Load balancing#

Get answers to common technical questions about failover and high availability for Rackspace Private Cloud Powered By OpenStack (RPCO) 17.1.

Do the services need to be modified to enable load balancing in one-arm mode?


What load balancing options can Rackspace provide in a customer data center?

RPCO supports two models for load balancing in a customer data center:

  • Unmanaged Load Balancers (a.k.a customer managed)

  • Managed Load Balancers

    • Support for hardware based HAProxy

    • Support for hardware based F5 BigIP

    • Support for OpenStack Octavia (self-service)

What is the load balancer configuration for the reference architecture?

The load balancers should be a high availability (HA) pair, so that at least two load balancers are available to balance the traffic across the infrastructure.

Can the same load balancer be used for service and VM traffic?

This depends on the devices being used. If the HAProxy option is employed, those devices cannot be used to balance data plane traffic (they are restricted to control plane only). If the F5 BigIP devices are being utilized, those devices can be used to balance data plane traffic. The only limitation is the capacity of the physical devices. OpenStack Octavia is only available for data plane traffic at this time.

What services should be load balanced?

All services should be load balanced, even if the service is running on a single container. This ensures that the traffic is managed by the load balancer and that failovers are graceful.

What if customers require API access to the load balancer to manage floating IP addresses and VIP pools?

RPCO does not provide customer API access to any control plane load balancers. RPCO also does not support customer API access to the F5 Local Traffic Manager (LTM). A separate customer-managed LTM or Octavia must be configured to handle VM (data plane) Load Balancing.

What is OpenStack Octavia?

RPCO offers OpenStack Octavia as a self-service load balancing option. Octavia provides an active/standby, highly available load balancer by using service virtual machines provisioned in the compute resources of the cloud. The active and standby service virtual machines are deployed on separate compute hosts to ensure load balancer availability in the case of a host outage.

Are Octavia flow logs available?

At this time, RPCO does not store flow logs for traffic passing through an Octavia load balancer.

What networks can be used for Octavia load balancing?

Octavia allows you to specify the OpenStack Neutron network for the frontend virtual IP (VIP) and the backend member servers. In addition, you can assign floating IPs to the Octavia VIP port.

What Octavia providers are supported on RPCO?

Currently, only the reference implementation driver (amphora) is available in RPCO deployments. The amphora driver is based on the HAProxy load balancing engine.