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 TEST_USER=demo
  23 TEST_PW=demo
  24 TEST_PROJECT=demo
  25 TEST_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 "$TEST_USER" || {
  93         openstack user create "$TEST_USER" --password "$TEST_PW"
  94     }
  95     openstack project list | grep -q "$TEST_PROJECT" || {
  96         openstack project create "$TEST_PROJECT"
  97     }
  98     openstack user role list "$TEST_USER" --project "$TEST_PROJECT" \
  99     | grep -q "$TEST_ROLE" || {
 100         openstack role add "$TEST_ROLE" --user "$TEST_USER" \
 101                            --project "$TEST_PROJECT"
 102     }
 103 }
 104
 105 boot_vm() {
 106     nova list | grep -q " $VM_NAME " && return 0
 107     (
 108         # test VM done with test user, so can test non-admin
 109         export OS_USERNAME="$TEST_USER"
 110         export OS_PASSWORD="$TEST_PW"
 111         export OS_TENANT_NAME="$TEST_PROJECT"
 112         nova boot --flavor "$VM_FLAVOR" \
 113                   --image "$IMAGE_NAME" \
 114                   "$VM_NAME"
 115         sleep 1
 116     )
 117
 118 }
 119
 120 create_alarm() {
 121     ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 122     vm_id=$(nova list | grep " $VM_NAME " | awk '{print $2}')
 123     ceilometer alarm-event-create --name "$ALARM_NAME" \
 124         --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
 125         --description "VM failure" \
 126         --enabled True \
 127         --repeat-actions False \
 128         --severity "moderate" \
 129         --event-type compute.instance.update \
 130         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 131 }
 132
 133 start_monitor() {
 134     pgrep -f "python monitor.py" && return 0
 135     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 136         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 137 }
 138
 139 stop_monitor() {
 140     pgrep -f "python monitor.py" || return 0
 141     sudo kill $(pgrep -f "python monitor.py")
 142     cat monitor.log
 143 }
 144
 145 start_inspector() {
 146     pgrep -f "python inspector.py" && return 0
 147     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 148 }
 149
 150 stop_inspector() {
 151     pgrep -f "python inspector.py" || return 0
 152     kill $(pgrep -f "python inspector.py")
 153     cat inspector.log
 154 }
 155
 156 start_consumer() {
 157     pgrep -f "python consumer.py" && return 0
 158     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 159 }
 160
 161 stop_consumer() {
 162     pgrep -f "python consumer.py" || return 0
 163     kill $(pgrep -f "python consumer.py")
 164     cat consumer.log
 165 }
 166
 167 wait_for_vm_launch() {
 168     echo "waiting for vm launch..."
 169     while true
 170     do
 171         state=$(nova list | grep " $VM_NAME " | awk '{print $6}')
 172         [[ "$state" == "ACTIVE" ]] && return 0
 173         sleep 1
 174     done
 175 }
 176
 177 inject_failure() {
 178     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 179     cat > disable_network.sh << 'END_TXT'
 180 #!/bin/bash -x
 181 dev=$(sudo ip route | awk '/^default/{print $5}')
 182 sleep 1
 183 sudo ip link set $dev down
 184 sleep 180
 185 sudo ip link set $dev up
 186 sleep 1
 187 END_TXT
 188     chmod +x disable_network.sh
 189     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 190     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 191 }
 192
 193 calculate_notification_time() {
 194     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 195     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 196     echo "$notified $detected" | \
 197         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 198 }
 199
 200 check_host_status_down() {
 201     (
 202         # Switching to test user
 203         export OS_USERNAME="$TEST_USER"
 204         export OS_PASSWORD="$TEST_PW"
 205         export OS_TENANT_NAME="$TEST_PROJECT"
 206
 207         host_status_line=$(nova show $VM_NAME | grep "host_status")
 208         [[ $? -ne 0 ]] && {
 209             echo "ERROR: host_status not configured for owner in Nova policy.json"
 210         }
 211
 212         host_status=$(echo $host_status_line | awk '{print $4}')
 213         [[ "$host_status" == "DOWN" ]] && {
 214             echo "$VM_NAME showing host_status: $host_status"
 215         }
 216         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 217     )
 218 }
 219
 220 cleanup() {
 221     set +e
 222     echo "cleanup..."
 223     stop_monitor
 224     stop_inspector
 225     stop_consumer
 226
 227     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 228     sleep 1
 229     nova list | grep -q " $VM_NAME " && nova delete "$VM_NAME"
 230     sleep 1
 231     alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 232     sleep 1
 233     [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 234     sleep 1
 235     image_id=$(glance image-list | grep " $IMAGE_NAME " | awk '{print $2}')
 236     sleep 1
 237     [ -n "$image_id" ] && glance image-delete "$image_id"
 238     openstack role remove "$TEST_ROLE" --user "$TEST_USER" \
 239                               --project "$TEST_PROJECT"
 240     openstack project delete "$TEST_PROJECT"
 241     openstack user delete "$TEST_USER"
 242
 243     #TODO: add host status check via nova admin api
 244     echo "waiting disabled compute host back to be enabled..."
 245     sleep 180
 246     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 247         "[ -e disable_network.log ] && cat disable_network.log"
 248 }
 249
 250
 251 echo "Note: doctor/tests/run.sh has been executed."
 252
 253 prepare_compute_ssh
 254
 255 trap cleanup EXIT
 256
 257 echo "preparing VM image..."
 258 download_image
 259 register_image
 260
 261 echo "starting doctor sample components..."
 262 start_monitor
 263 start_inspector
 264 start_consumer
 265
 266 echo "creating test user..."
 267 create_test_user
 268
 269 echo "creating VM and alarm..."
 270 boot_vm
 271 create_alarm
 272 wait_for_vm_launch
 273
 274 sleep 60
 275 echo "injecting host failure..."
 276 inject_failure
 277 sleep 10
 278
 279 check_host_status_down
 280 calculate_notification_time
 281
 282 echo "done"