AWS Backup: plans, vaults, copies, locks, and restores you have actually tested
Every service has its own snapshot button: RDS snapshots, DynamoDB backups, EBS snapshots, EFS backups. Pressed by hand they give you a different retention per team, no copies in another Region, and nobody who knows whether last night's backups ran. AWS Backup puts all of it behind one policy: plans say when and how long, selections say what (usually by tag), vaults hold the recovery points, copy actions send them to another Region or account, and Vault Lock stops anyone from deleting them early. A backup counts once a restore from it has been tested.
Need to know: a backup plan has rules: a schedule (cron, UTC), start and completion windows, a lifecycle (cold storage, delete after N days) and copy actions. A backup selection assigns resources to the plan, by tag or ARN, with the IAM role AWS Backup uses. A restore job always creates a new resource. Your RPO is the time between two recovery points (unless the rule uses continuous backup for point-in-time restore), and your RTO is how long a restore and the switch-over take.
What is protected, and where it goes
$ aws backup list-backup-plans --query 'BackupPlansList[].[BackupPlanName,BackupPlanId,LastExecutionDate]' --output text
lab-daily 0c92775b-91ef-4617-a132-062a3b8aa518 2026-09-22T05:00:00.000000+00:00
$ aws backup get-backup-plan --backup-plan-id $(aws backup list-backup-plans --query 'BackupPlansList[?BackupPlanName==`lab-daily`].BackupPlanId' --output text) --query 'BackupPlan.Rules[].{rule: RuleName, schedule: ScheduleExpression, vault: TargetBackupVaultName, lifecycle: Lifecycle, copies: CopyActions[].DestinationBackupVaultArn}'
[
{
"rule": "daily-35d",
"schedule": "cron(0 5 ? * * *)",
"vault": "lab-prod",
"lifecycle": {
"DeleteAfterDays": 35
},
"copies": [
"arn:aws:backup:eu-west-1:111122223333:backup-vault:lab-dr"
]
}
]
$ aws backup list-backup-selections --backup-plan-id $(aws backup list-backup-plans --query 'BackupPlansList[?BackupPlanName==`lab-daily`].BackupPlanId' --output text) --query 'BackupSelectionsList[].[SelectionName,IamRoleArn]' --output text
tagged-daily arn:aws:iam::111122223333:role/service-role/AWSBackupDefaultServiceRole
$ aws backup list-protected-resources --query 'Results[].[ResourceType,ResourceArn,LastBackupTime]' --output text
DynamoDB arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders 2026-09-22T05:02:30.000000+00:00
lab-daily runs at 05:00 UTC every day (cron(0 5 ? * * *)), keeps recovery points 35 days in the vault lab-prod, and copies each one to lab-dr in eu-west-1, where the copy is kept 90 days. Its selection takes every resource tagged backup=daily, which is how a new database gets protected without anyone editing the plan. A resource that is not tagged is not protected, and nothing tells you. AWS Config rules and Backup Audit Manager reports exist to catch that.
Vaults and recovery points
$ aws backup describe-backup-vault --backup-vault-name lab-prod --query '{key: EncryptionKeyArn, points: NumberOfRecoveryPoints, locked: Locked, min: MinRetentionDays, max: MaxRetentionDays}'
{
"key": "arn:aws:kms:eu-central-1:111122223333:key/84e7e9e3-d54a-4b27-b155-2f25306535f9",
"points": 2,
"locked": false,
"min": null,
"max": null
}
$ aws backup list-recovery-points-by-backup-vault --backup-vault-name lab-prod --query 'RecoveryPoints[].[ResourceType,Status,CreationDate,CalculatedLifecycle.DeleteAt]' --output text
DynamoDB COMPLETED 2026-09-22T05:00:00.000000+00:00 2026-10-27T05:02:30.000000+00:00
DynamoDB COMPLETED 2026-09-21T05:00:00.000000+00:00 2026-10-26T05:02:30.000000+00:00
$ aws backup list-recovery-points-by-backup-vault --backup-vault-name lab-dr --region eu-west-1 --query 'RecoveryPoints[].[ResourceType,Status,CreationDate]' --output text
DynamoDB COMPLETED 2026-09-22T05:03:00.000000+00:00
DynamoDB COMPLETED 2026-09-21T05:03:00.000000+00:00
A vault is a container with its own KMS key and access policy. For RDS, a recovery point is a DB snapshot (awsbackup:job-...); for DynamoDB, a table backup; for EBS, a snapshot. Last night's copies are in eu-west-1 too: if Frankfurt became unusable, these are what you would restore from.
An on-demand backup, and a restore
Before a risky migration, take a backup now instead of waiting for 05:00:
$ aws backup start-backup-job --backup-vault-name lab-prod --resource-arn arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders --iam-role-arn arn:aws:iam::111122223333:role/service-role/AWSBackupDefaultServiceRole --lifecycle DeleteAfterDays=14 --query BackupJobId --output text
13BF16D6-21B7-4774-9B26-C2E2CA92976D
$ aws backup list-backup-jobs --by-state CREATED --query 'BackupJobs[].[BackupJobId,State,ResourceType]' --output text
13BF16D6-21B7-4774-9B26-C2E2CA92976D CREATED DynamoDB
$ sleep 70
$ aws backup list-backup-jobs --by-resource-arn arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders --query 'BackupJobs[0].[State,PercentDone]' --output text
RUNNING 11.0
$ sleep 100
$ aws backup list-backup-jobs --by-resource-arn arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders --query 'BackupJobs[0].[State,PercentDone,RecoveryPointArn]' --output text
COMPLETED 100.0 arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders/backup/1790107264000-05c90025
A job goes CREATED -> RUNNING -> COMPLETED (or FAILED, or EXPIRED if it could not start inside the start window). Now restore the newest DynamoDB recovery point into a new table. The restore metadata describes what can be set for the new resource:
$ RP=$(aws backup list-recovery-points-by-resource --resource-arn arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders --query 'RecoveryPoints[0].RecoveryPointArn' --output text)
$ aws backup get-recovery-point-restore-metadata --backup-vault-name lab-prod --recovery-point-arn $RP --query RestoreMetadata
{
"targetTableName": "shop-orders",
"encryptionType": "Default"
}
$ aws backup start-restore-job --recovery-point-arn $RP --iam-role-arn arn:aws:iam::111122223333:role/service-role/AWSBackupDefaultServiceRole --metadata targetTableName=try-orders-restored --query RestoreJobId --output text
5212049A-F266-414A-81A7-6049AA782327
$ sleep 130
$ aws backup list-restore-jobs --query 'RestoreJobs[0].[Status,CreatedResourceArn]' --output text
COMPLETED arn:aws:dynamodb:eu-central-1:111122223333:table/try-orders-restored
$ aws dynamodb describe-table --table-name try-orders-restored --query 'Table.[TableStatus,ItemCount]' --output text
ACTIVE 33
That is a restore test, done by hand. A test checks three things: the data is there (count the items, run the application's own checks), the time it took (your measured RTO), and the steps (roles, metadata, security groups, the switch-over) are written down. AWS Backup's restore testing plans run restores like this on a schedule, validate them and delete the test resources, so "we have backups" becomes "we restored one last Tuesday in 11 minutes".
Locks: backups an attacker cannot delete
$ aws backup put-backup-vault-lock-configuration --backup-vault-name lab-prod --min-retention-days 7 --max-retention-days 365
$ aws backup describe-backup-vault --backup-vault-name lab-prod --query '[Locked,MinRetentionDays,MaxRetentionDays,LockDate]' --output text
True 7 365 None
$ aws backup delete-recovery-point --backup-vault-name lab-prod --recovery-point-arn $(aws backup list-recovery-points-by-backup-vault --backup-vault-name lab-prod --query 'RecoveryPoints[0].RecoveryPointArn' --output text)
aws: [ERROR]: An error occurred (InvalidRequestException) when calling the DeleteRecoveryPoint operation: Recovery point arn:aws:dynamodb:eu-central-1:111122223333:table/shop-orders/backup/1790107264000-05c90025 cannot be deleted: backup vault lab-prod is locked and the recovery point is younger than its MinRetentionDays (7).
Vault Lock enforces a minimum and maximum retention on every recovery point in the vault and refuses early deletes. Without --changeable-for-days it is governance mode: a user with the right IAM permissions can still remove the lock. With --changeable-for-days N (at least 3) it becomes compliance mode once those days have passed: nobody, not the root user and not AWS Support, can remove it or delete a recovery point early. That is the backup equivalent of S3 Object Lock compliance mode, and the reason ransomware groups target backups first. Two more layers: cross-account copies (to a vault in a separate backup account, which the organization's backup policy can enforce) and a vault access policy that denies backup:DeleteRecoveryPoint.
RPO, RTO and the plan
| requirement | what the plan needs |
|---|---|
| RPO 24 h | a daily rule |
| RPO 1 h | an hourly rule (cron(0 * ? * * *)), or continuous backup |
| RPO minutes | continuous backup (EnableContinuousBackup: point-in-time restore for RDS, S3, Aurora, SAP HANA), or the database's own PITR / replicas |
| survive a Region | copy actions to another Region (and restore there) |
| survive a compromised account | copies to another account + Vault Lock (compliance) |
| keep 7 years | lifecycle to cold storage, DeleteAfterDays 2555 |
| RTO | measured by restore tests, not promised by the plan |
Which DR strategy fits which RPO and RTO (backup and restore, pilot light, warm standby, multi-site active-active) is the design question of AWS VI. Backups are the floor under all of them.
On the exam (SOA-C03 2.3, SAA-C03 2.2, CLF-C02 3.6): "centrally manage backups across EBS, RDS, DynamoDB, EFS..." = AWS Backup (plan + tag-based selection). "Backups in another Region or account" = copy actions. "Prevent anyone, including root, from deleting backups" = Vault Lock in compliance mode. "Point-in-time restore through AWS Backup" = continuous backup. "Verify that backups can be restored" = restore testing. A restore always creates a new resource.
In an interview: "How do you know your backups work?" - "Because restores are tested on a schedule: AWS Backup restore testing (or our own job) restores the latest recovery points into a sandbox, runs data checks, records how long it took, and alarms if it fails. Plus alarms on failed or expired backup jobs, a report of resources without a backup plan, and copies in a separate account that the production admins cannot delete."
You can now: read a backup plan's rules, selections, lifecycle and copy actions, find the recovery points in a vault (and in the DR Region), run an on-demand backup and a restore job and treat it as a restore test, lock a vault, and derive the RPO a plan really gives.