Resilience and Disaster Recovery Cheat Sheet

A vendor-neutral resilience and disaster recovery cheat sheet for RTO, RPO, high availability, backup, replication, standby patterns, failover, and testing.

Use this page for rapid architecture review after reading the full Resilience and Disaster Recovery guide. The objective is not to memorize product names. It is to identify the failure scenario, business tolerance, state-protection need, and operating evidence that determine a defensible design.

Core distinctions

Concept Primary question Does not prove
Availability Can users successfully use the service now? That data can be restored after corruption
Reliability Does the service behave correctly over time? That every large disaster is covered
Durability Will committed data remain intact? That the application remains online
Recoverability Can the required service and data be restored? That recovery is fast enough
High availability (HA) Can service continue through expected local failures? Regional or cyber-disaster recovery
Disaster recovery (DR) Can acceptable service be restored after a major disruption? Continuous service during normal failures
Fault tolerance Can the system continue through a defined failure with little interruption? Protection outside that defined failure scope
Resilience Can the system absorb, adapt to, and recover from disruption? That any single feature makes the whole system resilient

Decision order

  1. Name the business service or critical user flow.
  2. Name the failure scenario and scope: process, host, rack, zone, region, provider, dependency, operator error, corruption, or cyberattack.
  3. Define the acceptable service state: full service, degraded mode, read-only mode, or manual fallback.
  4. Set RTO and RPO for that flow.
  5. Select redundancy, replication, backup, and recovery patterns that address the named scenario.
  6. Include identity, keys, DNS, certificates, configuration, artifacts, network paths, and external dependencies.
  7. Test the complete path and retain evidence that the objective was met.

Starting with a vendor feature reverses the design process. Start with the outcome and failure instead.

Recovery objectives

Objective Meaning Design pressure
RTO Maximum acceptable time until the required service is restored Detection, decisions, automation, standby capacity, routing, validation
RPO Maximum acceptable interval of data loss in the recovered state Backup frequency, replication lag, transaction capture, reconciliation
MTD or MTPD Longest disruption the business can tolerate before impact becomes unacceptable Outer business limit for technical and business recovery
WRT Time after technical restoration to validate data and resume business work Reconciliation, integrity checks, process restart
MTTR Observed average time to restore or repair Operational evidence; not a guarantee for each event
Availability SLO Target successful service over a measurement window Normal reliability and error-budget decisions

RPO looks backward from the incident to the latest acceptable recoverable state. RTO looks forward from the disruption to validated service restoration.

HA, DR, backup, and replication

Mechanism Best at Weak when used alone
Local redundancy Component, host, rack, or zone failures Larger shared failures and corruption
High availability Continuing service through expected failures Historical recovery and severe disasters
Replication Maintaining another usable copy with low lag Deletion, corruption, or malicious changes that replicate
Backup Recovering an earlier point in time Immediate continuity and low RTO
Disaster recovery Restoring service after a defined large disruption Preventing every interruption
Immutable or isolated recovery copy Reversing destructive or unauthorized changes Fast service restoration unless the recovery path is prepared

Backup and replication are complementary. Replication can reduce recovery time and data loss for infrastructure failures. Versioned, immutable, and administratively isolated backups provide a path back before logical damage occurred.

Failure-scope ladder

Failure scope Typical protection pattern Hidden shared dependency to check
Process or instance Restart, replacement, health checks, stateless design Shared storage or configuration
Rack or local facility Redundant power, network, and placement Common switches, power, or control plane
Zone or data center Multi-zone service and data placement Single-zone egress, appliance, or state tier
Region or site Cross-region or cross-site recovery Identity, DNS, keys, artifact registry, decision authority
Provider or platform Portable data, alternate operating path, exit plan where justified Common identity, domain registrar, or software supply chain
Logical corruption or deletion Versioning, point-in-time recovery, protected backups Replication carrying damage forward
Ransomware or privileged compromise Immutable copies, administrative isolation, clean-room recovery Compromised keys, backup control plane, or recovery credentials

Redundant components that depend on the same fragile control plane are not independent. Draw dependencies, not only regions and zones.

Recovery strategy patterns

Pattern Normal state Relative cost Typical recovery profile Main risk
Backup and restore Recovery environment absent or minimal Lowest Slowest RTO; RPO depends on recoverable copies Rebuild and restore take longer than assumed
Pilot light Core stateful services or minimal foundation active Low to medium Faster than full rebuild Scale-up path, quotas, and configuration drift
Warm standby Smaller complete environment running Medium to high Faster RTO with continuous replication Standby capacity or dependencies not ready
Hot standby Full or near-full secondary environment ready High Rapid failover Cost and operational drift
Active-active Multiple locations serve production traffic Highest Potentially smallest interruption Data conflicts, split brain, routing, and operational complexity
Cyber recovery Isolated copies and clean recovery environment Varies Optimized for trusted restoration, not only speed Restoring compromised data or credentials

Choose the least complex pattern that demonstrably meets the target. Near-zero objectives have significant consistency, cost, and operational consequences.

Active topology chooser

Topology Use when Verify before choosing
Active-passive Simpler write ownership and failover are acceptable Promotion, fencing, capacity, traffic switch, stale clients
Active-active reads Read scale or local latency matters; writes can remain centralized Replica lag and read-after-write expectations
Active-active partitioned writes Ownership can be divided by tenant, key, or geography Rebalancing, ownership changes, hot partitions
Active-active shared writes The data model supports multi-writer conflict handling Consistency, conflict resolution, duplicate operations, split brain

Fencing is mandatory when two nodes might believe they are primary. Detecting failure is not enough; the old writer must be prevented from continuing before the replacement safely accepts writes.

Replication and consistency

Choice Benefit Cost or risk
Synchronous replication Tight data-loss objective within the protected scope Write latency and reduced availability during partitions
Asynchronous replication Lower latency and longer-distance practicality Replication lag and possible acknowledged-write loss
Single-writer topology Simpler ordering and conflict control Promotion and writer-location dependency
Multi-writer topology Local writes and broader active service Conflicts, convergence, and application complexity
Strong consistency Current agreed value before success Coordination latency and partition trade-offs
Eventual consistency Availability and geographic scale Stale reads and explicit conflict semantics

Do not infer an RPO of zero from the word replicated. Ask whether replication is synchronous, which writes are acknowledged, what happens during a partition, and how promotion handles queued or divergent changes.

Failover and failback sequence

  1. Detect a symptom and determine whether it represents a real failure.
  2. Assess scope, dependencies, data state, and security implications.
  3. Declare the incident and authorize recovery.
  4. Fence unsafe writers or isolate compromised systems.
  5. Restore, promote, or activate the recovery environment.
  6. Reconnect dependencies and move traffic deliberately.
  7. Validate data integrity, security controls, and critical business flows.
  8. Monitor the recovered service under production load.
  9. Reconcile data and decide whether failback is necessary.
  10. Treat failback as another risky migration with its own plan and validation.

Automatic failover is appropriate only when health signals are trustworthy and false positives are less dangerous than delayed recovery. Otherwise, automation should prepare and validate while an authorized operator makes the final decision.

What recovery testing must prove

Evidence Question it answers
Restore completion and integrity checks Can protected data actually be recovered?
Measured end-to-end recovery time Did the service meet RTO, not merely infrastructure startup time?
Recovered point or measured data loss Did the result meet RPO?
Critical-flow tests Can users complete the required business operations?
Identity, keys, DNS, and integration checks Did hidden dependencies survive?
Capacity and load results Can the recovery environment handle real demand?
Decision and communication record Were roles and authority usable under pressure?
Failback or permanent-migration plan Can the organization leave the recovery state safely?

Use multiple exercise types: document review, tabletop, component restore, partial failover, and full service exercise. A successful tabletop does not prove that data can be restored; a successful database restore does not prove that the business service can resume.

Common traps

Trap Corrective rule
“Multi-zone means disaster recovery” Multi-zone HA and cross-region recovery protect different failure scopes.
“Replication replaces backup” Replication can copy corruption and deletion; retain recoverable historical states.
“Durable storage means the application is available” Durability protects data; availability requires a functioning service path.
“The standby exists, so RTO is met” Measure detection, authorization, activation, routing, and validation end to end.
“RPO equals backup interval” Effective RPO depends on successful, consistent, accessible, restorable copies.
“Active-active is always best” It adds conflict, consistency, routing, and operating complexity.
“Failback is just reversing failover” Data divergence and renewed risk make failback a separate controlled migration.
“A plan is evidence” Only exercised recovery with measured results demonstrates capability.

Review path

Primary background references: NIST SP 800-34 Rev. 1, the Google Cloud disaster recovery planning guide, the Google SRE book, Azure reliability guidance, and AWS disaster recovery patterns.

Revised on Friday, September 11, 2026