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 [[ $? -ne 0 ]] ; then
  93         echo "ERROR: can not ssh to computer host"
  94         exit 1
  95     fi
  96 }
  97
  98 get_consumer_ip() {
  99     CONSUMER_IP=$(ip route get $COMPUTE_IP | awk '/ src /{print $NF}')
 100
 101     if [[ -z "$CONSUMER_IP" ]]; then
 102         echo "ERROR: Could not get CONSUMER_IP."
 103         exit 1
 104     fi
 105 }
 106
 107 download_image() {
 108     [ -e "$IMAGE_FILE" ] && return 0
 109     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 110 }
 111
 112 register_image() {
 113     openstack image list | grep -q " $IMAGE_NAME " && return 0
 114     openstack image create "$IMAGE_NAME" \
 115                            --public \
 116                            --disk-format "$IMAGE_FORMAT" \
 117                            --container-format bare \
 118                            --file "$IMAGE_FILE"
 119 }
 120
 121 create_test_user() {
 122     openstack user list | grep -q " $DOCTOR_USER " || {
 123         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 124     }
 125     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 126         openstack project create "$DOCTOR_PROJECT"
 127     }
 128     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 129     | grep -q " $DOCTOR_ROLE " || {
 130         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 131                            --project "$DOCTOR_PROJECT"
 132     }
 133 }
 134
 135 boot_vm() {
 136     # test VM done with test user, so can test non-admin
 137     openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
 138     openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
 139                             --image "$IMAGE_NAME" \
 140                             "$VM_NAME"
 141     sleep 1
 142 }
 143
 144 create_alarm() {
 145     # get vm_id as test user
 146     ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
 147     vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
 148     ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
 149         --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 150         --description "VM failure" \
 151         --enabled True \
 152         --repeat-actions False \
 153         --severity "moderate" \
 154         --event-type compute.instance.update \
 155         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 156 }
 157
 158
 159 start_monitor() {
 160     pgrep -f "python monitor.py" && return 0
 161     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 162         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 163 }
 164
 165 stop_monitor() {
 166     pgrep -f "python monitor.py" || return 0
 167     sudo kill $(pgrep -f "python monitor.py")
 168     cat monitor.log
 169 }
 170
 171 start_inspector() {
 172     pgrep -f "python inspector.py" && return 0
 173     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 174 }
 175
 176 stop_inspector() {
 177     pgrep -f "python inspector.py" || return 0
 178     kill $(pgrep -f "python inspector.py")
 179     cat inspector.log
 180 }
 181
 182 start_consumer() {
 183     pgrep -f "python consumer.py" && return 0
 184     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 185 }
 186
 187 stop_consumer() {
 188     pgrep -f "python consumer.py" || return 0
 189     kill $(pgrep -f "python consumer.py")
 190     cat consumer.log
 191 }
 192
 193 wait_for_vm_launch() {
 194     echo "waiting for vm launch..."
 195
 196     count=0
 197     while [[ ${count} -lt 60 ]]
 198     do
 199         state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
 200         [[ "$state" == "ACTIVE" ]] && return 0
 201         [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 202         count=$(($count+1))
 203         sleep 1
 204     done
 205     echo "ERROR: time out while waiting for vm launch"
 206     exit 1
 207 }
 208
 209 inject_failure() {
 210     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 211     cat > disable_network.sh << 'END_TXT'
 212 #!/bin/bash -x
 213 dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
 214 sleep 1
 215 sudo ip link set $dev down
 216 sleep 180
 217 sudo ip link set $dev up
 218 sleep 1
 219 END_TXT
 220     sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
 221     chmod +x disable_network.sh
 222     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 223     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 224 }
 225
 226 calculate_notification_time() {
 227     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 228     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 229     echo "$notified $detected" | \
 230         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 231 }
 232
 233 check_host_status() {
 234     expected_state=$1
 235
 236     host_status_line=$(openstack $as_doctor_user --os-compute-api-version 2.16 \
 237                        server show $VM_NAME | grep "host_status")
 238     host_status=$(echo $host_status_line | awk '{print $4}')
 239     if [ -z "$host_status" ] ; then
 240         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 241         exit 1
 242     elif [[ "$host_status" != "$expected_state" ]] ; then
 243         echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
 244         exit 1
 245     else
 246         echo "$VM_NAME showing host_status: $host_status"
 247     fi
 248 }
 249
 250 cleanup() {
 251     set +e
 252     echo "cleanup..."
 253     stop_monitor
 254     stop_inspector
 255     stop_consumer
 256
 257     echo "waiting disabled compute host back to be enabled..."
 258     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 259     sleep 180
 260     check_host_status "UP"
 261     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 262         "[ -e disable_network.log ] && cat disable_network.log"
 263
 264     openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
 265     sleep 1
 266     alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 267     sleep 1
 268     [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
 269     sleep 1
 270
 271     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 272     sleep 1
 273     [ -n "$image_id" ] && openstack image delete "$image_id"
 274     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 275                               --project "$DOCTOR_PROJECT"
 276     openstack project delete "$DOCTOR_PROJECT"
 277     openstack user delete "$DOCTOR_USER"
 278 }
 279
 280
 281 echo "Note: doctor/tests/run.sh has been executed."
 282
 283 trap cleanup EXIT
 284
 285 echo "preparing VM image..."
 286 download_image
 287 register_image
 288
 289 echo "creating test user..."
 290 create_test_user
 291
 292 echo "creating VM..."
 293 boot_vm
 294 wait_for_vm_launch
 295
 296 echo "get computer host info and prepare to ssh..."
 297 get_compute_host_info
 298 prepare_compute_ssh
 299
 300 echo "creating alarm..."
 301 get_consumer_ip
 302 create_alarm
 303
 304 echo "starting doctor sample components..."
 305 start_monitor
 306 start_inspector
 307 start_consumer
 308
 309 sleep 60
 310 echo "injecting host failure..."
 311 inject_failure
 312 sleep 10
 313
 314 check_host_status "DOWN"
 315 calculate_notification_time
 316
 317 echo "done"