This document is intended for developers and operators. For an end-user guide, please see the end-user quick-start guide and cookbook in this documentation repository.
Octavia in devstack with a default (non-HA) configuration will deploy one amphora VM per loadbalancer deployed. The current default amphora image also requires at least 1GB of RAM to run effectively. As such it is important that your devstack environment has enough resources dedicated to it to run all its necessary components. For most devstack environments, the limiting resource will be RAM. At the present time, we recommend at least 12GB of RAM for the standard devstack defaults, or 8GB of RAM if cinder and swift are disabled. More is recommended if you also want to run a couple of application server VMs (so that Octavia has something to load balance within your devstack environment).
Also, because the current implementation of Octavia delivers load balancing services using amphorae that run as Nova virtual machines, it is effectively mandatory to enable nested virtualization. The software will work with software emulated CPUs, but be unusably slow. The idea is to make sure the BIOS of the systems you’re running your devstack on have virtualization features enabled (Intel VT-x, AMD-V, etc.), and the virtualization software you’re using exposes these features to the guest VM (sometimes called nested virtualization). For more information, see: http://docs.openstack.org/developer/devstack/guides/devstack-with-nested-kvm.html
The devstack environment we recommend should be running Ubuntu Linux 14.04 or later. These instructions may work for other Linux operating systems or environments. However, most people doing development on Octavia are using Ubuntu for their test environment, so you will probably have the easiest time getting your devstack working with that OS.
This document is not a definitive guide for deploying Octavia in every production environment. There are many ways to deploy Octavia depending on the specifics and limitations of your situation. For example, in our experience, large production environments often have restrictions, hidden “features” or other elements in the network topology which mean the default Neutron networking stack (with which Octavia was designed to operate) must be modified or replaced with a custom networking solution. This may also mean that for your particular environment, you may need to write your own custom networking driver to plug into Octavia. Obviously, instructions for doing this are beyond the scope of this document.
We hope this document provides the cloud operator or distribution creator with a basic understanding of how the Octavia components fit together practically. Through this, it should become more obvious how components of Octavia can be divided or duplicated across physical hardware in a production cloud environment to aid in achieving scalability and resiliency for the Octavia load balancing system.
In the interest of keeping this guide somewhat high-level and avoiding obsolescence or operator/distribution-specific environment assumptions by specifying exact commands that should be run to accomplish the tasks below, we will instead just describe what needs to be done and leave it to the cloud operator or distribution creator to “do the right thing” to accomplish the task for their environment. If you need guidance on specific commands to run to accomplish the tasks described below, we recommend reading through the plugin.sh script in devstack subdirectory of this project. The devstack plugin exercises all the essential components of Octavia in the right order, and this guide will mostly be an elaboration of this process.
The scope of this guide is to provide a basic overview of setting up all the components of Octavia in a production environment, assuming that the default in-tree drivers and components (including a “standard” Neutron install) are going to be used.
For the purposes of this guide, we will therefore assume the following core components have already been set up for your production OpenStack environment:
By default Octavia will use the ‘neutron’ user for keystone authentication, and the admin user for interactions with all other serivces. However, it doesn’t actually share neutron’s database or otherwise access Neutron outside of Neutron’s API, so a dedicated ‘octavia’ keystone user should generally be created for Octavia to use.
You must:
Octavia makes use of an “LB Network” exclusively as a management network that the controller uses to talk to amphorae and vice versa. All the amphorae that Octavia deploys will have interfaces and IP addresses on this network. Therefore, it’s important that the subnet deployed on this network be sufficiently large to allow for the maximum number of amphorae and controllers likely to be deployed throughout the lifespan of the cloud installation.
At the present time, only IPv4 subnets have been tested as the LB Network (for example: 172.31.0.0/16), though there are plans to eventually support IPv6 subnets for the LB Network.
The LB Network is isolated from tenant networks on the amphorae by means of network namespaces on the amphorae. Therefore, operators need not be concerned about overlapping subnet ranges with tenant networks.
You must also create a Neutron security group which will be applied to amphorae created on the LB network. It needs to allow amphorae to send UDP heartbeat packets to the health monitor (by default, UDP port 5555), and ingress on the amphora’s API (by default, TCP port 9443). It can also be helpful to allow SSH access to the amphorae from the controller for troubleshooting purposes (ie. TCP port 22), though this is not strictly necessary in production environments.
Amphorae will send periodic health checks to the controller’s health manager. Any firewall protecting the interface on which the health manager listens must allow these packets from amphorae on the LB Network (by default, UDP port 5555).
Finally, you need to add routing or interfaces to this network such that the Octavia controller (which will be described below) is able to communicate with hosts on this network. This also implies you should have some idea where you’re going to run the Octavia controller components.
You must:
Octavia deploys amphorae based on a virtual machine disk image. By default we use the OpenStack diskimage-builder project for this. Scripts to accomplish this are within the diskimage-create directory of this repository. In addition to creating the disk image, configure a Nova flavor to use for amphorae, and upload the disk image to glance.
You must:
This seems somewhat obvious, but the important things to note here are that you should put this somewhere on the network where it will have access to the database (to be initialized below), the oslo messaging system, and the LB network. Octavia uses the standard python setuptools, so installation of the software itself should be straightforward.
Running multiple instances of the individual Octavia controller components on seperate physical hosts is recommended in order to provide scalability and availability of the controller software.
One important security note: In 0.9 of Octavia, the Octavia API is designed to be consumed only by the Neutron-LBaaS v2 Octavia driver. As such, there is presently no authentication required to use the Octavia API, and therefore the Octavia API should only be accessible on trusted network segments (specifically, the segment that runs the neutron-services daemons.)
The Octavia controller presently consists of several components which may be split across several physical machines. For the 0.9 release of Octavia, the important (and potentially separable) components are the controller worker, housekeeper, health manager and API controller. Please see the component diagrams elsewhere in this repository’s documentation for detailed descriptions of each. Please use the following table for hints on which controller components need access to outside resources:
| Component | Resource | ||
|---|---|---|---|
| LB Network | Database | OSLO messaging | |
| API | No | Yes | Yes |
| controller worker | Yes | Yes | Yes |
| health monitor | Yes | Yes | No |
| housekeeper | Yes | Yes | No |
In addition to talking to each other via OSLO messaging, various controller components must also communicate with other OpenStack components, like nova, neutron, barbican, etc. via their APIs.
You must:
Octavia presently allows for one method for the controller to communicate with amphorae: The amphora REST API. Both amphora API and Octavia controller do bi-directional certificate-based authentication in order to authenticate and encrypt communication. You must therefore create appropriate TLS certificates which will be used for key signing, authentication, and encryption. There is a helper script to do this in this repository under: bin/create_certficiates.sh
Please note that certificates created with this helper script may not meet your organization’s security policies, since they are self-signed certificates with arbitrary bit lengths, expiration dates, etc. Operators should obviously follow their own security guidelines in creating these certificates.
In addition to the above, it can sometimes be useful for cloud operators to log into running amphorae to troubleshoot problems. The standard method for doing this is to use SSH from the host running the controller worker. In order to do this, you must create an SSH public/private key pair specific to your cloud (for obvious security reasons). You must add this keypair to nova. You must then also update octavia.conf with the keypair name you used when adding it to nova so that amphorae are initialized with it on boot.
See the Troubleshooting Tips section below for an example of how an operator can SSH into an amphora.
You must:
Going into all of the specifics of how Octavia can be configured is actually beyond the scope of this document. For full documentation of this, please see the configuration reference: Octavia Configuration Options
A configuration template can be found in etc/octavia.conf in this repository.
It’s also important to note that this configuration file will need to be updated with UUIDs of the LB network, amphora security group, amphora image tag, SSH key path, TLS certificate path, database credentials, etc.
At a minimum, the configuration should specify the following, beyond the defaults. Your specific environment may require more than this:
| Section | Configuration parameter |
|---|---|
| DEFAULT | transport_url |
| database | connection |
| certificates | ca_certificate |
| certificates | ca_private_key |
| certificates | ca_private_key_passphrase |
| controller_worker | amp_boot_network_list |
| controller_worker | amp_flavor_id |
| controller_worker | amp_image_owner_id |
| controller_worker | amp_image_tag |
| controller_worker | amp_secgroup_list |
| controller_worker | amp_ssh_key_name [1] |
| controller_worker | amphora_driver |
| controller_worker | compute_driver |
| controller_worker | loadbalancer_topology |
| controller_worker | network_driver |
| haproxy_amphora | client_cert |
| haproxy_amphora | server_ca |
| health_manager | bind_ip |
| health_manager | controller_ip_port_list |
| health_manager | heartbeat_key |
| house_keeping | spare_amphora_pool_size |
| keystone_authtoken | admin_password |
| keystone_authtoken | admin_tenant_name |
| keystone_authtoken | admin_user |
| keystone_authtoken | auth_uri |
| keystone_authtoken | auth_version |
| oslo_messaging | topic |
| oslo_messaging_rabbit | rabbit_host |
| oslo_messaging_rabbit | rabbit_userid |
| oslo_messaging_rabbit | rabbit_password |
| [1] | This is technically optional, but extremely useful for troubleshooting. |
You must:
One configuration directive deserves some extra consideration in this document:
Depending on the specifics of your production environment, you may decide to run Octavia with a non-empty “spares pool.” Since the time it takes to spin up a new amphora can be non-trivial in some cloud environments (and the reliability of such operations can sometimes be less than ideal), this directive instructs Octavia to attempt to maintain a certain number of amphorae running in an idle, unconfigured state. These amphora will run base amphora health checks and wait for configuration from the Octavia controller. The overall effect of this is to greatly reduce the time it takes and increase the reliability of deploying a new load balancing service on demand. This comes at the cost of having a number of deployed amphorae which consume resources but are not actively providing load balancing services, and at the cost of not being able to use Nova anti-affinity features for ACTIVE-STANDBY load balancer topologies.
This is controlled through alembic migrations under the octavia/db directory in this repository. A tool has been created to aid in the initialization of the octavia database. This should be available under /usr/local/bin/octavia-db-manage on the host on which the octavia controller worker is installed. Note that this tool looks at the /etc/octavia/octavia.conf file for its database credentials, so initializing the database must happen after Octavia is configured.
It’s also important to note here that all of the components of the Octavia controller will need direct access to the database (including the API handler), so you must ensure these components are able to communicate with whichever host is housing your database.
You must:
We recommend using upstart / systemd scripts to ensure the components of the Octavia controller are all started and kept running. It of course doesn’t hurt to first start by running these manually to ensure configuration and communication is working between all the components.
You must:
This is fairly straightforward. Neutron LBaaS needs to be directed to use the Octavia service provider. There should be a line like the following in /etc/neutron/neutron_lbaas.conf file’s [service providers] section:
service_provider = LOADBALANCERV2:Octavia:neutron_lbaas.drivers.octavia.driver.OctaviaDriver:default
In addition to the above you must add the octavia API base_url to the [octavia] section of /etc/neutron/neutron.conf. For example:
[octavia]
base_url=http://127.0.0.1:9876
You must:
This isn’t strictly necessary for all cloud installations, however, if yours makes use of the Horizon GUI interface for tenants, it is probably also a good idea to make sure that it is configured with the Neutron-LBaaS v2 extension.
You may:
If all of the above instructions have been followed, it should now be possible to deploy load balancing services using the python neutronclient CLI, communicating with the neutron-lbaas v2 API.
Example:
# neutron lbaas-loadbalancer-create --name lb1 private-subnet
# neutron lbaas-loadbalancer-show lb1 # Keep re-running this command until this shows a provisioning status of 'ACTIVE'
# neutron lbaas-listener-create --name listener1 --loadbalancer lb1 --protocol HTTP --protocol-port 80
Upon executing the above, log files should indicate that an amphora is deployed to house the load balancer, and that this load balancer is further modified to include a listener. The amphora should be visible to the octavia or admin tenant using the nova list command, and the listener should respond on the load balancer’s IP on port 80 (with an error 503 in this case, since no pool or members have been defined yet—but this is usually enough to see that the Octavia load balancing system is working). For more information on configuring load balancing services as a tenant, please see the end-user quick-start guide and cookbook.
The troubleshooting hints in this section are meant primarily for developers or operators troubleshooting underlying Octavia components, rather than end-users or tenants troubleshooting the load balancing service itself.
If you are using the reference amphora image, it may be helpful to log into running amphorae when troubleshooting service problems. To do this, first discover the lb_network_ip address of the amphora you would like to SSH into by looking in the amphora table in the octavia database. Then from the host housing the controller worker, run:
ssh -i /etc/octavia/.ssh/octavia_ssh_key ubuntu@[lb_network_ip]