1)
 110 }
 111
 112 download_image() {
 113     [ -e "$IMAGE_FILE" ] && return 0
 114     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 115 }
 116
 117 register_image() {
 118     openstack image list | grep -q " $IMAGE_NAME " && return 0
 119     openstack image create "$IMAGE_NAME" \
 120                            --public \
 121                            --disk-format "$IMAGE_FORMAT" \
 122                            --container-format bare \
 123                            --file "$IMAGE_FILE"
 124 }
 125
 126 create_test_user() {
 127     openstack user list | grep -q " $DOCTOR_USER " || {
 128         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 129     }
 130     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 131         openstack project create "$DOCTOR_PROJECT"
 132     }
 133     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 134     | grep -q " $DOCTOR_ROLE " || {
 135         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 136                            --project "$DOCTOR_PROJECT"
 137     }
 138 }
 139
 140 boot_vm() {
 141     # test VM done with test user, so can test non-admin
 142     openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
 143     openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
 144                             --image "$IMAGE_NAME" \
 145                             "$VM_NAME"
 146     sleep 1
 147 }
 148
 149 create_alarm() {
 150     # get vm_id as test user
 151     ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
 152     vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
 153     ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
 154         --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 155         --description "VM failure" \
 156         --enabled True \
 157         --repeat-actions False \
 158         --severity "moderate" \
 159         --event-type compute.instance.update \
 160         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 161 }
 162
 163
 164 start_monitor() {
 165     pgrep -f "python monitor.py" && return 0
 166     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 167         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 168 }
 169
 170 stop_monitor() {
 171     pgrep -f "python monitor.py" || return 0
 172     sudo kill $(pgrep -f "python monitor.py")
 173     cat monitor.log
 174 }
 175
 176 start_inspector() {
 177     pgrep -f "python inspector.py" && return 0
 178     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 179 }
 180
 181 stop_inspector() {
 182     pgrep -f "python inspector.py" || return 0
 183     kill $(pgrep -f "python inspector.py")
 184     cat inspector.log
 185 }
 186
 187 start_consumer() {
 188     pgrep -f "python consumer.py" && return 0
 189     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 190 }
 191
 192 stop_consumer() {
 193     pgrep -f "python consumer.py" || return 0
 194     kill $(pgrep -f "python consumer.py")
 195     cat consumer.log
 196 }
 197
 198 wait_for_vm_launch() {
 199     echo "waiting for vm launch..."
 200
 201     count=0
 202     while [[ ${count} -lt 60 ]]
 203     do
 204         state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
 205         [[ "$state" == "ACTIVE" ]] && return 0
 206         [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 207         count=$(($count+1))
 208         sleep 1
 209     done
 210     echo "ERROR: time out while waiting for vm launch"
 211     exit 1
 212 }
 213
 214 inject_failure() {
 215     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 216     cat > disable_network.sh << 'END_TXT'
 217 #!/bin/bash -x
 218 dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
 219 sleep 1
 220 sudo ip link set $dev down
 221 sleep 180
 222 sudo ip link set $dev up
 223 sleep 1
 224 END_TXT
 225     sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
 226     chmod +x disable_network.sh
 227     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 228     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 229 }
 230
 231 calculate_notification_time() {
 232     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 233     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 234     echo "$notified $detected" | \
 235         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 236 }
 237
 238 check_host_status() {
 239     expected_state=$1
 240
 241     host_status_line=$(openstack $as_doctor_user --os-compute-api-version 2.16 \
 242                        server show $VM_NAME | grep "host_status")
 243     host_status=$(echo $host_status_line | awk '{print $4}')
 244     if [ -z "$host_status" ] ; then
 245         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 246         exit 1
 247     elif [[ "$host_status" != "$expected_state" ]] ; then
 248         echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
 249         exit 1
 250     else
 251         echo "$VM_NAME showing host_status: $host_status"
 252     fi
 253 }
 254
 255 cleanup() {
 256     set +e
 257     echo "cleanup..."
 258     stop_monitor
 259     stop_inspector
 260     stop_consumer
 261
 262     echo "waiting disabled compute host back to be enabled..."
 263     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 264     sleep 180
 265     check_host_status "UP"
 266     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 267         "[ -e disable_network.log ] && cat disable_network.log"
 268
 269     openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
 270     sleep 1
 271     alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 272     sleep 1
 273     [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
 274     sleep 1
 275
 276     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 277     sleep 1
 278     [ -n "$image_id" ] && openstack image delete "$image_id"
 279     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 280                               --project "$DOCTOR_PROJECT"
 281     openstack project delete "$DOCTOR_PROJECT"
 282     openstack user delete "$DOCTOR_USER"
 283 }
 284
 285
 286 echo "Note: doctor/tests/run.sh has been executed."
 287
 288 trap cleanup EXIT
 289
 290 echo "preparing VM image..."
 291 download_image
 292 register_image
 293
 294 echo "creating test user..."
 295 create_test_user
 296
 297 echo "creating VM..."
 298 boot_vm
 299 wait_for_vm_launch
 300
 301 echo "get computer host info and prepare to ssh..."
 302 get_compute_host_info
 303 prepare_compute_ssh
 304
 305 echo "creating alarm..."
 306 get_consumer_ip
 307 create_alarm
 308
 309 echo "starting doctor sample components..."
 310 start_monitor
 311 start_inspector
 312 start_consumer
 313
 314 sleep 60
 315 echo "injecting host failure..."
 316 inject_failure
 317 sleep 10
 318
 319 check_host_status "DOWN"
 320 calculate_notification_time
 321
 322 echo "done"