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 $1}')
  48     )
  49     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  50         COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  51         if [[ "$INSTALLER_IP" == "none" ]] ; then
  52             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  53             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  54         fi
  55         COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  56              "source stackrc; \
  57              openstack server show $COMPUTE_HOST \
  58              | awk '/ ctlplane network /{print \$5}'")
  59     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  60         COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
  61         COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  62         if [[ -z "$COMPUTE_IP" ]]; then
  63             echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  64             exit 1
  65         fi
  66     fi
  67
  68     # verify connectivity to target compute host
  69     ping -c 1 "$COMPUTE_IP"
  70     if [[ $? -ne 0 ]] ; then
  71         echo "ERROR: can not ping to computer host"
  72         exit 1
  73     fi
  74 }
  75
  76 prepare_compute_ssh() {
  77     ssh_opts_cpu="$ssh_opts"
  78
  79     # get ssh key from installer node
  80     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  81         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  82     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  83         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  84     fi
  85
  86     sudo chown $(whoami):$(whoami) instack_key
  87     chmod 400 instack_key
  88     ssh_opts_cpu+=" -i instack_key"
  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 change_to_doctor_user() {
 141     export OS_USERNAME="$DOCTOR_USER"
 142     export OS_PASSWORD="$DOCTOR_PW"
 143     export OS_PROJECT_NAME="$DOCTOR_PROJECT"
 144     export OS_TENANT_NAME="$DOCTOR_PROJECT"
 145 }
 146
 147 boot_vm() {
 148     (
 149         # test VM done with test user, so can test non-admin
 150         change_to_doctor_user
 151         openstack server list | grep -q " $VM_NAME " && return 0
 152         openstack server create --flavor "$VM_FLAVOR" \
 153                                 --image "$IMAGE_NAME" \
 154                                 "$VM_NAME"
 155         sleep 1
 156     )
 157
 158 }
 159
 160 create_alarm() {
 161     (
 162         # get vm_id as test user
 163         change_to_doctor_user
 164         ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 165         vm_id=$(openstack server list | grep " $VM_NAME " | awk '{print $2}')
 166         ceilometer alarm-event-create --name "$ALARM_NAME" \
 167             --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 168             --description "VM failure" \
 169             --enabled True \
 170             --repeat-actions False \
 171             --severity "moderate" \
 172             --event-type compute.instance.update \
 173             -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 174     )
 175 }
 176
 177
 178 start_monitor() {
 179     pgrep -f "python monitor.py" && return 0
 180     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 181         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 182 }
 183
 184 stop_monitor() {
 185     pgrep -f "python monitor.py" || return 0
 186     sudo kill $(pgrep -f "python monitor.py")
 187     cat monitor.log
 188 }
 189
 190 start_inspector() {
 191     pgrep -f "python inspector.py" && return 0
 192     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 193 }
 194
 195 stop_inspector() {
 196     pgrep -f "python inspector.py" || return 0
 197     kill $(pgrep -f "python inspector.py")
 198     cat inspector.log
 199 }
 200
 201 start_consumer() {
 202     pgrep -f "python consumer.py" && return 0
 203     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 204 }
 205
 206 stop_consumer() {
 207     pgrep -f "python consumer.py" || return 0
 208     kill $(pgrep -f "python consumer.py")
 209     cat consumer.log
 210 }
 211
 212 wait_for_vm_launch() {
 213     echo "waiting for vm launch..."
 214
 215     (
 216         # get VM state as test user
 217         change_to_doctor_user
 218
 219         count=0
 220         while [[ ${count} -lt 60 ]]
 221         do
 222             state=$(openstack server list | grep " $VM_NAME " | awk '{print $6}')
 223             [[ "$state" == "ACTIVE" ]] && return 0
 224             [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 225             count=$(($count+1))
 226             sleep 1
 227         done
 228         echo "ERROR: time out while waiting for vm launch"
 229         exit 1
 230     )
 231 }
 232
 233 inject_failure() {
 234     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 235     cat > disable_network.sh << 'END_TXT'
 236 #!/bin/bash -x
 237 dev=$(sudo ip route | awk '/^default/{print $5}')
 238 sleep 1
 239 sudo ip link set $dev down
 240 sleep 180
 241 sudo ip link set $dev up
 242 sleep 1
 243 END_TXT
 244     chmod +x disable_network.sh
 245     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 246     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 247 }
 248
 249 calculate_notification_time() {
 250     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 251     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 252     echo "$notified $detected" | \
 253         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 254 }
 255
 256 check_host_status() {
 257     expect_state=$1
 258     (
 259         change_to_doctor_user
 260
 261         host_status_line=$(openstack --os-compute-api-version 2.16 server show $VM_NAME | grep "host_status")
 262         if [[ $? -ne 0 ]] ; then
 263             echo "ERROR: host_status not configured for owner in Nova policy.json"
 264             exit 1
 265         fi
 266
 267         host_status=$(echo $host_status_line | awk '{print $4}')
 268         if [ -z "$host_status" ] ; then
 269             echo "ERROR: host_status not reported by: nova show $VM_NAME"
 270             exit 1
 271         elif [[ "$host_status" != "$expect_state" ]] ; then
 272             echo "ERROR: host_status:$host_status not equal to expect_state: $expect_state"
 273             exit 1
 274         else
 275             echo "$VM_NAME showing host_status: $host_status"
 276         fi
 277     )
 278 }
 279
 280 cleanup() {
 281     set +e
 282     echo "cleanup..."
 283     stop_monitor
 284     stop_inspector
 285     stop_consumer
 286
 287     echo "waiting disabled compute host back to be enabled..."
 288     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 289     sleep 180
 290     check_host_status "UP"
 291     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 292         "[ -e disable_network.log ] && cat disable_network.log"
 293
 294     (
 295         change_to_doctor_user
 296         openstack server list | grep -q " $VM_NAME " && openstack server delete "$VM_NAME"
 297         sleep 1
 298         alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 299         sleep 1
 300         [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 301         sleep 1
 302     )
 303     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 304     sleep 1
 305     [ -n "$image_id" ] && openstack image delete "$image_id"
 306     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 307                               --project "$DOCTOR_PROJECT"
 308     openstack project delete "$DOCTOR_PROJECT"
 309     openstack user delete "$DOCTOR_USER"
 310 }
 311
 312
 313 echo "Note: doctor/tests/run.sh has been executed."
 314
 315 trap cleanup EXIT
 316
 317 echo "preparing VM image..."
 318 download_image
 319 register_image
 320
 321 echo "creating test user..."
 322 create_test_user
 323
 324 echo "creating VM..."
 325 boot_vm
 326 wait_for_vm_launch
 327
 328 echo "get computer host info and prepare to ssh..."
 329 get_compute_host_info
 330 prepare_compute_ssh
 331
 332 echo "creating alarm..."
 333 get_consumer_ip
 334 create_alarm
 335
 336 echo "starting doctor sample components..."
 337 start_monitor
 338 start_inspector
 339 start_consumer
 340
 341 sleep 60
 342 echo "injecting host failure..."
 343 inject_failure
 344 sleep 10
 345
 346 check_host_status "DOWN"
 347 calculate_notification_time
 348
 349 echo "done"