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 #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 $4 }')
  45     if [[ -z "$COMPUTE_HOST" ]] ; then
  46         echo "ERROR: failed to get compute hostname"
  47         exit 1
  48     fi
  49     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  50         compute_host_in_undercloud=${COMPUTE_HOST%%.*}
  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 $compute_host_in_undercloud \
  59              | awk '/ ctlplane network /{print \$5}'")
  60     elif [[ "$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 $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  65             exit 1
  66         fi
  67     fi
  68     echo "COMPUTE_HOST=$COMPUTE_HOST"
  69     echo "COMPUTE_IP=$COMPUTE_IP"
  70
  71     # verify connectivity to target compute host
  72     ping -c 1 "$COMPUTE_IP"
  73     if [[ $? -ne 0 ]] ; then
  74         echo "ERROR: can not ping to computer host"
  75         exit 1
  76     fi
  77 }
  78
  79 prepare_compute_ssh() {
  80     ssh_opts_cpu="$ssh_opts"
  81
  82     # get ssh key from installer node
  83     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  84         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  85         sudo chown $(whoami):$(whoami) instack_key
  86         chmod 400 instack_key
  87         ssh_opts_cpu+=" -i instack_key"
  88     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  89         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  90     fi
  91
  92     # verify ssh to target compute host
  93     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
  94     if [[ $? -ne 0 ]] ; then
  95         echo "ERROR: can not ssh to computer host"
  96         exit 1
  97     fi
  98 }
  99
 100 get_consumer_ip() {
 101     CONSUMER_IP=$(ip route get $COMPUTE_IP | awk '/ src /{print $NF}')
 102     echo "CONSUMER_IP=$CONSUMER_IP"
 103
 104     if [[ -z "$CONSUMER_IP" ]]; then
 105         echo "ERROR: Could not get CONSUMER_IP."
 106         exit 1
 107     fi
 108 }
 109
 110 download_image() {
 111     [ -e "$IMAGE_FILE" ] && return 0
 112     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 113 }
 114
 115 register_image() {
 116     openstack image list | grep -q " $IMAGE_NAME " && return 0
 117     openstack image create "$IMAGE_NAME" \
 118                            --public \
 119                            --disk-format "$IMAGE_FORMAT" \
 120                            --container-format bare \
 121                            --file "$IMAGE_FILE"
 122 }
 123
 124 create_test_user() {
 125     openstack user list | grep -q " $DOCTOR_USER " || {
 126         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW"
 127     }
 128     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 129         openstack project create "$DOCTOR_PROJECT"
 130     }
 131     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 132     | grep -q " $DOCTOR_ROLE " || {
 133         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 134                            --project "$DOCTOR_PROJECT"
 135     }
 136 }
 137
 138 boot_vm() {
 139     # test VM done with test user, so can test non-admin
 140     openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
 141     openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
 142                             --image "$IMAGE_NAME" \
 143                             "$VM_NAME"
 144     sleep 1
 145 }
 146
 147 create_alarm() {
 148     # get vm_id as test user
 149     ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
 150     vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
 151     ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
 152         --alarm-action "http://$CONSUMER_IP:$CONSUMER_PORT/failure" \
 153         --description "VM failure" \
 154         --enabled True \
 155         --repeat-actions False \
 156         --severity "moderate" \
 157         --event-type compute.instance.update \
 158         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 159 }
 160
 161
 162 start_monitor() {
 163     pgrep -f "python monitor.py" && return 0
 164     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 165         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 166 }
 167
 168 stop_monitor() {
 169     pgrep -f "python monitor.py" || return 0
 170     sudo kill $(pgrep -f "python monitor.py")
 171     cat monitor.log
 172 }
 173
 174 start_inspector() {
 175     pgrep -f "python inspector.py" && return 0
 176     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 177 }
 178
 179 stop_inspector() {
 180     pgrep -f "python inspector.py" || return 0
 181     kill $(pgrep -f "python inspector.py")
 182     cat inspector.log
 183 }
 184
 185 start_consumer() {
 186     pgrep -f "python consumer.py" && return 0
 187     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 188 }
 189
 190 stop_consumer() {
 191     pgrep -f "python consumer.py" || return 0
 192     kill $(pgrep -f "python consumer.py")
 193     cat consumer.log
 194 }
 195
 196 wait_for_vm_launch() {
 197     echo "waiting for vm launch..."
 198
 199     count=0
 200     while [[ ${count} -lt 60 ]]
 201     do
 202         state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
 203         [[ "$state" == "ACTIVE" ]] && return 0
 204         [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 205         count=$(($count+1))
 206         sleep 1
 207     done
 208     echo "ERROR: time out while waiting for vm launch"
 209     exit 1
 210 }
 211
 212 inject_failure() {
 213     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 214     cat > disable_network.sh << 'END_TXT'
 215 #!/bin/bash -x
 216 dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
 217 sleep 1
 218 sudo ip link set $dev down
 219 sleep 180
 220 sudo ip link set $dev up
 221 sleep 1
 222 END_TXT
 223     sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
 224     chmod +x disable_network.sh
 225     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 226     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 227 }
 228
 229 calculate_notification_time() {
 230     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 231     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 232     if ! grep -q "doctor consumer notified at" consumer.log ; then
 233         echo "ERROR: consumer hasn't received fault notification."
 234         exit 1
 235     fi
 236     echo "$notified $detected" | \
 237         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 238 }
 239
 240 check_host_status() {
 241     expected_state=$1
 242
 243     host_status_line=$(openstack $as_doctor_user --os-compute-api-version 2.16 \
 244                        server show $VM_NAME | grep "host_status")
 245     host_status=$(echo $host_status_line | awk '{print $4}')
 246     if [ -z "$host_status" ] ; then
 247         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 248         exit 1
 249     elif [[ "$expected_state" =~ "$host_status" ]] ; then
 250         echo "$VM_NAME showing host_status: $host_status"
 251     else
 252         echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
 253         exit 1
 254     fi
 255 }
 256
 257 cleanup() {
 258     set +e
 259     echo "cleanup..."
 260     stop_monitor
 261     stop_inspector
 262     stop_consumer
 263
 264     echo "waiting disabled compute host back to be enabled..."
 265     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 266     sleep 240
 267     check_host_status "UP"
 268     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 269         "[ -e disable_network.log ] && cat disable_network.log"
 270
 271     openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
 272     sleep 1
 273     alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 274     sleep 1
 275     [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
 276     sleep 1
 277
 278     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 279     sleep 1
 280     [ -n "$image_id" ] && openstack image delete "$image_id"
 281     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 282                               --project "$DOCTOR_PROJECT"
 283     openstack project delete "$DOCTOR_PROJECT"
 284     openstack user delete "$DOCTOR_USER"
 285 }
 286
 287
 288 echo "Note: doctor/tests/run.sh has been executed."
 289
 290 trap cleanup EXIT
 291
 292 echo "preparing VM image..."
 293 download_image
 294 register_image
 295
 296 echo "creating test user..."
 297 create_test_user
 298
 299 echo "creating VM..."
 300 boot_vm
 301 wait_for_vm_launch
 302 openstack $as_doctor_user server show $VM_NAME
 303
 304 echo "get computer host info and prepare to ssh..."
 305 get_compute_host_info
 306 prepare_compute_ssh
 307
 308 echo "creating alarm..."
 309 get_consumer_ip
 310 create_alarm
 311
 312 echo "starting doctor sample components..."
 313 start_monitor
 314 start_inspector
 315 start_consumer
 316
 317 sleep 60
 318 echo "injecting host failure..."
 319 inject_failure
 320 sleep 60
 321
 322 check_host_status "(DOWN|UNKNOWN)"
 323 calculate_notification_time
 324
 325 echo "done"