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 COMPUTE_USER=${COMPUTE_USER:-none}
  32
  33 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  34
  35 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  36     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  37     exit 1
  38 fi
  39
  40 get_compute_host_info() {
  41     (
  42         change_to_doctor_user
  43
  44         # get computer host info which VM boot in
  45         export COMPUTE_HOST=$(openstack server show $VM_NAME | \
  46                 grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }' |
  47                 awk -F '.' '{print $1}')
  48     )
  49     if [[ -z "$COMPUTE_HOST" ]] ; then
  50         echo "ERROR: failed to get compute hostname"
  51         exit 1
  52     fi
  53     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  54         COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  55         if [[ "$INSTALLER_IP" == "none" ]] ; then
  56             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  57             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  58         fi
  59         COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  60              "source stackrc; \
  61              openstack server show $COMPUTE_HOST \
  62              | awk '/ ctlplane network /{print \$5}'")
  63     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  64         COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
  65         COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  66         if [[ -z "$COMPUTE_IP" ]]; then
  67             echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  68             exit 1
  69         fi
  70     fi
  71
  72     # verify connectivity to target compute host
  73     ping -c 1 "$COMPUTE_IP"
  74     if [[ $? -ne 0 ]] ; then
  75         echo "ERROR: can not ping to computer host"
  76         exit 1
  77     fi
  78 }
  79
  80 prepare_compute_ssh() {
  81     ssh_opts_cpu="$ssh_opts"
  82
  83     # get ssh key from installer node
  84     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  85         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa 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     sudo chown $(whoami):$(whoami) instack_key
  91     chmod 400 instack_key
  92     ssh_opts_cpu+=" -i instack_key"
  93
  94     # verify ssh to target compute host
  95     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
  96     if [[ $? -ne 0 ]] ; then
  97         echo "ERROR: can not ssh to computer host"
  98         exit 1
  99     fi
 100 }
 101
 102 get_consumer_ip() {
 103     #get network of the compute_ip address
 104     network_pre=${COMPUTE_IP%.*}
 105     network=${network_pre}.0/24
 106
 107     #if there is a dedicated route, use it
 108     dev=$(ip route |grep $network | awk '{print $3}' |head -n 1)
 109
 110     #if there is no route, use default route
 111     [[ -z $dev ]] && dev=$(ip route |grep ^default | awk '{print $3}' |head -n 1)
 112
 113     CONSUMER_IP=$(ip addr show $dev |grep inet | grep $network_pre | awk '{print $2}' | cut -d'/' -f1)
 114 }
 115
 116 download_image() {
 117     [ -e "$IMAGE_FILE" ] && return 0
 118     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 119 }
 120
 121 register_image() {
 122     openstack image list | grep -q " $IMAGE_NAME " && return 0
 123     openstack image create "$IMAGE_NAME" \
 124                            --public \
 125                            --disk-format "$IMAGE_FORMAT" \
 126                            --container-format bare \
 127                            --file "$IMAGE_FILE"
 128 }
 129
 130 create_test_user() {
 131     openstack user list | grep -q " $DOCTOR_USER " || {
 132         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 133     }
 134     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 135         openstack project create "$DOCTOR_PROJECT"
 136     }
 137     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 138     | grep -q " $DOCTOR_ROLE " || {
 139         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 140                            --project "$DOCTOR_PROJECT"
 141     }
 142 }
 143
 144 change_to_doctor_user() {
 145     export OS_USERNAME="$DOCTOR_USER"
 146     export OS_PASSWORD="$DOCTOR_PW"
 147     export OS_PROJECT_NAME="$DOCTOR_PROJECT"
 148     export OS_TENANT_NAME="$DOCTOR_PROJECT"
 149 }
 150
 151 boot_vm() {
 152     (
 153         # test VM done with test user, so can test non-admin
 154         change_to_doctor_user
 155         openstack server list | grep -q " $VM_NAME " && return 0
 156         openstack server create --flavor "$VM_FLAVOR" \
 157                                 --image "$IMAGE_NAME" \
 158                                 "$VM_NAME"
 159         sleep 1
 160     )
 161
 162 }
 163
 164 create_alarm() {
 165     (
 166         # get vm_id as test user
 167         change_to_doctor_user
 168         ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 169         vm_id=$(openstack server list | grep " $VM_NAME " | awk '{print $2}')
 170         ceilometer alarm-event-create --name "$ALARM_NAME" \
 171             --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 172             --description "VM failure" \
 173             --enabled True \
 174             --repeat-actions False \
 175             --severity "moderate" \
 176             --event-type compute.instance.update \
 177             -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 178     )
 179 }
 180
 181
 182 start_monitor() {
 183     pgrep -f "python monitor.py" && return 0
 184     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 185         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 186 }
 187
 188 stop_monitor() {
 189     pgrep -f "python monitor.py" || return 0
 190     sudo kill $(pgrep -f "python monitor.py")
 191     cat monitor.log
 192 }
 193
 194 start_inspector() {
 195     pgrep -f "python inspector.py" && return 0
 196     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 197 }
 198
 199 stop_inspector() {
 200     pgrep -f "python inspector.py" || return 0
 201     kill $(pgrep -f "python inspector.py")
 202     cat inspector.log
 203 }
 204
 205 start_consumer() {
 206     pgrep -f "python consumer.py" && return 0
 207     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 208 }
 209
 210 stop_consumer() {
 211     pgrep -f "python consumer.py" || return 0
 212     kill $(pgrep -f "python consumer.py")
 213     cat consumer.log
 214 }
 215
 216 wait_for_vm_launch() {
 217     echo "waiting for vm launch..."
 218
 219     (
 220         # get VM state as test user
 221         change_to_doctor_user
 222
 223         count=0
 224         while [[ ${count} -lt 60 ]]
 225         do
 226             state=$(openstack server list | grep " $VM_NAME " | awk '{print $6}')
 227             [[ "$state" == "ACTIVE" ]] && return 0
 228             [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 229             count=$(($count+1))
 230             sleep 1
 231         done
 232         echo "ERROR: time out while waiting for vm launch"
 233         exit 1
 234     )
 235 }
 236
 237 inject_failure() {
 238     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 239     cat > disable_network.sh << 'END_TXT'
 240 #!/bin/bash -x
 241 dev=$(sudo ip route | awk '/^default/{print $5}')
 242 sleep 1
 243 sudo ip link set $dev down
 244 sleep 180
 245 sudo ip link set $dev up
 246 sleep 1
 247 END_TXT
 248     chmod +x disable_network.sh
 249     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 250     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 251 }
 252
 253 calculate_notification_time() {
 254     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 255     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 256     echo "$notified $detected" | \
 257         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 258 }
 259
 260 check_host_status() {
 261     expect_state=$1
 262     (
 263         change_to_doctor_user
 264
 265         host_status_line=$(openstack --os-compute-api-version 2.16 server show $VM_NAME | grep "host_status")
 266         if [[ $? -ne 0 ]] ; then
 267             echo "ERROR: host_status not configured for owner in Nova policy.json"
 268             exit 1
 269         fi
 270
 271         host_status=$(echo $host_status_line | awk '{print $4}')
 272         if [ -z "$host_status" ] ; then
 273             echo "ERROR: host_status not reported by: nova show $VM_NAME"
 274             exit 1
 275         elif [[ "$host_status" != "$expect_state" ]] ; then
 276             echo "ERROR: host_status:$host_status not equal to expect_state: $expect_state"
 277             exit 1
 278         else
 279             echo "$VM_NAME showing host_status: $host_status"
 280         fi
 281     )
 282 }
 283
 284 cleanup() {
 285     set +e
 286     echo "cleanup..."
 287     stop_monitor
 288     stop_inspector
 289     stop_consumer
 290
 291     echo "waiting disabled compute host back to be enabled..."
 292     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 293     sleep 180
 294     check_host_status "UP"
 295     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 296         "[ -e disable_network.log ] && cat disable_network.log"
 297
 298     (
 299         change_to_doctor_user
 300         openstack server list | grep -q " $VM_NAME " && openstack server delete "$VM_NAME"
 301         sleep 1
 302         alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 303         sleep 1
 304         [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 305         sleep 1
 306     )
 307     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 308     sleep 1
 309     [ -n "$image_id" ] && openstack image delete "$image_id"
 310     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 311                               --project "$DOCTOR_PROJECT"
 312     openstack project delete "$DOCTOR_PROJECT"
 313     openstack user delete "$DOCTOR_USER"
 314 }
 315
 316
 317 echo "Note: doctor/tests/run.sh has been executed."
 318
 319 trap cleanup EXIT
 320
 321 echo "preparing VM image..."
 322 download_image
 323 register_image
 324
 325 echo "creating test user..."
 326 create_test_user
 327
 328 echo "creating VM..."
 329 boot_vm
 330 wait_for_vm_launch
 331
 332 echo "get computer host info and prepare to ssh..."
 333 get_compute_host_info
 334 prepare_compute_ssh
 335
 336 echo "creating alarm..."
 337 get_consumer_ip
 338 create_alarm
 339
 340 echo "starting doctor sample components..."
 341 start_monitor
 342 start_inspector
 343 start_consumer
 344
 345 sleep 60
 346 echo "injecting host failure..."
 347 inject_failure
 348 sleep 10
 349
 350 check_host_status "DOWN"
 351 calculate_notification_time
 352
 353 echo "done"