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
  34 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  35     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  36     exit 1
  37 fi
  38
  39 get_compute_host_info() {
  40     (
  41         change_to_doctor_user
  42
  43         # get computer host info which VM boot in
  44         export COMPUTE_HOST=$(openstack server show $VM_NAME | \
  45                 grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }' |
  46                 awk -F '.' '{print $1}')
  47     )
  48     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  49         COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  50         if [[ "$INSTALLER_IP" == "none" ]] ; then
  51             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  52             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  53         fi
  54         COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  55              "source stackrc; \
  56              openstack server show $COMPUTE_HOST \
  57              | awk '/ ctlplane network /{print \$5}'")
  58     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  59         COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
  60         COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  61         if [[ -z "$COMPUTE_IP" ]]; then
  62             echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  63             exit 1
  64         fi
  65     fi
  66
  67     # verify connectivity to target compute host
  68     ping -c 1 "$COMPUTE_IP"
  69     if [[ $? -ne 0 ]] ; then
  70         echo "ERROR: can not ping to computer host"
  71         exit 1
  72     fi
  73 }
  74
  75 prepare_compute_ssh() {
  76     ssh_opts_cpu="$ssh_opts"
  77
  78     # get ssh key from installer node
  79     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  80         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  81     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  82         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  83     fi
  84
  85     sudo chown $(whoami):$(whoami) instack_key
  86     chmod 400 instack_key
  87     ssh_opts_cpu+=" -i instack_key"
  88
  89     # verify ssh to target compute host
  90     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
  91     if [[ $? -ne 0 ]] ; then
  92         echo "ERROR: can not ssh to computer host"
  93         exit 1
  94     fi
  95 }
  96
  97 get_consumer_ip() {
  98     #get network of the compute_ip address
  99     network_pre=${COMPUTE_IP%.*}
 100     network=${network_pre}.0/24
 101
 102     #if there is a dedicated route, use it
 103     dev=$(ip route |grep $network | awk '{print $3}' |head -n 1)
 104
 105     #if there is no route, use default route
 106     [[ -z $dev ]] && dev=$(ip route |grep ^default | awk '{print $3}' |head -n 1)
 107
 108     CONSUMER_IP=$(ip addr show $dev |grep inet | grep $network_pre | awk '{print $2}' | cut -d'/' -f1)
 109 }
 110
 111 download_image() {
 112     [ -e "$IMAGE_FILE" ] && return 0
 113     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 114 }
 115
 116 register_image() {
 117     openstack image list | grep -q " $IMAGE_NAME " && return 0
 118     openstack image create "$IMAGE_NAME" \
 119                            --public \
 120                            --disk-format "$IMAGE_FORMAT" \
 121                            --container-format bare \
 122                            --file "$IMAGE_FILE"
 123 }
 124
 125 create_test_user() {
 126     openstack user list | grep -q " $DOCTOR_USER " || {
 127         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 128     }
 129     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 130         openstack project create "$DOCTOR_PROJECT"
 131     }
 132     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 133     | grep -q " $DOCTOR_ROLE " || {
 134         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 135                            --project "$DOCTOR_PROJECT"
 136     }
 137 }
 138
 139 change_to_doctor_user() {
 140     export OS_USERNAME="$DOCTOR_USER"
 141     export OS_PASSWORD="$DOCTOR_PW"
 142     export OS_PROJECT_NAME="$DOCTOR_PROJECT"
 143     export OS_TENANT_NAME="$DOCTOR_PROJECT"
 144 }
 145
 146 boot_vm() {
 147     (
 148         # test VM done with test user, so can test non-admin
 149         change_to_doctor_user
 150         openstack server list | grep -q " $VM_NAME " && return 0
 151         openstack server create --flavor "$VM_FLAVOR" \
 152                                 --image "$IMAGE_NAME" \
 153                                 "$VM_NAME"
 154         sleep 1
 155     )
 156
 157 }
 158
 159 create_alarm() {
 160     (
 161         # get vm_id as test user
 162         change_to_doctor_user
 163         ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 164         vm_id=$(openstack server list | grep " $VM_NAME " | awk '{print $2}')
 165         ceilometer alarm-event-create --name "$ALARM_NAME" \
 166             --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 167             --description "VM failure" \
 168             --enabled True \
 169             --repeat-actions False \
 170             --severity "moderate" \
 171             --event-type compute.instance.update \
 172             -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 173     )
 174 }
 175
 176
 177 start_monitor() {
 178     pgrep -f "python monitor.py" && return 0
 179     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 180         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 181 }
 182
 183 stop_monitor() {
 184     pgrep -f "python monitor.py" || return 0
 185     sudo kill $(pgrep -f "python monitor.py")
 186     cat monitor.log
 187 }
 188
 189 start_inspector() {
 190     pgrep -f "python inspector.py" && return 0
 191     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 192 }
 193
 194 stop_inspector() {
 195     pgrep -f "python inspector.py" || return 0
 196     kill $(pgrep -f "python inspector.py")
 197     cat inspector.log
 198 }
 199
 200 start_consumer() {
 201     pgrep -f "python consumer.py" && return 0
 202     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 203 }
 204
 205 stop_consumer() {
 206     pgrep -f "python consumer.py" || return 0
 207     kill $(pgrep -f "python consumer.py")
 208     cat consumer.log
 209 }
 210
 211 wait_for_vm_launch() {
 212     echo "waiting for vm launch..."
 213
 214     (
 215         # get VM state as test user
 216         change_to_doctor_user
 217
 218         count=0
 219         while [[ ${count} -lt 60 ]]
 220         do
 221             state=$(openstack server list | grep " $VM_NAME " | awk '{print $6}')
 222             [[ "$state" == "ACTIVE" ]] && return 0
 223             [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 224             count=$(($count+1))
 225             sleep 1
 226         done
 227         echo "ERROR: time out while waiting for vm launch"
 228         exit 1
 229     )
 230 }
 231
 232 inject_failure() {
 233     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 234     cat > disable_network.sh << 'END_TXT'
 235 #!/bin/bash -x
 236 dev=$(sudo ip route | awk '/^default/{print $5}')
 237 sleep 1
 238 sudo ip link set $dev down
 239 sleep 180
 240 sudo ip link set $dev up
 241 sleep 1
 242 END_TXT
 243     chmod +x disable_network.sh
 244     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 245     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 246 }
 247
 248 calculate_notification_time() {
 249     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 250     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 251     echo "$notified $detected" | \
 252         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 253 }
 254
 255 check_host_status() {
 256     expect_state=$1
 257     (
 258         change_to_doctor_user
 259
 260         host_status_line=$(openstack --os-compute-api-version 2.16 server show $VM_NAME | grep "host_status")
 261         if [[ $? -ne 0 ]] ; then
 262             echo "ERROR: host_status not configured for owner in Nova policy.json"
 263             exit 1
 264         fi
 265
 266         host_status=$(echo $host_status_line | awk '{print $4}')
 267         if [ -z "$host_status" ] ; then
 268             echo "ERROR: host_status not reported by: nova show $VM_NAME"
 269             exit 1
 270         elif [[ "$host_status" != "$expect_state" ]] ; then
 271             echo "ERROR: host_status:$host_status not equal to expect_state: $expect_state"
 272             exit 1
 273         else
 274             echo "$VM_NAME showing host_status: $host_status"
 275         fi
 276     )
 277 }
 278
 279 cleanup() {
 280     set +e
 281     echo "cleanup..."
 282     stop_monitor
 283     stop_inspector
 284     stop_consumer
 285
 286     echo "waiting disabled compute host back to be enabled..."
 287     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 288     sleep 180
 289     check_host_status "UP"
 290     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 291         "[ -e disable_network.log ] && cat disable_network.log"
 292
 293     (
 294         change_to_doctor_user
 295         openstack server list | grep -q " $VM_NAME " && openstack server delete "$VM_NAME"
 296         sleep 1
 297         alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 298         sleep 1
 299         [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 300         sleep 1
 301     )
 302     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 303     sleep 1
 304     [ -n "$image_id" ] && openstack image delete "$image_id"
 305     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 306                               --project "$DOCTOR_PROJECT"
 307     openstack project delete "$DOCTOR_PROJECT"
 308     openstack user delete "$DOCTOR_USER"
 309 }
 310
 311
 312 echo "Note: doctor/tests/run.sh has been executed."
 313
 314 trap cleanup EXIT
 315
 316 echo "preparing VM image..."
 317 download_image
 318 register_image
 319
 320 echo "creating test user..."
 321 create_test_user
 322
 323 echo "creating VM..."
 324 boot_vm
 325 wait_for_vm_launch
 326
 327 echo "get computer host info and prepare to ssh..."
 328 get_compute_host_info
 329 prepare_compute_ssh
 330
 331 echo "creating alarm..."
 332 get_consumer_ip
 333 create_alarm
 334
 335 echo "starting doctor sample components..."
 336 start_monitor
 337 start_inspector
 338 start_consumer
 339
 340 sleep 60
 341 echo "injecting host failure..."
 342 inject_failure
 343 sleep 10
 344
 345 check_host_status "DOWN"
 346 calculate_notification_time
 347
 348 echo "done"