#!/bin/bash # # This script is meant to be run once after running start for the first # time. This script downloads a cirros image and registers it. Then it # configures networking and nova quotas to allow 40 m1.small instances # to be created. ARCH=$(uname -m) IMAGE_PATH=/opt/cache/files/ IMAGE_URL=http://download.cirros-cloud.net/0.4.0/ IMAGE=cirros-0.4.0-${ARCH}-disk.img IMAGE_NAME=cirros IMAGE_TYPE=linux # This EXT_NET_CIDR is your public network,that you want to connect to the internet via. EXT_NET_CIDR='10.0.2.0/24' EXT_NET_RANGE='start=10.0.2.150,end=10.0.2.199' EXT_NET_GATEWAY='10.0.2.1' # Sanitize language settings to avoid commands bailing out # with "unsupported locale setting" errors. unset LANG unset LANGUAGE LC_ALL=C export LC_ALL for i in curl openstack; do if [[ ! $(type ${i} 2>/dev/null) ]]; then if [ "${i}" == 'curl' ]; then echo "Please install ${i} before proceeding" else echo "Please install python-${i}client before proceeding" fi exit fi done # Test for credentials set if [[ "${OS_USERNAME}" == "" ]]; then echo "No Keystone credentials specified. Try running source openrc" exit fi # Test to ensure configure script is run only once if openstack image list | grep -q cirros; then echo "This tool should only be run once per deployment." exit fi echo Checking for locally available cirros image. # Let's first try to see if the image is available locally # nodepool nodes caches them in $IMAGE_PATH if ! [ -f "${IMAGE_PATH}/${IMAGE}" ]; then IMAGE_PATH='./' if ! [ -f "${IMAGE_PATH}/${IMAGE}" ]; then echo None found, downloading cirros image. curl -L -o ${IMAGE_PATH}/${IMAGE} ${IMAGE_URL}/${IMAGE} fi else echo Using cached cirros image from the nodepool node. fi EXTRA_PROPERTIES= if [ ${ARCH} == aarch64 ]; then EXTRA_PROPERTIES="--property hw_firmware_type=uefi" fi echo Creating glance image. openstack image create --disk-format qcow2 --container-format bare --public \ --property os_type=${IMAGE_TYPE} ${EXTRA_PROPERTIES} --file ${IMAGE_PATH}/${IMAGE} ${IMAGE_NAME} echo Configuring neutron. openstack network create --external --provider-physical-network physnet1 \ --provider-network-type flat public1 openstack subnet create --no-dhcp \ --allocation-pool ${EXT_NET_RANGE} --network public1 \ --subnet-range ${EXT_NET_CIDR} --gateway ${EXT_NET_GATEWAY} public1-subnet openstack network create --provider-network-type vxlan demo-net openstack subnet create --subnet-range 10.0.0.0/24 --network demo-net \ --gateway 10.0.0.1 --dns-nameserver 8.8.8.8 demo-subnet openstack router create demo-router openstack router add subnet demo-router demo-subnet openstack router set --external-gateway public1 demo-router # Get admin user and tenant IDs ADMIN_USER_ID=$(openstack user list | awk '/ admin / {print $2}') ADMIN_PROJECT_ID=$(openstack project list | awk '/ admin / {print $2}') ADMIN_SEC_GROUP=$(openstack security group list --project ${ADMIN_PROJECT_ID} | awk '/ default / {print $2}') # Sec Group Config openstack security group rule create --ingress --ethertype IPv4 \ --protocol icmp ${ADMIN_SEC_GROUP} openstack security group rule create --ingress --ethertype IPv4 \ --protocol tcp --dst-port 22 ${ADMIN_SEC_GROUP} # Open heat-cfn so it can run on a different host openstack security group rule create --ingress --ethertype IPv4 \ --protocol tcp --dst-port 8000 ${ADMIN_SEC_GROUP} openstack security group rule create --ingress --ethertype IPv4 \ --protocol tcp --dst-port 8080 ${ADMIN_SEC_GROUP} if [ ! -f ~/.ssh/id_rsa.pub ]; then echo Generating ssh key. ssh-keygen -t rsa -f ~/.ssh/id_rsa fi if [ -r ~/.ssh/id_rsa.pub ]; then echo Configuring nova public key and quotas. openstack keypair create --public-key ~/.ssh/id_rsa.pub mykey fi # Increase the quota to allow 40 m1.small instances to be created # 40 instances openstack quota set --instances 40 ${ADMIN_PROJECT_ID} # 40 cores openstack quota set --cores 40 ${ADMIN_PROJECT_ID} # 96GB ram openstack quota set --ram 96000 ${ADMIN_PROJECT_ID} # add default flavors, if they don't already exist if ! openstack flavor list | grep -q m1.tiny; then openstack flavor create --id 1 --ram 512 --disk 1 --vcpus 1 m1.tiny openstack flavor create --id 2 --ram 2048 --disk 20 --vcpus 1 m1.small openstack flavor create --id 3 --ram 4096 --disk 40 --vcpus 2 m1.medium openstack flavor create --id 4 --ram 8192 --disk 80 --vcpus 4 m1.large openstack flavor create --id 5 --ram 16384 --disk 160 --vcpus 8 m1.xlarge fi cat << EOF Done. To deploy a demo instance, run: openstack server create \\ --image ${IMAGE_NAME} \\ --flavor m1.tiny \\ --key-name mykey \\ --network demo-net \\ demo1 EOF