Meta-OpenStack ============== Description ----------- The meta-openstack layers provide support for building the OpenStack packages. It contains recipes for the nova, glance, keystone, cinder, quantum, swift and horizon components and their dependencies. The main meta-openstack layer, works in conjunction with the meta-openstack* layers to configure and deploy a system. Components ---------- * All the openstack packages are python packages. They can be found in the recipes-devtools/python folder. Each component has been split into multiple packages similar to the system used by other Linux distributions. * The configuration files for each package can be found in the files folder for each package. The debug and verbose options have been enabled for capturing meaningful information in the logs. The packages have been configured following the model used by devstack. Each package has to initialize a database before it is used for the first time. This is done using a postinstall script that is run the first time the image is booted. This is why the image boot time is longer the first time. * System-V initscripts are also provided in order to start the services at boot time. * Systemd support is not complete. * We used postgresql (package) for the database backend. The layer contains an initscript that starts the DB server at boot time. Tests were done using sqlite3 and MySQL and there were no errors. * The RabbitMQ server is used for the AMQP message queues. The server starts at boot time with the default configuration. * The layer also contains three packagegroups: ** packagegroup-cloud-controller - required packages for building a controller node. This provides all functionality except the actual hosting of the virtual machines or network services. This includes the database server, AMQP server and all the openstack services except nova-compute. ** packagegroup-cloud-compute - packages for a processing node. This node runs the compute service as well as the network service agent (in our case, the Open vSwitch plugin agent). This server also manages the KVM hypervisor. ** packagegroup-cloud-network: this provides networking services like DHCP, layer 2 switching, layer 3 routing and metadata connectivity. * When creating an image, multiple packagegroups can be used to obtain a target that has the functionality of both a controller and compute node. Dependencies ------------ * This layer depends on components from the poky, meta-virtualization and meta-openembedded layers. You can find the exact URIs of the repos and the necessary revisions in the README file. Building an image ----------------- * There are three new target images: openstack-image-compute, openstack-image-network and openstack-image-controller that contain the packagegroups with the same name, that have been described in the previous section. * Once a buildir has been initialized you have to append the necessary layers to the bblayers.conf file: /meta-virtualization \ /meta-cloud-services \ /meta-cloud-services/meta-openstack--deploy \ /meta-cloud-services/meta-openstack \ /meta-cloud-services/meta-openstack-qemu \ # optional, add if using qemu /meta-cloud-services/meta-openstack \ /meta-openembedded/meta-oe \ /meta-openembedded/meta-networking \ /meta-openembedded/meta-python \ /meta-openembedded/meta-filesystems \ /meta-openembedded/meta-webserver \ /meta-openembedded/meta-ruby * All images must use systemd init system. After the builddir has been initialized you have to append the necessary variables to ensure that systemd will be used in your images: DISTRO_FEATURES:append = " systemd" DISTRO_FEATURES_BACKFILL_CONSIDERED += "sysvinit" VIRTUAL-RUNTIME_init_manager = "systemd" VIRTUAL-RUNTIME_initscripts = "systemd-compat-units" Additionally activiate the meta-virtualization layer: DISTRO_FEATURES:append = " virtualization kvm" Package configurations ---------------------- The identity.sh script creates the necessary users, services and endpoints for the Keystone identity system. If you want to customize the usernames or passwords don't forget to change the information in the configuration files for the services as well. The hosts.bbclass contains the IP addresses of the compute and controller nodes that will seed the system. It also contains the IP address of the node being built "MY_IP". Override this class in a layer to provide values that are specific to your configuration. The defaults are suitable for a 2 node system launched via runqemu. If deploying to a simulated system, add the qemu deployment layer to the bblayers.conf file, after the node type deployment layer. * Sample Guest Image * If a sample guest image is desired on a control node, the following can be added to local.conf: IMAGE_INSTALL:append = " cirros-guest-image" * Cinder Additional Packages * On compute node, by default openstack-image-compute doesn't include initiator iscsi package. If open-iscsi-user recipe exists in layers, the following can be added to local.conf: CINDER_EXTRA_FEATURES += " open-iscsi-user" If glusterfs recipi exists in layers, the following can be added to local.conf: CINDER_EXTRA_FEATURES += " glusterfs" Running an image ---------------- To test the images, you can run them using the runqemu script (on a machine with appropriate accleration). In order to use the command line clients (nova, keystone, glance etc) some environmental variables have to be set. These are required by the openstack services to connect to the identity service and authenticate the user. These can be found in /root/.bashrc or /etc/nova/openrc. * Controller node * All the installed OpenStack services nova(except compute), keystone, glance, cinder, quantum, swift horizon should be running after a successful boot. % bitbake openstack-image-controller % runqemu qemux86-64 openstack-image-controller kvm nographic qemuparams="-m 4096" The dashboard component is listening for new connections on port 8080. You can connect to it using any browser. * Compute Node * All the installed OpenStack compute services nova, quantum, should be running after a successful boot. % bitbake openstack-image-compute % runqemu qemux86-64 openstack-image-compute kvm nographic qemuparams="-m 4096 -smp 4" * Image Launch * Assuing that the cirros-guest-image has been added to the control image, the following steps will validate a simple compute node guest launch: % . /etc/nova/openrc % glance image-create --name myFirstImage --is-public true \ --container-format bare --disk-format qcow2 --file /root/images/cirros-*-x86_64-disk.img % neutron net-create mynetwork % nova boot --image myFirstImage --flavor 1 myinstance * Cinder Multi-backend * Cinder currently is configured to support multi-backend: lvm-iscsi, nfs, glusterfs, and ceph rbd. When a Cinder volume is created, it's needed to be specified which backend it belongs to through "--volume_type" option passed in "cinder create" command. The Cinder volume types for these backends can be created following the steps: % . /etc/nova/openrc % cinder type-create lvm_iscsi % cinder type-key lvm_iscsi set volume_backend_name=LVM_iSCSI % cinder type-create nfs % cinder type-key nfs set volume_backend_name=Generic_NFS % cinder type-create glusterfs % cinder type-key glusterfs set volume_backend_name=GlusterFS % cinder type-create cephrbd % cinder type-key cephrbd set volume_backend_name=RBD_CEPH For example, to create 1G Cinder volume in lvm-iscsi backend: % cinder create --volume_type lvm_iscsi --display_name=lvm_vol 1 !! Hint !! When using a multi-node setup it is recommended that each host have a different hostname and that every host knows the other hosts. * Notes * Keystone token query: % sudo -u postgres psql -d keystone -c "select count(*) from token" Ceilometer: % ceilometer meter-list % ceilometer sample-list --meter cpu