High Availability Servers are generally defined within the business context, and may include scenarios for ‘high availability’ such as:
Although ‘design papers’ prefer that ‘server grade’ equipment be used on all ‘server systems’ we generally do not have that flexibility, and OpenBSD high avalability technologies are ‘service grade’ agnostic and designed to take effective use of disparate systems (i.e. the High Availability tools are designed to allow lower-end equipment to augment, supplement services by higher end/priced equipment, or completely replace them.)
The following are some notes on where OpenBSD may be a solution for High Availability. These are evolving notes, so please confirm through other sources before you pursue anything from these notes.
This is the simplest solution to install, though potentially most difficult to maintain or verify. In the warm failover system we build System-B as a copy of the live System-A and deploy these systems.
Advantage: Simplest to roll-out
Disadvantage: Requires manual intervention in failover scenario.
Two methods exist for switching from a failed Server-A to warm Server-B.
Service recovery will be to unplug the failed system Server-A and plug in the warm functioning system Server-B. Due to contention on the network for IP addresses, both Server-A and Server-B the systems cannot be ‘live’ by using only a duplicate of the existing system.
To allow both Server-A and Server-B to be both physically connected, at least the IP addresses on the servers need to be different and ‘switching’ Server-B into “live” mode will require a system administrator to log onto and disable relevant IP configurations on Server-A and enable correct configuration on Server-B
A variant of option (b) would be to retain the IP addresses but use the border firewalls to route to what is selected as the ‘live’ system.
Examples for these options would be to roll-out commodity boxes, with current software as failover for aging systems.
OpenBSD base OS support’s load-balancing traffic in and out, such that it can ‘share’ traffic amongst servers. Additional load-balancing tools are available through the Ports Package Collection.
Traffic ‘sharing’ is transparent to the destination servers, and the interdepence of those servers determine the complexity required for deployment. Services with no interdepence between multiple servers (such as static web sites, download servers, MX Proxies) can be readily scaled through adding more boxes to the service pool.
Advantage: Increases the capacity of your service linearly depending on number of additional servers added to the ‘pool.’
Likewise, hosts can be removed or maintained while service is provisioned via other servers in the ‘load pool.’
The Hot Failover of hosts is supported by OpenBSD using CARP. which in it’s initial implementation is supported by at least the Packet Filter Firewall.
Servers in the ‘hot failover pool’ use a sharing system for their IP Addresses. One host is selected as the ‘primary/master’ host servicing all requests, while the rest of the machines are ‘live standby’. The CARP service manages which host is the active master, applications need to manage feeding data between hosts to ensure that host can “take-over” should the master server fail.
Any failure on the ‘master’ host to service network traffic is immediately noticed by other servers and another server ‘takes over’ the ‘master’ mode.
Data/State synchronisation is commonly through a shared storage device (such as a SAN.) Where an application or service has in memory data that needs to be replicated to the live standby hosts, the application needs a mechanism to distribute that data.