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 DOCTOR_ROLE=_member_
  26
  27 SUPPORTED_INSTALLER_TYPES="apex local"
  28 INSTALLER_TYPE=${INSTALLER_TYPE:-apex}
  29 INSTALLER_IP=${INSTALLER_IP:-none}
  30 COMPUTE_HOST=${COMPUTE_HOST:-overcloud-novacompute-0}
  31 COMPUTE_IP=${COMPUTE_IP:-none}
  32 COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  33 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  34
  35 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  36     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  37     exit 1
  38 fi
  39
  40 prepare_compute_ssh() {
  41     ssh_opts_cpu="$ssh_opts"
  42
  43     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  44         if [[ "$INSTALLER_IP" == "none" ]] ; then
  45             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  46             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  47         fi
  48
  49         if [[ "$COMPUTE_IP" == "none" ]] ; then
  50             COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  51                          "source stackrc; \
  52                           nova show $COMPUTE_HOST \
  53                           | awk '/ ctlplane network /{print \$5}'")
  54         fi
  55
  56         # get ssh key from installer node
  57         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  58         sudo chown $(whoami):$(whoami) instack_key
  59         chmod 400 instack_key
  60         ssh_opts_cpu+=" -i instack_key"
  61     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  62         if [[ "$COMPUTE_IP" == "none" ]] ; then
  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         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  71     fi
  72
  73     # verify connectivity to target compute host
  74     ping -c 1 "$COMPUTE_IP"
  75 }
  76
  77 download_image() {
  78     [ -e "$IMAGE_FILE" ] && return 0
  79     wget "$IMAGE_URL" -o "$IMAGE_FILE"
  80 }
  81
  82 register_image() {
  83     glance image-list | grep -q " $IMAGE_NAME " && return 0
  84     glance image-create --name "$IMAGE_NAME" \
  85                         --visibility public \
  86                         --disk-format "$IMAGE_FORMAT" \
  87                         --container-format bare \
  88                         --file "$IMAGE_FILE"
  89 }
  90
  91 create_test_user() {
  92     openstack user list | grep -q " $DOCTOR_USER " || {
  93         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
  94     }
  95     openstack project list | grep -q " $DOCTOR_PROJECT " || {
  96         openstack project create "$DOCTOR_PROJECT"
  97     }
  98     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
  99     | grep -q " $DOCTOR_ROLE " || {
 100         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 101                            --project "$DOCTOR_PROJECT"
 102     }
 103 }
 104
 105 change_to_doctor_user() {
 106     export OS_USERNAME="$DOCTOR_USER"
 107     export OS_PASSWORD="$DOCTOR_PW"
 108     export OS_PROJECT_NAME="$DOCTOR_PROJECT"
 109     export OS_TENANT_NAME="$DOCTOR_PROJECT"
 110 }
 111
 112 boot_vm() {
 113     (
 114         # test VM done with test user, so can test non-admin
 115         change_to_doctor_user
 116         nova list | grep -q " $VM_NAME " && return 0
 117         nova boot --flavor "$VM_FLAVOR" \
 118                   --image "$IMAGE_NAME" \
 119                   "$VM_NAME"
 120         sleep 1
 121     )
 122
 123 }
 124
 125 create_alarm() {
 126     (
 127         # get vm_id as test user
 128         change_to_doctor_user
 129         ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 130         vm_id=$(nova list | grep " $VM_NAME " | awk '{print $2}')
 131         ceilometer alarm-event-create --name "$ALARM_NAME" \
 132             --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
 133             --description "VM failure" \
 134             --enabled True \
 135             --repeat-actions False \
 136             --severity "moderate" \
 137             --event-type compute.instance.update \
 138             -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 139     )
 140 }
 141
 142
 143 start_monitor() {
 144     pgrep -f "python monitor.py" && return 0
 145     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 146         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 147 }
 148
 149 stop_monitor() {
 150     pgrep -f "python monitor.py" || return 0
 151     sudo kill $(pgrep -f "python monitor.py")
 152     cat monitor.log
 153 }
 154
 155 start_inspector() {
 156     pgrep -f "python inspector.py" && return 0
 157     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 158 }
 159
 160 stop_inspector() {
 161     pgrep -f "python inspector.py" || return 0
 162     kill $(pgrep -f "python inspector.py")
 163     cat inspector.log
 164 }
 165
 166 start_consumer() {
 167     pgrep -f "python consumer.py" && return 0
 168     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 169 }
 170
 171 stop_consumer() {
 172     pgrep -f "python consumer.py" || return 0
 173     kill $(pgrep -f "python consumer.py")
 174     cat consumer.log
 175 }
 176
 177 wait_for_vm_launch() {
 178     echo "waiting for vm launch..."
 179
 180     (
 181         # get VM state as test user
 182         change_to_doctor_user
 183
 184         count=0
 185         while [[ ${count} -lt 60 ]]
 186         do
 187             state=$(nova list | grep " $VM_NAME " | awk '{print $6}')
 188             [[ "$state" == "ACTIVE" ]] && return 0
 189             [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 190             count=$(($count+1))
 191             sleep 1
 192         done
 193         echo "ERROR: time out while waiting for vm launch"
 194         exit 1
 195     )
 196 }
 197
 198 inject_failure() {
 199     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 200     cat > disable_network.sh << 'END_TXT'
 201 #!/bin/bash -x
 202 dev=$(sudo ip route | awk '/^default/{print $5}')
 203 sleep 1
 204 sudo ip link set $dev down
 205 sleep 180
 206 sudo ip link set $dev up
 207 sleep 1
 208 END_TXT
 209     chmod +x disable_network.sh
 210     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 211     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 212 }
 213
 214 calculate_notification_time() {
 215     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 216     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 217     echo "$notified $detected" | \
 218         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 219 }
 220
 221 check_host_status_down() {
 222     (
 223         change_to_doctor_user
 224
 225         host_status_line=$(nova show $VM_NAME | grep "host_status")
 226         [[ $? -ne 0 ]] && {
 227             echo "ERROR: host_status not configured for owner in Nova policy.json"
 228         }
 229
 230         host_status=$(echo $host_status_line | awk '{print $4}')
 231         [[ "$host_status" == "DOWN" ]] && {
 232             echo "$VM_NAME showing host_status: $host_status"
 233         }
 234         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 235     )
 236 }
 237
 238 cleanup() {
 239     set +e
 240     echo "cleanup..."
 241     stop_monitor
 242     stop_inspector
 243     stop_consumer
 244
 245     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 246     sleep 1
 247     (
 248         change_to_doctor_user
 249         nova list | grep -q " $VM_NAME " && nova delete "$VM_NAME"
 250         sleep 1
 251         alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 252         sleep 1
 253         [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 254         sleep 1
 255     )
 256     image_id=$(glance image-list | grep " $IMAGE_NAME " | awk '{print $2}')
 257     sleep 1
 258     [ -n "$image_id" ] && glance image-delete "$image_id"
 259     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 260                               --project "$DOCTOR_PROJECT"
 261     openstack project delete "$DOCTOR_PROJECT"
 262     openstack user delete "$DOCTOR_USER"
 263
 264     #TODO: add host status check via nova admin api
 265     echo "waiting disabled compute host back to be enabled..."
 266     sleep 180
 267     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 268         "[ -e disable_network.log ] && cat disable_network.log"
 269 }
 270
 271
 272 echo "Note: doctor/tests/run.sh has been executed."
 273
 274 prepare_compute_ssh
 275
 276 trap cleanup EXIT
 277
 278 echo "preparing VM image..."
 279 download_image
 280 register_image
 281
 282 echo "starting doctor sample components..."
 283 start_monitor
 284 start_inspector
 285 start_consumer
 286
 287 echo "creating test user..."
 288 create_test_user
 289
 290 echo "creating VM and alarm..."
 291 boot_vm
 292 create_alarm
 293 wait_for_vm_launch
 294
 295 sleep 60
 296 echo "injecting host failure..."
 297 inject_failure
 298 sleep 10
 299
 300 check_host_status_down
 301 calculate_notification_time
 302
 303 echo "done"