# http://www.apache.org/licenses/LICENSE-2.0
##############################################################################
-[[ "${CI_DEBUG:-true}" == "true" ]] && set -x
+[[ "${CI_DEBUG:-true}" == [Tt]rue ]] && set -x
IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-disk.img
IMAGE_NAME=cirros
SUPPORTED_INSTALLER_TYPES="apex local"
INSTALLER_TYPE=${INSTALLER_TYPE:-apex}
INSTALLER_IP=${INSTALLER_IP:-none}
-COMPUTE_USER=${COMPUTE_USER:-none}
ssh_opts="-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
get_compute_host_info() {
# get computer host info which VM boot in
COMPUTE_HOST=$(openstack $as_doctor_user server show $VM_NAME |
- grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }' |
- awk -F '.' '{print $1}')
+ grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }')
if [[ -z "$COMPUTE_HOST" ]] ; then
echo "ERROR: failed to get compute hostname"
exit 1
fi
if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
+ compute_host_in_undercloud=${COMPUTE_HOST%%.*}
COMPUTE_USER=${COMPUTE_USER:-heat-admin}
if [[ "$INSTALLER_IP" == "none" ]] ; then
instack_mac=$(sudo virsh domiflist instack | awk '/default/{print $5}')
fi
COMPUTE_IP=$(sudo ssh $ssh_opts $INSTALLER_IP \
"source stackrc; \
- openstack server show $COMPUTE_HOST \
+ nova show $compute_host_in_undercloud \
| awk '/ ctlplane network /{print \$5}'")
elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
COMPUTE_USER=${COMPUTE_USER:-$(whoami)}
exit 1
fi
fi
+ echo "COMPUTE_HOST=$COMPUTE_HOST"
+ echo "COMPUTE_IP=$COMPUTE_IP"
# verify connectivity to target compute host
ping -c 1 "$COMPUTE_IP"
# get ssh key from installer node
if [[ "$INSTALLER_TYPE" == "apex" ]] ; then
sudo scp $ssh_opts root@"$INSTALLER_IP":/home/stack/.ssh/id_rsa instack_key
+ sudo chown $(whoami):$(whoami) instack_key
+ chmod 400 instack_key
+ ssh_opts_cpu+=" -i instack_key"
elif [[ "$INSTALLER_TYPE" == "local" ]] ; then
echo "INSTALLER_TYPE set to 'local'. Assuming SSH keys already exchanged with $COMPUTE_HOST"
fi
- sudo chown $(whoami):$(whoami) instack_key
- chmod 400 instack_key
- ssh_opts_cpu+=" -i instack_key"
-
# verify ssh to target compute host
ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'exit'
if [[ $? -ne 0 ]] ; then
}
get_consumer_ip() {
- #get network of the compute_ip address
- network_pre=${COMPUTE_IP%.*}
- network=${network_pre}.0/24
-
- #if there is a dedicated route, use it
- dev=$(ip route |grep $network | awk '{print $3}' |head -n 1)
-
- #if there is no route, use default route
- [[ -z $dev ]] && dev=$(ip route |grep ^default | awk '{print $3}' |head -n 1)
+ CONSUMER_IP=$(ip route get $COMPUTE_IP | awk '/ src /{print $NF}')
+ echo "CONSUMER_IP=$CONSUMER_IP"
- CONSUMER_IP=$(ip addr show $dev |grep inet | grep $network_pre | awk '{print $2}' | cut -d'/' -f1)
+ if [[ -z "$CONSUMER_IP" ]]; then
+ echo "ERROR: Could not get CONSUMER_IP."
+ exit 1
+ fi
}
download_image() {
echo "disabling network of compute host [$COMPUTE_HOST] for 3 mins..."
cat > disable_network.sh << 'END_TXT'
#!/bin/bash -x
-dev=$(sudo ip route | awk '/^default/{print $5}')
+dev=$(sudo ip a | awk '/ @COMPUTE_IP@\//{print $7}')
sleep 1
sudo ip link set $dev down
sleep 180
sudo ip link set $dev up
sleep 1
END_TXT
+ sed -i -e "s/@COMPUTE_IP@/$COMPUTE_IP/" disable_network.sh
chmod +x disable_network.sh
scp $ssh_opts_cpu disable_network.sh "$COMPUTE_USER@$COMPUTE_IP:"
ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'nohup ./disable_network.sh > disable_network.log 2>&1 &'
calculate_notification_time() {
detected=$(grep "doctor monitor detected at" monitor.log | awk '{print $5}')
notified=$(grep "doctor consumer notified at" consumer.log | awk '{print $5}')
+ if ! grep -q "doctor consumer notified at" consumer.log ; then
+ echo "ERROR: consumer hasn't received fault notification."
+ exit 1
+ fi
echo "$notified $detected" | \
awk '{d = $1 - $2; if (d < 1 && d > 0) print d " OK"; else print d " NG"}'
}
if [ -z "$host_status" ] ; then
echo "ERROR: host_status not reported by: nova show $VM_NAME"
exit 1
- elif [[ "$host_status" != "$expected_state" ]] ; then
+ elif [[ "$expected_state" =~ "$host_status" ]] ; then
+ echo "$VM_NAME showing host_status: $host_status"
+ else
echo "ERROR: host_status:$host_status not equal to expected_state: $expected_state"
exit 1
- else
- echo "$VM_NAME showing host_status: $host_status"
fi
}
echo "waiting disabled compute host back to be enabled..."
python ./nova_force_down.py "$COMPUTE_HOST" --unset
- sleep 180
+ sleep 240
check_host_status "UP"
ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" \
"[ -e disable_network.log ] && cat disable_network.log"
- openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack server delete "$VM_NAME"
+ openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
sleep 1
- alarm_id=$(ceilometer alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
+ alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
sleep 1
- [ -n "$alarm_id" ] && ceilometer alarm-delete "$alarm_id"
+ [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
sleep 1
image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
echo "creating VM..."
boot_vm
wait_for_vm_launch
+openstack $as_doctor_user server show $VM_NAME
echo "get computer host info and prepare to ssh..."
get_compute_host_info
sleep 60
echo "injecting host failure..."
inject_failure
-sleep 10
+sleep 60
-check_host_status "DOWN"
+check_host_status "(DOWN|UNKNOWN)"
calculate_notification_time
echo "done"
Code Review / doctor.git / blobdiff