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 fuel local"
  29 INSTALLER_TYPE=${INSTALLER_TYPE:-local}
  30 INSTALLER_IP=${INSTALLER_IP:-none}
  31
  32 SUPPORTED_INSPECTOR_TYPES="sample congress"
  33 INSPECTOR_TYPE=${INSPECTOR_TYPE:-sample}
  34
  35 ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
  36 as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
  37                 --os-tenant-name $DOCTOR_PROJECT"
  38
  39 if [[ ! "$SUPPORTED_INSTALLER_TYPES" =~ "$INSTALLER_TYPE" ]] ; then
  40     echo "ERROR: INSTALLER_TYPE=$INSTALLER_TYPE is not supported."
  41     exit 1
  42 fi
  43
  44 if [[ ! "$SUPPORTED_INSPECTOR_TYPES" =~ "$INSPECTOR_TYPE" ]] ; then
  45     echo "ERROR: INSPECTOR_TYPE=$INSPECTOR_TYPE is not supported."
  46     exit 1
  47 fi
  48
  49 get_compute_host_info() {
  50     # get computer host info which VM boot in
  51     COMPUTE_HOST=$(openstack $as_doctor_user server show $VM_NAME |
  52                    grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }')
  53     compute_host_in_undercloud=${COMPUTE_HOST%%.*}
  54     if [[ -z "$COMPUTE_HOST" ]] ; then
  55         echo "ERROR: failed to get compute hostname"
  56         exit 1
  57     fi
  58
  59     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
  60         COMPUTE_USER=${COMPUTE_USER:-heat-admin}
  61         if [[ "$INSTALLER_IP" == "none" ]] ; then
  62             instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
  63             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  64         fi
  65         COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
  66              "source stackrc; \
  67              nova show $compute_host_in_undercloud \
  68              | awk '/ ctlplane network /{print \$5}'")
  69     elif [[ "$INSTALLER_TYPE" == "fuel" ]] ; then
  70         COMPUTE_USER=${COMPUTE_USER:-root}
  71         if [[ "$INSTALLER_IP" == "none" ]] ; then
  72             instack_mac=$(sudo virsh domiflist fuel-opnfv | awk '/pxebr/{print $5}')
  73             INSTALLER_IP=$(/usr/sbin/arp -e | grep ${instack_mac} | awk '{print $1}')
  74         fi
  75         node_id=$(echo $compute_host_in_undercloud | cut -d "-" -f 2)
  76         COMPUTE_IP=$(sshpass -p r00tme ssh 2>/dev/null $ssh_opts root@${INSTALLER_IP} \
  77              "fuel node|awk -F '|' -v id=$node_id '{if (\$1 == id) print \$5}' |xargs")
  78     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
  79         COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
  80         COMPUTE_IP=$(getent hosts "$COMPUTE_HOST" | awk '{ print $1 }')
  81     fi
  82
  83     if [[ -z "$COMPUTE_IP" ]]; then
  84         echo "ERROR: Could not resolve $COMPUTE_HOST. Either manually set COMPUTE_IP or enable DNS resolution."
  85         exit 1
  86     fi
  87     echo "COMPUTE_HOST=$COMPUTE_HOST"
  88     echo "COMPUTE_IP=$COMPUTE_IP"
  89
  90     # verify connectivity to target compute host
  91     ping -c 1 "$COMPUTE_IP"
  92     if [[ $? -ne 0 ]] ; then
  93         echo "ERROR: can not ping to computer host"
  94         exit 1
  95     fi
  96 }
  97
  98 prepare_compute_ssh() {
  99     ssh_opts_cpu="$ssh_opts"
 100
 101     # get ssh key from installer node
 102     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
 103         sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
 104         sudo chown $(whoami):$(whoami) instack_key
 105         chmod 400 instack_key
 106         ssh_opts_cpu+=" -i instack_key"
 107     elif [[ "$INSTALLER_TYPE" == "fuel" ]] ; then
 108         sshpass -p r00tme scp $ssh_opts root@${INSTALLER_IP}:.ssh/id_rsa instack_key
 109         sudo chown $(whoami):$(whoami) instack_key
 110         chmod 400 instack_key
 111         ssh_opts_cpu+=" -i instack_key"
 112     elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
 113         echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
 114     fi
 115
 116     # verify ssh to target compute host
 117     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
 118     if [[ $? -ne 0 ]] ; then
 119         echo "ERROR: can not ssh to computer host"
 120         exit 1
 121     fi
 122 }
 123
 124 get_consumer_ip() {
 125     CONSUMER_IP=$(sudo ssh $ssh_opts root@$INSTALLER_IP \
 126                   "ip route get $COMPUTE_IP | awk '/ src /{print \$NF}'")
 127     echo "CONSUMER_IP=$CONSUMER_IP"
 128
 129     if [[ -z "$CONSUMER_IP" ]]; then
 130         echo "ERROR: Could not get CONSUMER_IP."
 131         exit 1
 132     fi
 133 }
 134
 135 download_image() {
 136     [ -e "$IMAGE_FILE" ] && return 0
 137     wget "$IMAGE_URL" -o "$IMAGE_FILE"
 138 }
 139
 140 register_image() {
 141     openstack image list | grep -q " $IMAGE_NAME " && return 0
 142     openstack image create "$IMAGE_NAME" \
 143                            --public \
 144                            --disk-format "$IMAGE_FORMAT" \
 145                            --container-format bare \
 146                            --file "$IMAGE_FILE"
 147 }
 148
 149 create_test_user() {
 150     openstack project list | grep -q " $DOCTOR_PROJECT " || {
 151         openstack project create "$DOCTOR_PROJECT"
 152     }
 153     openstack user list | grep -q " $DOCTOR_USER " || {
 154         openstack user create "$DOCTOR_USER" --password "$DOCTOR_PW" \
 155                               --project "$DOCTOR_PROJECT"
 156     }
 157     openstack user role list "$DOCTOR_USER" --project "$DOCTOR_PROJECT" \
 158     | grep -q " $DOCTOR_ROLE " || {
 159         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 160                            --project "$DOCTOR_PROJECT"
 161     }
 162 }
 163
 164 boot_vm() {
 165     # test VM done with test user, so can test non-admin
 166     openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
 167     openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
 168                             --image "$IMAGE_NAME" \
 169                             "$VM_NAME"
 170     sleep 1
 171 }
 172
 173 create_alarm() {
 174     # get vm_id as test user
 175     ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
 176     vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
 177     # TODO(r-mibu): change notification endpoint from localhost to the consumer
 178     # IP address (functest container).
 179     ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
 180         --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
 181         --description "VM failure" \
 182         --enabled True \
 183         --repeat-actions False \
 184         --severity "moderate" \
 185         --event-type compute.instance.update \
 186         -q "traits.state=string::error; traits.instance_id=string::$vm_id"
 187 }
 188
 189 print_log() {
 190     log_file=$1
 191     echo "$log_file:"
 192     sed -e 's/^/    /' "$log_file"
 193 }
 194
 195 start_monitor() {
 196     pgrep -f "python monitor.py" && return 0
 197     sudo -E python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" "$INSPECTOR_TYPE" \
 198         "http://127.0.0.1:$INSPECTOR_PORT/events" > monitor.log 2>&1 &
 199 }
 200
 201 stop_monitor() {
 202     pgrep -f "python monitor.py" || return 0
 203     sudo kill $(pgrep -f "python monitor.py")
 204     print_log monitor.log
 205 }
 206
 207 congress_add_rule() {
 208     name=$1
 209     policy=$2
 210     rule=$3
 211
 212     if ! openstack congress policy rule list $policy | grep -q -e "// Name: $name$" ; then
 213         openstack congress policy rule create --name $name $policy "$rule"
 214     fi
 215 }
 216
 217 congress_del_rule() {
 218     name=$1
 219     policy=$2
 220
 221     if openstack congress policy rule list $policy | grep -q -e "^// Name: $name$" ; then
 222         openstack congress policy rule delete $policy $name
 223     fi
 224 }
 225
 226 congress_setup_rules() {
 227     congress_add_rule host_down classification \
 228         'host_down(host) :-
 229             doctor:events(hostname=host, type="compute.host.down", status="down")'
 230
 231     congress_add_rule active_instance_in_host classification \
 232         'active_instance_in_host(vmid, host) :-
 233             nova:servers(id=vmid, host_name=host, status="ACTIVE")'
 234
 235     congress_add_rule host_force_down classification \
 236         'execute[nova:services.force_down(host, "nova-compute", "True")] :-
 237             host_down(host)'
 238
 239     congress_add_rule error_vm_states classification \
 240         'execute[nova:servers.reset_state(vmid, "error")] :-
 241             host_down(host),
 242             active_instance_in_host(vmid, host)'
 243 }
 244
 245 start_inspector() {
 246     if [[ "$INSPECTOR_TYPE" == "sample" ]] ; then
 247         pgrep -f "python inspector.py" && return 0
 248         python inspector.py "$INSPECTOR_PORT" > inspector.log 2>&1 &
 249     elif [[ "$INSPECTOR_TYPE" == "congress" ]] ; then
 250         nova_api_min_version="2.11"
 251         nova_api_version=$(openstack congress datasource list | \
 252                            grep nova | grep -Po "(?<='api_version': ')[^']*")
 253         [[ -z $nova_api_version ]] && nova_api_version="2.0"
 254         if [[ "$nova_api_version" < "$nova_api_min_version" ]]; then
 255             echo "ERROR: Congress Nova datasource API version < $nova_api_min_version ($nova_api_version)"
 256             exit 1
 257         fi
 258         openstack congress driver list | grep -q " doctor "
 259         openstack congress datasource list | grep -q " doctor " || {
 260             openstack congress datasource create doctor doctor
 261         }
 262         congress_setup_rules
 263     fi
 264 }
 265
 266 stop_inspector() {
 267     if [[ "$INSPECTOR_TYPE" == "sample" ]] ; then
 268         pgrep -f "python inspector.py" || return 0
 269         kill $(pgrep -f "python inspector.py")
 270         print_log inspector.log
 271     elif [[ "$INSPECTOR_TYPE" == "congress" ]] ; then
 272         congress_del_rule host_force_down classification
 273         congress_del_rule error_vm_states classification
 274         congress_del_rule active_instance_in_host classification
 275         congress_del_rule host_down classification
 276     fi
 277 }
 278
 279 start_consumer() {
 280     pgrep -f "python consumer.py" && return 0
 281     python consumer.py "$CONSUMER_PORT" > consumer.log 2>&1 &
 282
 283     # NOTE(r-mibu): create tunnel to the controller nodes, so that we can
 284     # avoid some network problems dpends on infra and installers.
 285     # This tunnel will be terminated by stop_consumer() or after 10 mins passed.
 286     if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
 287         CONTROLLER_IPS=$(sudo ssh $ssh_opts $INSTALLER_IP \
 288                          "source stackrc; \
 289                          nova list | grep ' overcloud-controller-[0-9] ' \
 290                          | sed -e 's/^.*ctlplane=//' -e 's/ *|\$//'")
 291     fi
 292     if [[ -z "$CONTROLLER_IPS" ]]; then
 293         echo "ERROR: Could not get CONTROLLER_IPS."
 294         exit 1
 295     fi
 296     for ip in $CONTROLLER_IPS
 297     do
 298         forward_rule="-R $CONSUMER_PORT:localhost:$CONSUMER_PORT"
 299         tunnel_command="sudo ssh $ssh_opts_cpu $COMPUTE_USER@$ip $forward_rule sleep 600"
 300         $tunnel_command > "ssh_tunnel.${ip}.log" 2>&1 < /dev/null &
 301     done
 302 }
 303
 304 stop_consumer() {
 305     pgrep -f "python consumer.py" || return 0
 306     kill $(pgrep -f "python consumer.py")
 307     print_log consumer.log
 308
 309     # NOTE(r-mibu): terminate tunnels to the controller nodes
 310     for ip in $CONTROLLER_IPS
 311     do
 312         forward_rule="-R $CONSUMER_PORT:localhost:$CONSUMER_PORT"
 313         tunnel_command="sudo ssh $ssh_opts_cpu $COMPUTE_USER@$ip $forward_rule sleep 600"
 314         kill $(pgrep -f "$tunnel_command")
 315         print_log "ssh_tunnel.${ip}.log"
 316     done
 317 }
 318
 319 wait_for_vm_launch() {
 320     echo "waiting for vm launch..."
 321
 322     count=0
 323     while [[ ${count} -lt 60 ]]
 324     do
 325         state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
 326         [[ "$state" == "ACTIVE" ]] && return 0
 327         [[ "$state" == "ERROR" ]] && echo "vm state is ERROR" && exit 1
 328         count=$(($count+1))
 329         sleep 1
 330     done
 331     echo "ERROR: time out while waiting for vm launch"
 332     exit 1
 333 }
 334
 335 inject_failure() {
 336     echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
 337     cat > disable_network.sh << 'END_TXT'
 338 #!/bin/bash -x
 339 dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
 340 sleep 1
 341 sudo ip link set $dev down
 342 sleep 180
 343 sudo ip link set $dev up
 344 sleep 1
 345 END_TXT
 346     sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
 347     chmod +x disable_network.sh
 348     scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
 349     ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
 350 }
 351
 352 calculate_notification_time() {
 353     detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
 354     notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
 355     if ! grep -q "doctor consumer notified at" consumer.log ; then
 356         echo "ERROR: consumer hasn't received fault notification."
 357         exit 1
 358     fi
 359     echo "$notified $detected" | \
 360         awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
 361 }
 362
 363 check_host_status() {
 364     expected_state=$1
 365
 366     host_status_line=$(openstack $as_doctor_user --os-compute-api-version 2.16 \
 367                        server show $VM_NAME | grep "host_status")
 368     host_status=$(echo $host_status_line | awk '{print $4}')
 369     if [ -z "$host_status" ] ; then
 370         echo "ERROR: host_status not reported by: nova show $VM_NAME"
 371         exit 1
 372     elif [[ "$expected_state" =~ "$host_status" ]] ; then
 373         echo "$VM_NAME showing host_status: $host_status"
 374     else
 375         echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
 376         exit 1
 377     fi
 378 }
 379
 380 cleanup() {
 381     set +e
 382     echo "cleanup..."
 383     stop_monitor
 384     stop_inspector
 385     stop_consumer
 386
 387     echo "waiting disabled compute host back to be enabled..."
 388     python ./nova_force_down.py "$COMPUTE_HOST" --unset
 389     sleep 240
 390     check_host_status "UP"
 391     scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
 392     print_log disable_network.log
 393
 394     openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
 395     sleep 1
 396     alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
 397     sleep 1
 398     [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
 399     sleep 1
 400
 401     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
 402     sleep 1
 403     [ -n "$image_id" ] && openstack image delete "$image_id"
 404     openstack role remove "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
 405                               --project "$DOCTOR_PROJECT"
 406     openstack project delete "$DOCTOR_PROJECT"
 407     openstack user delete "$DOCTOR_USER"
 408 }
 409
 410
 411 echo "Note: doctor/tests/run.sh has been executed."
 412
 413 trap cleanup EXIT
 414
 415 echo "preparing VM image..."
 416 download_image
 417 register_image
 418
 419 echo "creating test user..."
 420 create_test_user
 421
 422 echo "creating VM..."
 423 boot_vm
 424 wait_for_vm_launch
 425 openstack $as_doctor_user server show $VM_NAME
 426
 427 echo "get computer host info and prepare to ssh..."
 428 get_compute_host_info
 429 prepare_compute_ssh
 430
 431 echo "creating alarm..."
 432 get_consumer_ip
 433 create_alarm
 434
 435 echo "starting doctor sample components..."
 436 start_inspector
 437 start_monitor
 438 start_consumer
 439
 440 sleep 60
 441 echo "injecting host failure..."
 442 inject_failure
 443 sleep 60
 444
 445 check_host_status "(DOWN|UNKNOWN)"
 446 calculate_notification_time
 447
 448 echo "done"