Do you really need three full data centers to build a resilient database architecture?
Not necessarily.
For many organizations, a two-site, three-data-center design can provide a practical starting point: local high availability handles infrastructure failures, while a remote site provides disaster recovery.
With GBase Database (GBase 8s), the architecture can be designed in three layers.
1. Local HA: Protect Against Hardware Failure
The primary data center can use a shared-storage cluster to protect against local server or database-node failures.
Because multiple nodes access the same persistent data, failover does not depend on copying the entire dataset to another node.
For appropriately configured deployments, this architecture can support stringent recovery objectives such as RPO=0 and RTO in the seconds range.
The key point is that local HA and disaster recovery solve different problems.
2. Remote DR: Prepare for Site-Level Failures
Local HA cannot protect against a complete data-center outage.
A second site in another location provides an additional layer of protection. Data can be replicated asynchronously to the remote DR environment.
The trade-off is straightforward:
Asynchronous replication introduces a recovery-point window.
The actual RPO depends on replication latency, network conditions, workload characteristics, and the recovery design.
This means the remote DR environment should be sized around business recovery objectives—not simply copied from the primary environment.
3. Witness: Prevent Split-Brain
A third location does not necessarily need to be another full-scale data center.
A lightweight witness node can participate in cluster arbitration when network isolation creates uncertainty about which site should remain active.
Its role is not to store the production dataset. Its role is to help the system make a consistent decision about cluster ownership.
This can make the third location considerably lighter than a complete production site, depending on the HA and DR architecture.
Why Start Lightweight?
A common mistake in disaster recovery planning is designing the final architecture before understanding the actual business requirements.
A more practical approach is to build in layers:
Local HA → Remote DR → Multi-site expansion
Each layer addresses a different failure scenario and can be evaluated against specific RPO, RTO, cost, and operational requirements.
For GBase Database (GBase 8s) deployments, the goal isn't simply to maximize the number of data centers.
The goal is to build the simplest architecture that meets the business's recovery objectives, then add redundancy as those requirements evolve.
