zero gravitySOLUTIONS
← All insightsARCHITECTURE / ARCHITECTURE DECISION GUIDE

Backup recovery or a second live region.

QUICK ANSWER

A second Region is justified when tested recovery evidence shows that its preparation closes a real gap. Compare whole-application recovery and writer authority, and retain backups because replication can also copy destructive changes.

Inside this article 7 sections

Choose between backup recovery and a second live Region by testing the whole application's recovery objective, not only database availability. A second Region reduces some preparation work during an incident, but it adds replication, capacity, routing, and ownership decisions before the incident begins.

The central questions are how much accepted data may be unavailable or lost, how long the product may remain disrupted, and how the team prevents two locations from making conflicting authoritative writes.

Define the failure the design must survive

Distinguish a process crash, an Availability Zone failure, a regional disruption, and a destructive application change. They do not require identical remedies. A second live Region can replicate an accidental deletion or corrupt business update, so it does not replace protected backups and point-in-time recovery.

For an illustrative B2B approval application, define recovery around accepted approvals and their audit records. The user interface can be restored quickly while recent approvals remain missing. That is not a successful recovery if those approvals are the product's authoritative business evidence.

Write recovery time and recovery point objectives for the application, then identify the data and dependencies required to satisfy them. Do not borrow a provider's availability statement and present it as the product's recovery result.

Original recovery choice matrix · whole-application responsibilities
ResponsibilityBackup and restoreWarm secondary Region
Application capacityProvision and validate during recoveryMaintain and scale an existing deployment
Data readinessRestore to an available recovery pointInspect replication state before promotion
Write authorityPrevent old primary from resuming conflicting workFence old writer before secondary takes authority
Ongoing effortBackup integrity and restore rehearsalsReplication, drift, capacity, and failover rehearsals
The second Region moves work earlier in time; it does not remove the need to validate data, authority, and customer behavior.

Compare preparation levels without promising recovery times

Backup and restore keeps recoverable data and the means to rebuild the application. A pilot-light design keeps selected core components ready while other capacity is created or activated during recovery. Warm standby keeps a functional but potentially smaller deployment available. Active-active operation serves traffic from multiple locations and needs a more demanding consistency design.

These categories describe preparation, not guaranteed recovery durations. The actual time depends on data volume, capacity availability, automation, access, and validation. Measure the chosen design through a rehearsal rather than assigning it a recovery time from a generic comparison chart.

Start with the least complex design that can demonstrate the required result. If restore testing misses the objective because provisioning is slow, maintaining selected infrastructure may help. If data reconciliation dominates, another running application fleet may not address the main delay.

Make write authority explicit

During failover, decide who can accept new approvals and how the old primary is prevented from continuing. Routing users away does not necessarily stop background workers, scheduled jobs, or existing connections in the old location.

Use an authority mechanism appropriate to the data platform and failure model. It may involve database promotion, fencing tokens, disabled credentials, or an application control plane. The mechanism must remain usable during the failure being considered and should not depend solely on a successful call to the impaired Region.

Define what happens to operations whose outcome is uncertain at the transition. Preserve operation identifiers and reconcile committed effects. Reissuing every pending approval or payment in the secondary location can duplicate actions that succeeded before the primary became unreachable.

Inventory regional and external dependencies

Include queues, object storage, secrets, encryption keys, images, logging, identity integration, scheduled jobs, and outbound provider configuration. A replicated database does not automatically replicate queued work or guarantee that the secondary role can read the objects referenced by its rows.

Check external allowlists and callback URLs. A provider may permit only the primary egress address or send webhooks to a route that still terminates there. Test the secondary integration path before relying on it during an incident.

Keep the standby configuration current and inspect drift. A deployment that has not run a representative customer workflow for months may be missing a schema migration or secret refresh. The credential rotation guide applies to both primary and recovery consumers.

Rehearse promotion and return

In a controlled environment, simulate loss of the primary path. Record detection, decision, authority transfer, capacity activation, routing, data checks, and customer workflow validation as separate milestones. Include a recently accepted operation whose completion is ambiguous.

Verify the recovered application can read old data, accept a new write, complete background work, and enforce account permissions. Observe whether delayed messages from the old location can arrive after promotion and whether they remain safe to process.

Plan the return path too. Returning to the original Region is another migration with data direction and writer ownership decisions. Avoid a simplistic “switch DNS back” instruction after the secondary has accepted new authoritative writes.

Include failover capacity reservations or quota readiness in the review where the workload needs them. A standby that can run a small health check may still lack the capacity to serve the full customer population after promotion, especially during a wider disruption.

Buy a recovery assessment before a second operating footprint

A useful first engagement defines the failure scenario, the business recovery objectives, and the dependency boundary before recommending a second Region. Bring a representative customer workflow and identify the data and external systems it needs. Agree whether the assessment includes a controlled recovery exercise, the design of that exercise, or only a review of existing evidence. Those scopes support different levels of confidence.

The expected output is a recovery dependency map, observed results from any executed exercise, a record of manual interventions, and options for closing the gaps. Ask the team to include write authority and the return procedure, not only traffic redirection. Where a dependency cannot be exercised, the limitation should remain in the acceptance record. An untested provider assumption must not disappear into an overall recovery-time claim.

A restore improvement may satisfy the business objective with less ongoing work. A standby design may be justified when the measured gap remains unacceptable. We can help compare those paths and scope the next implementation around the missing capability. The buyer should leave knowing what was demonstrated, what operating responsibility each option adds, and which subsequent exercise would establish readiness.

Choose from observed recovery evidence

Report the tested failure, observed recovery point, total recovery time, manual interventions, and unresolved dependencies. Compare those results with the business objective. A warm secondary is justified when it closes a demonstrated gap at an acceptable operating cost.

The PostgreSQL restore protocol provides the database-level foundation. Architecture design can extend that evidence into whole-application recovery, including routing, write authority, and a return procedure that preserves the work customers entrusted to the product.

FROM INSIGHT TO ACTIONSoftware architecture consulting

Choose recovery architecture from evidence you can inspect.

We can map the application dependencies, design or run a controlled exercise, and turn the findings into a practical recovery implementation plan.

Explore architecture consultingDiscuss your architecture

Start with your goals.
We’ll work through the next step together.