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 [[ "${CI_DEBUG:-true}" == [Tt]rue ]] && set -x
  12
  13 IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-disk.img
  14 IMAGE_NAME=cirros
  15 IMAGE_FILE="${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
  31 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  32 as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
  33                 --os-tenant-name $DOCTOR_PROJECT"
  34
  35 if [[ ! "$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     # get computer host info which VM boot in
  42     COMPUTE_HOST=$(openstack $as_doctor_user server show $VM_NAME |
  43                    grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }')
  44     if [[ -z "$COMPUTE_HOST" ]] ; then
  45         echo "ERROR: failed to get compute hostname"
  46         exit 1
  47     fi
  48     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  49         compute_host_in_undercloud=${COMPUTE_HOST%%.*}
  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              nova show $compute_host_in_undercloud \
  58              | awk '/ ctlplane network /{print \$5}'")
  59     elif [[ "$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 $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  64             exit 1
  65         fi
  66     fi
  67     echo "COMPUTE_HOST=$COMPUTE_HOST"
  68     echo "COMPUTE_IP=$COMPUTE_IP"
  69
  70     # verify connectivity to target compute host
  71     ping -c 1 "$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         sudo chown $(whoami):$(whoami) instack_key
  85         chmod 400 instack_key
  86         ssh_opts_cpu+=" -i instack_key"
  87     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  88         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  89     fi
  90
  91     # verify ssh to target compute host
  92     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
  93     if [[ $? -ne 0 ]] ; then
  94         echo "ERROR: can not ssh to computer host"
  95         exit 1
  96     fi
  97 }
  98
  99 get_consumer_ip() {
 100     CONSUMER_IP=$(ip route get $COMPUTE_IP | awk '/ src /{print $NF}')
 101     echo "CONSUMER_IP=$CONSUMER_IP"
 102
 103     if [[ -z "$CONSUMER_IP" ]]; then
 104         echo "ERROR: Could not get CONSUMER_IP."
 105         exit 1
 106     fi
 107 }
 108
 109 download_image() {
 110     [ -e "$IMAGE_FILE" ] && return 0
 111     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 112 }
 113
 114 register_image() {
 115     openstack image list | grep -q " $IMAGE_NAME " && return 0
 116     openstack image create "$IMAGE_NAME" \
 117                            --public \
 118                            --disk-format "$IMAGE_FORMAT" \
 119                            --container-format bare \
 120                            --file "$IMAGE_FILE"
 121 }
 122
 123 create_test_user() {
 124     openstack user list | grep -q " $DOCTOR_USER " || {
 125         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 126     }
 127     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 128         openstack project create "$DOCTOR_PROJECT"
 129     }
 130     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 131     | grep -q " $DOCTOR_ROLE " || {
 132         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 133                            --project "$DOCTOR_PROJECT"
 134     }
 135 }
 136
 137 boot_vm() {
 138     # test VM done with test user, so can test non-admin
 139     openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
 140     openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
 141                             --image "$IMAGE_NAME" \
 142                             "$VM_NAME"
 143     sleep 1
 144 }
 145
 146 create_alarm() {
 147     # get vm_id as test user
 148     ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
 149     vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
 150     ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
 151         --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 152         --description "VM failure" \
 153         --enabled True \
 154         --repeat-actions False \
 155         --severity "moderate" \
 156         --event-type compute.instance.update \
 157         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 158 }
 159
 160
 161 start_monitor() {
 162     pgrep -f "python monitor.py" && return 0
 163     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 164         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 165 }
 166
 167 stop_monitor() {
 168     pgrep -f "python monitor.py" || return 0
 169     sudo kill $(pgrep -f "python monitor.py")
 170     cat monitor.log
 171 }
 172
 173 start_inspector() {
 174     pgrep -f "python inspector.py" && return 0
 175     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 176 }
 177
 178 stop_inspector() {
 179     pgrep -f "python inspector.py" || return 0
 180     kill $(pgrep -f "python inspector.py")
 181     cat inspector.log
 182 }
 183
 184 start_consumer() {
 185     pgrep -f "python consumer.py" && return 0
 186     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 187 }
 188
 189 stop_consumer() {
 190     pgrep -f "python consumer.py" || return 0
 191     kill $(pgrep -f "python consumer.py")
 192     cat consumer.log
 193 }
 194
 195 wait_for_vm_launch() {
 196     echo "waiting for vm launch..."
 197
 198     count=0
 199     while [[ ${count} -lt 60 ]]
 200     do
 201         state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
 202         [[ "$state" == "ACTIVE" ]] && return 0
 203         [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 204         count=$(($count+1))
 205         sleep 1
 206     done
 207     echo "ERROR: time out while waiting for vm launch"
 208     exit 1
 209 }
 210
 211 inject_failure() {
 212     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 213     cat > disable_network.sh << 'END_TXT'
 214 #!/bin/bash -x
 215 dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
 216 sleep 1
 217 sudo ip link set $dev down
 218 sleep 180
 219 sudo ip link set $dev up
 220 sleep 1
 221 END_TXT
 222     sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
 223     chmod +x disable_network.sh
 224     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 225     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 226 }
 227
 228 calculate_notification_time() {
 229     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 230     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 231     if ! grep -q "doctor consumer notified at" consumer.log ; then
 232         echo "ERROR: consumer hasn't received fault notification."
 233         exit 1
 234     fi
 235     echo "$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 $VM_NAME"
 247         exit 1
 248     elif [[ "$expected_state" =~ "$host_status" ]] ; then
 249         echo "$VM_NAME showing host_status: $host_status"
 250     else
 251         echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
 252         exit 1
 253     fi
 254 }
 255
 256 cleanup() {
 257     set +e
 258     echo "cleanup..."
 259     stop_monitor
 260     stop_inspector
 261     stop_consumer
 262
 263     echo "waiting disabled compute host back to be enabled..."
 264     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 265     sleep 240
 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 $as_doctor_user server delete "$VM_NAME"
 271     sleep 1
 272     alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 273     sleep 1
 274     [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$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 openstack $as_doctor_user server show $VM_NAME
 302
 303 echo "get computer host info and prepare to ssh..."
 304 get_compute_host_info
 305 prepare_compute_ssh
 306
 307 echo "creating alarm..."
 308 get_consumer_ip
 309 create_alarm
 310
 311 echo "starting doctor sample components..."
 312 start_monitor
 313 start_inspector
 314 start_consumer
 315
 316 sleep 60
 317 echo "injecting host failure..."
 318 inject_failure
 319 sleep 60
 320
 321 check_host_status "(DOWN|UNKNOWN)"
 322 calculate_notification_time
 323
 324 echo "done"