Easy ReviewSAA-C03
Design Resilient Architectures16

Disaster recovery

Set recovery targets, choose a recovery strategy, and rehearse failover and failback.

Este conteúdo de estudo ainda não foi traduzido, então aparece em inglês. Ajude a traduzir

Disaster recovery prepares a workload for a serious disruption. Start with acceptable downtime and data loss, then prepare and test the data, application resources and traffic changes needed to restore service.

RTO and RPO

Disaster recoveryThe preparation and actions used to restore a workload after a serious disruption. restores a workload after a serious disruption. Recovery time objectiveThe longest service interruption a business is prepared to accept. (RTO) is the downtime the business can accept. Recovery point objectiveThe oldest recoverable data a business can accept at interruption, measured as a time gap. (RPO) is the data loss it can accept, expressed as time.

How it works

  1. Set both objectives from business impact, dependencies and compliance needs.
  2. Choose a backup interval or replication design that can preserve recent enough data. Asynchronous replicationCopying changes to another data store after the source accepts them. copies changes after the source accepts them, so some changes can still be in transit when it fails.
  3. Choose a recovery procedure that restores service within the downtime target.
  4. Test the procedure. Compare the time until service returns with RTO and the age of recoverable data with RPO.

Exam cues

  • If the question says data loss must stay within a stated interval, think RPO, because it limits the age of the recovery point.
  • If the question says service must return within a stated interval, think RTO, because it limits downtime.
  • If the question says asynchronous replication guarantees no lost writes, think replication lagThe delay between a source change and its arrival at a replica., because unreplicated changes can be lost.

Remember

  • RTO measures time forward from interruption to restoration.
  • RPO measures the acceptable gap between the recoverable data and interruption.
  • Backup frequency affects the recoverable point. A configured schedule still needs a recovery test.

Trap: A low RPO does not prove fast service restoration. Data freshness and recovery duration are separate targets.

Da documentação da AWS (8)
  • Recovery time objective (RTO) is the longest acceptable interval from service interruption to service restoration.
  • Recovery point objective (RPO) limits the acceptable age of the last recoverable data when service is interrupted.
  • Set recovery objectives from business impact, cost, dependencies and compliance requirements.
  • Backup frequency determines the recoverable point for a backup-based recovery plan.
  • Aurora Global Database asynchronous replication can lose unreplicated changes during unplanned failover; loss depends on replication lag.
  • Test recovery paths to show that the workload recovers within its time target and to the required data state.
  • RTO constrains downtime while RPO constrains lost data; choosing a stricter target than the business needs adds cost and complexity.
  • Disaster recovery prepares a workload to respond to a serious disruption using business objectives for downtime and data loss.

Recovery strategy selection

A disaster recoveryThe preparation and actions used to restore a workload after a serious disruption. strategy determines how much of a workload is ready at the recovery site before a failure. Choose it from acceptable downtime, acceptable data loss and the cost of maintaining that readiness.

How it works

  1. Set recovery time objectiveThe longest service interruption a business is prepared to accept. (RTO), the acceptable downtime, and recovery point objectiveThe oldest recoverable data a business can accept at interruption, measured as a time gap. (RPO), the acceptable data loss in time.
  2. Select the preparation level below. Infrastructure as codeInfrastructure definitions stored as templates that can deploy the required resources. (IaC) means templates that can recreate infrastructure; keep application code, configuration and machine images available too.
  3. Test data recovery and the application’s ability to handle the failoverThe move of production traffic from the primary environment to a recovery environment. load. The ranges below describe patterns, not a guaranteed workload result.
Strategy Ready before failure Recovery action Indicative RPO / RTO
Backup and restore Backups and deployment material Rebuild and restore Hours / up to a day
Pilot lightA recovery design with data and core infrastructure ready but more resources needed to process requests. Data replication and core infrastructure Deploy or start the missing application resources, then scale Minutes / tens of minutes
Warm standbyA working recovery deployment kept at reduced capacity until it must take over. Working workload at reduced capacity Scale up Seconds / minutes
Multi-site active-activeA design in which workloads in multiple Regions serve traffic during normal operation. Workloads serving traffic in multiple RegionsA separate geographic area containing multiple isolated Availability Zones. Use surviving Regions Toward zero / toward zero, at the highest cost

Hot standbyA warm standby scaled to serve the full production load. is a warm standby scaled to serve the full production load. Multi-site active-active also needs data synchronization and a way to handle concurrent write conflicts. DynamoDB global tables accept regional writes. Their default multi-Region eventual consistency (MREC) mode copies changes asynchronously and resolves concurrent changes with last writer winsConflict handling that retains the item update with the latest internal timestamp., which keeps the item version with the latest internal timestamp. Aurora global database write forwarding sends secondary writes to the primary cluster.

Exam cues

  • If the question says the recovery copy cannot process requests until application resources start, think pilot light, because only core resources are ready.
  • If the question says a small working stack already serves requests, think warm standby, because recovery mainly needs more capacity.
  • If the question says each Region must accept database writes, think DynamoDB global tables, because Aurora write forwarding still changes data at its primary.
  • If the question says cost is tight and rebuilding meets the targets, think backup and restore, because a stricter strategy adds unnecessary cost.
  • If the question says lowest-cost DR where only the data and server images may run in the recovery Region, think pilot light with infrastructure as code, because servers are deployed from templates only at failover.

Remember

  • Warm standby becomes hot standby when fully scaled and serving the full production load.
  • Aurora unplanned recovery can take minutes and lose unreplicated changes. RDS read replicaA read-only database copy that receives changes from a source database. promotion includes a reboot and can take several minutes or longer.
  • Measure full recovery, including data, application capacity and traffic movement.

Trap: A strategy’s typical RTO is not a promise. A database promotion time alone does not establish the application’s RTO.

Failover to a recovery Region (pilot light or warm standby)
Primary RegionRecovery Regionasync replication1normal2failover3App fleetfull capacityiPrimary databaseread/writeiRoute 53 / ARCfailover routingiStandby appoff or scaled downiReplica databasealways oni
  1. Replicate data continuously while keeping core infrastructure ready in the recovery Region.

  2. During normal operation, Route 53 directs clients to the primary Region.

  3. On failover, promote the replica, scale up the standby fleet and shift traffic to the recovery Region.

Detalhes das caixas (5)
App fleet

Serves all traffic during normal operation.

Primary database

The primary database accepts writes. This example uses asynchronous replication to a recovery Region.

Route 53 / ARC

Route 53 health checks can fail over automatically. Application Recovery Controller routing controls give a data plane on/off switch for manual failover.

Standby app

Pilot light: servers are not running, only core pieces exist, so you start and scale them. Warm standby: a small working fleet already runs, so you only scale up.

Replica database

The replica receives changes asynchronously. Unreplicated changes can be lost during failover. Keep point-in-time backups as well.

Da documentação da AWS (24)
  • Backup and restore keeps backups and redeploys infrastructure, configuration and application code during recovery.
  • Infrastructure as code (IaC) defines infrastructure in deployable templates; recovery also needs code, configuration and machine images.
  • Pilot light keeps replicated data and core resources available but needs additional actions before it can process requests.
  • Warm standby keeps a working workload at reduced capacity and scales it up for recovery.
  • A warm standby that is fully scaled and serving full production load is a hot standby.
  • Multi-site active-active serves traffic from multiple Regions and must handle data synchronization and write conflicts.
  • Select a recovery strategy that meets business recovery objectives without unnecessary cost or complexity.
  • AWS illustrates backup and restore with RPO in hours and RTO of up to a day; these are indicative strategy ranges.
  • AWS illustrates pilot light with RPO in minutes and RTO in tens of minutes.
  • AWS illustrates warm standby with RPO in seconds and RTO in minutes.
  • AWS illustrates multi-site active-active as pushing RPO and RTO toward zero at the highest operational cost.
  • Aurora Global Database recovery can take minutes, with potential data loss determined by asynchronous replication lag.
  • RDS read replica promotion reboots the instance and can take several minutes or longer, depending on replica size.
  • DynamoDB global tables accept writes in replica Regions and copy changes to other replicas.
  • DynamoDB global tables default to multi-Region eventual consistency (MREC), which uses asynchronous replication and last-writer-wins conflict resolution.
  • Aurora global database write forwarding sends writes from secondary clusters to the primary cluster; data changes first at the primary.
  • Recovery testing must establish that the secondary site can handle the failover load and meet the recovery targets.
  • RTO is the maximum acceptable downtime between interruption and service restoration.
  • RPO is the maximum acceptable time gap between the last recoverable data and interruption.
  • Route 53 health checks and ARC routing controls provide failover through data plane operations.
  • Keep point-in-time backups or versioned data in the recovery site because continuous replication can copy destructive changes.
  • In pilot light, data stores replicate continuously to the recovery Region while application servers stay switched off; AWS recommends not deploying them at all and deploying them from infrastructure as code such as CloudFormation when failover starts.
  • For EC2 in pilot light, golden AMIs, for example from an EC2 Image Builder pipeline, are copied to the recovery Region ahead of time so servers can be launched there.
  • Failing over to a cross-Region RDS read replica means promoting it to the primary; for engines other than Aurora this takes a few minutes and includes a reboot.

Flashcards

toque para virar
Resposta

·

Todos os 16 flashcards em lista

What do RPO and RTO measure in AWS disaster recovery?

RPO limits lost data in units of time. RTO limits how long service can remain unavailable.

Does asynchronous cross-Region replication in Aurora Global Database guarantee zero data loss during an unplanned outage?

No. Changes not yet replicated can be lost; inspect replication lag when assessing recovery.

How should a business set RTO and RPO for an AWS workload?

Use business impact and recovery requirements, including dependencies, cost and compliance. Then test the recovery path.

An AWS application replicates its data frequently but takes hours to rebuild. Does a low RPO prove a low RTO?

No. RPO limits data loss; RTO limits downtime. Measure the entire recovery path.

An AWS workload requirement limits the age of recoverable data and the delay until service returns. Which recovery objective corresponds to each limit?

RPO limits the age of recoverable data; RTO limits the delay until service returns. Test both against the business targets.

An AWS workload resumes service within its downtime target but restores data older than its permitted loss window. Which objective did it miss?

RPO. Meeting RTO does not establish that the recovered data is recent enough.

How do AWS disaster recovery strategies progress from lower cost to faster typical recovery?

Backup and restore, pilot light, warm standby, then multi-site active-active. These patterns trade cost and complexity for lower typical RTO and RPO; measure the workload.

What must an AWS backup-and-restore recovery plan preserve besides application data?

Infrastructure templates, application code, configuration and machine images needed to redeploy the workload.

What separates AWS pilot light from warm standby disaster recovery?

Warm standby can process traffic immediately at reduced capacity. Pilot light needs additional resources started or deployed first.

How do warm standby, hot standby and multi-site active-active differ in AWS disaster recovery?

Warm standby runs at reduced capacity. Hot standby has full capacity ready. Multi-site active-active serves traffic from multiple Regions during normal operation.

How should a recovery plan compare Aurora Global Database failover with RDS read replica promotion?

Aurora recovery can take minutes and can lose unreplicated changes. RDS promotion reboots the replica and can take several minutes or longer. Test the complete workload rather than assuming a fixed duration.

How do DynamoDB global table writes differ from Aurora global database write forwarding?

DynamoDB replicas accept writes in their own Regions. Aurora secondary clusters forward writes to a single primary cluster. Default DynamoDB MREC tables resolve concurrent conflicts using last writer wins.

An AWS workload has a tight recovery budget and can tolerate rebuilding from backups. Which disaster recovery pattern should the team assess first?

Backup and restore. Preserve the recovery data, infrastructure templates and application configuration, then test that rebuilding meets both objectives.

An AWS application needs a working secondary Region before a failure, but cannot justify full standby capacity. Which disaster recovery pattern fits?

Warm standby. Keep a functioning copy at reduced capacity and test scaling it to the failover load.

An AWS active-active application loses a Region. Is recovery complete simply because the other Region was already serving traffic?

No. Verify that the surviving deployment can handle the failover load and that data synchronization and write conflict handling meet requirements.

A company wants the lowest-cost DR for an EC2 web tier and an RDS database whose RTO allows minutes to rebuild servers. What should exist in the recovery Region before a disaster, and what happens at failover?

Keep a cross-Region RDS read replica and copies of the golden AMIs there. At failover, deploy the web tier from a CloudFormation template, promote the replica and point DNS at the new stack.

Você concluiu todos os tópicos da SAA-C03!

Seu badge ficou dourado. Próximo passo: agendar a prova na AWS. Depois de passar, adicione-se ao Quem passou.

Compartilhar

Conte que você concluiu esta certificação

Concluí todos os tópicos da AWS Certified Solutions Architect - Associate (SAA-C03) no Easy Review. Venha estudar também!