In order for the solution to be ready for production, different systems need to be fault-tolerant. The Architecture Guide design utilizes both hardware-based and software-based redundancy. This includes, but is not limited to:
- Operating Systems are hosted on either a RAID 1 or RAID 10 hard drive set.
- Critical network connections from server to switch utilize network bonding.
- Multiple Controllers host the control plane services. Minimally, 3 Controller nodes are required.
- Control plane services are made highly available utilizing ha-proxy, Corosync, Pacemaker, and/or native resiliency.
- Red Hat Ceph storage utilizes a minimum of three (3) servers.
- Red Hat Ceph storage is used with either replication or erasure coding.
- Optional: Instance High Availability. This validated option utilizes remote pacemaker to monitor the Compute nodes. If preset criteria are met, the process of migrating instances off of the failing Compute nodes to others begins. If a Compute node completely fails, pacemaker can be configured to start the failed instances on different Compute nodes.
- Optional: Dell EMC Unity storage, SC series storage, and Dell EMC PowerMax storage arrays are highly available.
Note: The Solution Admin Host, and the server hosted on it (Red Hat OpenStack Director), are not fault tolerant, but are not required for continued functionality of the OpenStack cluster.