Step 6. Ensuring availability

Case: one server on a site has failed. The entire site functionality remains available, meaning all previous cases continue to work.

The input requirement is that the system must be highly available and provide full functionality when any server is switched off and when communication between any sites is lost. This is achieved by redundancy of all necessary roles on each site.

Let us consider the case of deploying the system on the following infrastructure.

sites

The distribution of the roles known at this step might, for example, look like this:

sites_ws_dc

In the role distribution scheme above, it is worth noting that even if one site simultaneously loses communication with other sites and one of its servers, the case from step 1 will still be executed there. The system will effectively switch to the single-server mode discussed in step 1 on that site. All functional services will be concentrated on one server, and each at its level will wait for communication with other servers and sites from the configuration known to the server.

Two servers on one site is the smallest infrastructure on which availability can be ensured through redundancy of all roles. Therefore, alternative ways of distributing roles in the example reduce to changing the redundancy order or partially abandoning redundancy. That is, there are no alternatives.

When servers and communication are fully available, any WS instance is a fully functional API access point: this is Active-Active redundancy; in the diagram, role instances are shown in the same color. DC and MIC are made redundant differently: only one role instance is active on the site, and the second is activated when the first becomes unavailable. This is Active-Passive redundancy (an alternative concept is role takeover, with the related terms FailOver and TakeOver); in the diagram, active and passive (standby) instances are shown in different colors.

active passiveAll system roles are made redundant according to one of the principles described above (Active-Active or Active-Passive). In either case, there can be 2, 3, 10, or any number of servers with the same role, but while for Active-Active this makes sense for even load distribution, for Active-Passive roles it is hardly reasonable to have more than 3 standby servers for the same role, unless the task is to ensure availability when any 3 servers fail simultaneously.

Table 1. Terms used
Term Definition

MDC

!

SDC

!

DC

!

Object DB

!

Proxying a request to another site

!

Multi-site domain service

!