Halo! Saya Ask Owen. Tanyakan tentang proyek, pengalaman, skill, pendidikan, atau cara menghubungi Owen.
Architecture
Disaster Recovery Is More Than Backups
A recovery plan becomes credible only when it connects dependencies, RTO, RPO, ownership, sequencing, and evidence from validation.
A backup is an input to recovery. It is not proof that a business service can return within the time and data-loss boundaries people expect.
Recovery planning becomes useful when it connects the application, data, infrastructure, identity, networking, external services, operational owners, and the order in which those pieces must return.
Begin with the business capability
Infrastructure inventories are necessary, but they are not the best starting point. I begin with the capability that must be restored and work backward through its dependencies.
For each capability, the map should answer:
- who owns the recovery decision;
- which user journey or operation matters;
- which applications and data stores support it;
- which shared services must be available first;
- which external dependencies can block validation;
- what evidence proves the capability is usable.
This prevents a plan from declaring success because individual servers are online while the end-to-end service is still unavailable.
RTO and RPO describe different constraints
Recovery Time Objective defines the target time for restoring a service. Recovery Point Objective defines the acceptable amount of data loss measured in time.
They influence different design choices. A shorter RTO can require faster failover, prepared capacity, and rehearsed decisions. A shorter RPO can require more frequent replication, different storage behavior, and stronger consistency controls.
Build a dependency graph, not a flat list
A flat asset list hides recovery order. A dependency graph exposes it.
One simplified sequence might be:
- Establish network, identity, name resolution, and access prerequisites.
- Restore the data layer and verify integrity.
- Restore shared platform services.
- Restore application components in dependency order.
- Reconnect external integrations where appropriate.
- Validate the business capability through a real transaction or controlled test.
The exact sequence varies, but the principle holds: every component should have a reason for its place in the order.
Assign decisions, not only tasks
Recovery work contains decisions that cannot be automated away. Someone must decide whether to declare an incident, start failover, accept degraded operation, or return to the primary environment.
A practical runbook names:
- the role authorized to initiate recovery;
- the technical owner for each dependency;
- the communication path;
- the validation owner;
- the criteria for continuing, pausing, or rolling back;
- the person who records deviations from the plan.
Clear ownership is a reliability mechanism. It reduces hesitation at the moments when time and information are limited.
Validate assumptions in layers
Not every validation needs to be a full failover. A useful program can combine several layers:
- restore tests for backups and snapshots;
- configuration and access checks;
- dependency walkthroughs;
- component recovery exercises;
- application-level validation;
- planned end-to-end recovery tests.
Each layer should produce evidence: timestamps, observed gaps, manual workarounds, unexpected dependencies, and follow-up owners.
Measure the recovery path without inventing confidence
Useful observations include the time spent at each checkpoint, how much manual coordination was required, which assumptions failed, and whether restored data met the intended point.
If an organization cannot publish exact figures, the public lesson can still be concrete: identify the dependency that changed the sequence, the validation that caught a gap, or the manual step that should become automation. Qualitative evidence is stronger than a fabricated percentage.
Keep the plan alive
Architecture, ownership, credentials, integrations, and data volumes change. A recovery document that is not reviewed and exercised becomes a description of an environment that no longer exists.
I treat recovery planning as a loop:
- Map the capability and dependencies.
- Define objectives and ownership.
- Design the recovery sequence.
- Validate a bounded part of the plan.
- Record evidence and gaps.
- Update the architecture and runbook.
The central idea is simple: recovery is a system behavior. Backups matter, but credible recovery requires an understood sequence, accountable decisions, and evidence that the intended service can actually return.
