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