tests/run.sh

   1 #!/bin/bash -ex
   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 IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-disk.img
  12 IMAGE_NAME=cirros
  13 IMAGE_FILE="${IMAGE_NAME}.img"
  14 IMAGE_FORMAT=qcow2
  15 VM_NAME=doctor_vm1
  16 VM_FLAVOR=m1.tiny
  17 ALARM_NAME=doctor_alarm1
  18 INSPECTOR_PORT=12345
  19 CONSUMER_PORT=12346
  20 TEST_USER=demo
  21 TEST_PW=demo
  22 TEST_TENANT=demo
  23 TEST_ROLE=_member_
  24
  25 SUPPORTED_INSTALLER_TYPES="apex local"
  26 INSTALLER_TYPE=${INSTALLER_TYPE:-apex}
  27 INSTALLER_IP=${INSTALLER_IP:-none}
  28 COMPUTE_HOST=${COMPUTE_HOST:-overcloud-novacompute-0}
  29 COMPUTE_IP=${COMPUTE_IP:-none}
  30 COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  31 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  32
  33 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  34     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  35     exit 1
  36 fi
  37
  38 prepare_compute_ssh() {
  39     ssh_opts_cpu="$ssh_opts"
  40
  41     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  42         if [[ "$INSTALLER_IP" == "none" ]] ; then
  43             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  44             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  45         fi
  46
  47         if [[ "$COMPUTE_IP" == "none" ]] ; then
  48             COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  49                          "source stackrc; \
  50                           nova show $COMPUTE_HOST \
  51                           | awk '/ ctlplane network /{print \$5}'")
  52         fi
  53
  54         # get ssh key from installer node
  55         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
  56         sudo chown $(whoami):$(whoami) instack_key
  57         chmod 400 instack_key
  58         ssh_opts_cpu+=" -i instack_key"
  59     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  60         if [[ "$COMPUTE_IP" == "none" ]] ; then
  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
  68         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
  69     fi
  70
  71     # verify connectivity to target compute host
  72     ping -c 1 "$COMPUTE_IP"
  73 }
  74
  75 download_image() {
  76     [ -e "$IMAGE_FILE" ] && return 0
  77     wget "$IMAGE_URL" -o "$IMAGE_FILE"
  78 }
  79
  80 register_image() {
  81     glance image-list | grep -q " $IMAGE_NAME " && return 0
  82     glance image-create --name "$IMAGE_NAME" \
  83                         --visibility public \
  84                         --disk-format "$IMAGE_FORMAT" \
  85                         --container-format bare \
  86                         --file "$IMAGE_FILE"
  87 }
  88
  89 create_test_user() {
  90     keystone user-list | grep -q "$TEST_USER" || {
  91         keystone user-create --name "$TEST_USER" --pass "$TEST_PW"
  92     }
  93     keystone tenant-list | grep -q "$TEST_TENANT" || {
  94         keystone tenant-create --name "$TEST_TENANT"
  95     }
  96     keystone user-role-list --user "$TEST_USER" --tenant "$TEST_TENANT" \
  97     | grep -q "$TEST_ROLE" || {
  98         keystone user-role-add --user "$TEST_USER" --role "$TEST_ROLE" \
  99                            --tenant "$TEST_TENANT"
 100     }
 101 }
 102
 103 boot_vm() {
 104     nova list | grep -q " $VM_NAME " && return 0
 105     (
 106         # test VM done with test user, so can test non-admin
 107         export OS_USERNAME="$TEST_USER"
 108         export OS_PASSWORD="$TEST_PW"
 109         export OS_TENANT_NAME="$TEST_TENANT"
 110         nova boot --flavor "$VM_FLAVOR" \
 111                   --image "$IMAGE_NAME" \
 112                   "$VM_NAME"
 113         sleep 1
 114     )
 115
 116 }
 117
 118 create_alarm() {
 119     ceilometer alarm-list | grep -q " $ALARM_NAME " && return 0
 120     vm_id=$(nova list | grep " $VM_NAME " | awk '{print $2}')
 121     ceilometer alarm-event-create --name "$ALARM_NAME" \
 122         --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
 123         --description "VM failure" \
 124         --enabled True \
 125         --repeat-actions False \
 126         --severity "moderate" \
 127         --event-type compute.instance.update \
 128         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 129 }
 130
 131 start_monitor() {
 132     pgrep -f "python monitor.py" && return 0
 133     sudo python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" \
 134         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 135 }
 136
 137 stop_monitor() {
 138     pgrep -f "python monitor.py" || return 0
 139     sudo kill $(pgrep -f "python monitor.py")
 140     cat monitor.log
 141 }
 142
 143 start_inspector() {
 144     pgrep -f "python inspector.py" && return 0
 145     python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 146 }
 147
 148 stop_inspector() {
 149     pgrep -f "python inspector.py" || return 0
 150     kill $(pgrep -f "python inspector.py")
 151     cat inspector.log
 152 }
 153
 154 start_consumer() {
 155     pgrep -f "python consumer.py" && return 0
 156     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 157 }
 158
 159 stop_consumer() {
 160     pgrep -f "python consumer.py" || return 0
 161     kill $(pgrep -f "python consumer.py")
 162     cat consumer.log
 163 }
 164
 165 wait_for_vm_launch() {
 166     echo "waiting for vm launch..."
 167     while true
 168     do
 169         state=$(nova list | grep " $VM_NAME " | awk '{print $6}')
 170         [[ "$state" == "ACTIVE" ]] && return 0
 171         sleep 1
 172     done
 173 }
 174
 175 inject_failure() {
 176     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 177     cat > disable_network.sh << 'END_TXT'
 178 #!/bin/bash -x
 179 dev=$(sudo ip route | awk '/^default/{print $5}')
 180 sleep 1
 181 sudo ip link set $dev down
 182 sleep 180
 183 sudo ip link set $dev up
 184 sleep 1
 185 END_TXT
 186     chmod +x disable_network.sh
 187     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 188     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 189 }
 190
 191 calculate_notification_time() {
 192     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 193     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 194     echo "$notified $detected" | \
 195         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 196 }
 197
 198 check_host_status_down() {
 199     (
 200         # Switching to test user
 201         export OS_USERNAME="$TEST_USER"
 202         export OS_PASSWORD="$TEST_PW"
 203         export OS_TENANT_NAME="$TEST_TENANT"
 204
 205         host_status_line=$(nova show $VM_NAME | grep "host_status")
 206         [[ $? -ne 0 ]] && {
 207             echo "ERROR: host_status not configured for owner in Nova policy.json"
 208         }
 209
 210         host_status=$(echo $host_status_line | awk '{print $4}')
 211         [[ "$host_status" == "DOWN" ]] && {
 212             echo "$VM_NAME showing host_status: $host_status"
 213         }
 214         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 215     )
 216 }
 217
 218 cleanup() {
 219     set +e
 220     echo "cleanup..."
 221     stop_monitor
 222     stop_inspector
 223     stop_consumer
 224
 225     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 226     sleep 1
 227     nova delete "$VM_NAME"
 228     sleep 1
 229     alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 230     sleep 1
 231     [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
 232     sleep 1
 233     image_id=$(glance image-list | grep " $IMAGE_NAME " | awk '{print $2}')
 234     sleep 1
 235     [ -n "$image_id" ] && glance image-delete "$image_id"
 236     keystone user-role-remove --user "$TEST_USER" --role "$TEST_ROLE" \
 237                               --tenant "$TEST_TENANT"
 238     keystone tenant-remove --name "$TEST_TENANT"
 239     keystone user-delete "$TEST_USER"
 240
 241     #TODO: add host status check via nova admin api
 242     echo "waiting disabled compute host back to be enabled..."
 243     sleep 180
 244     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
 245         "[ -e disable_network.log ] && cat disable_network.log"
 246 }
 247
 248
 249 echo "Note: doctor/tests/run.sh has been executed."
 250
 251 prepare_compute_ssh
 252
 253 trap cleanup EXIT
 254
 255 echo "preparing VM image..."
 256 download_image
 257 register_image
 258
 259 echo "starting doctor sample components..."
 260 start_monitor
 261 start_inspector
 262 start_consumer
 263
 264 echo "creating test user..."
 265 create_test_user
 266
 267 echo "creating VM and alarm..."
 268 boot_vm
 269 create_alarm
 270 wait_for_vm_launch
 271
 272 sleep 60
 273 echo "injecting host failure..."
 274 inject_failure
 275 sleep 10
 276
 277 check_host_status_down
 278 calculate_notification_time
 279
 280 echo "done"