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