Home › AWS › AWS data services › AWS Backup
AWS data services · Observability, audit and cost · about 10 min read
AWS Backup
One place to schedule, keep, copy and lock backups of AWS databases, volumes and buckets, with restore tests and audit reports.
In plain words
What it is
A backup is a copy of data kept somewhere safe, so that you can get the data back after a mistake, a failure or an attack. AWS services each have their own backup feature. RDS takes snapshots, DynamoDB has point-in-time recovery, and EBS volumes have snapshots. AWS Backup puts all of these behind one set of rules. You write a backup plan once, say which resources it covers, usually by tag, and AWS Backup takes the copies, keeps them as long as you said, moves old ones to cheaper storage and deletes them when they expire.
Backups land in a backup vault. A vault is a container with its own encryption key and its own access policy, separate from the source resource. If someone deletes the database, the backups in the vault remain. Vault Lock goes further. Once it is set in compliance mode, nobody, including the account's administrators, can delete backups early or shorten how long they are kept. That is the write-once-read-many model that ransomware defence and many regulations ask for.
A concrete example: a retail order platform runs its order database on Aurora PostgreSQL and its shopping-cart table on DynamoDB. A backup plan tagged backup=tier1 takes a backup every hour and keeps it for 35 days, keeps a monthly copy for seven years, and copies every backup to a locked vault in a separate backup account in another region. A restore test runs every week. It restores last night's Aurora backup to a test cluster, runs a row count, and records how long the restore took.
AWS Backup does not understand table formats. It backs up an S3 bucket as objects and versions, not as Delta tables with a transaction log. For lakehouse tables, Delta's own history, deep clone and S3 replication are the tools, and this page explains why the reference platform keeps personal data out of AWS Backup.
A bank's safe-deposit room with a time lock: you set the rules for what goes in and how long it stays, and once locked even the manager cannot take it out early.
Big picture
Where it fits
The lit lane is where AWS Backup works. Boxes on the left feed it; boxes on the right use what it produces.
AWS Backup sits in the band across the bottom of the picture, with observability, audit and cost, because it is an operations control that touches many services rather than a step data flows through. On the reference platform its scope is narrow on purpose.
The platform's own resilience plan for lakehouse data uses Delta time travel, deep clone of gold and the token vault to a second region, and S3 replication, with a written and tested restore drill. AWS Backup is not used for the token vault or for personal-data tables. The reason is erasure. When a person asks to be forgotten, the platform deletes their token and VACUUM removes old files within 30 days. A locked backup that keeps an old copy for seven years would hold that person's data out of reach of the erasure process. Every restore would also need the erasure log replayed straight after it.
Where AWS Backup fits well is the operational side of the platform. That includes app databases on RDS or Aurora that feed DMS, DynamoDB tables in the fast lane, EBS volumes on any self-managed servers, and configuration such as CloudFormation stacks. These are the systems whose loss would stop data flowing in. Note that the platform's metadata database runs on Databricks Lakebase, so it is protected by Databricks features, not AWS Backup.
How it works
The ideas you need
| Idea | What it means |
|---|---|
| Backup plan | A policy that says how often to back up, how long to keep each copy, when to move it to cold storage and where to copy it. |
| Resource assignment | Which resources a plan covers, usually chosen by tag so new databases are protected automatically. |
| Backup vault | A container for backups with its own KMS key and access policy, separate from the source resource. |
| Vault Lock | A setting that stops anyone deleting backups early or shortening retention. In compliance mode it cannot be undone after a grace period. |
| Logically air-gapped vault | A vault type that is locked by default and can be shared to another account for recovery even if the source account is compromised. |
| Point-in-time recovery | Continuous backup that lets you restore to any second within a window, for example the last 35 days, instead of only to a snapshot time. |
| Restore testing | A scheduled job that restores backups automatically and checks them, so you know restores work and how long they take. |
| Backup Audit Manager | Built-in controls and daily reports that show which resources are, and are not, backed up as the policy says. |
| Cold storage tier | Cheaper storage for older backups, with a minimum stay before deletion, available for some resource types. |
Choosing
When to use it, and when not
Use it when
- You run app databases on RDS, Aurora or DynamoDB and need backups on a schedule, with copies in another account and region.
- An auditor or regulator asks for proof that every tier-1 system is backed up and that backups cannot be deleted early.
- You have many accounts in AWS Organizations and want one backup policy applied to all of them.
- You need ransomware protection: immutable copies in a separate account that an attacker in the main account cannot reach.
- You want restores tested automatically on a schedule, with the time to restore recorded as evidence.
Avoid it when
- The data is personal and subject to erasure requests, and a long-locked backup would keep it beyond what the erasure process can reach.
- The data is a Delta or Iceberg table that you need restored as a consistent table; use table history, deep clone and replication instead.
- The bucket is a raw landing area that is deleted after a day by design; backing it up undoes the design.
- The data can be rebuilt cheaply from a source that is already protected, such as silver rebuilt from append-only bronze.
Production grade
What goes wrong in production
| Problem | What breaks | What to do |
|---|---|---|
| Backups that keep erased data | A person's record is erased in the lakehouse, but a seven-year locked backup of an S3 bucket still holds it. | Keep personal-data buckets and the token vault out of AWS Backup by tag rules, and document that choice in the privacy design. |
| Restores that were never tested | Backups show green for a year, then the first real restore takes nine hours and fails on a missing KMS grant. | Schedule restore testing for every tier-1 resource, record restore time against the recovery-time target, and test cross-account restores too. |
| Untagged resources left out | A new database is created without the backup tag and is never protected. | Use Organizations backup policies, enforce the tag with an SCP or Config rule, and alarm on Backup Audit Manager findings. |
| Vault Lock set too early | Compliance mode is turned on with the wrong retention, and the extra storage cannot be removed for years. | Start in governance mode, check costs and retention for a month, then switch to compliance mode deliberately. |
| Double backups and double cost | DynamoDB point-in-time recovery, RDS automated snapshots and AWS Backup all keep copies of the same data. | Decide per service which mechanism owns backup, and check AWS's notes on how Backup and the native feature share storage. |
| KMS key loss | The vault key, or the source key needed for restore, is deleted or locked away, and the backups cannot be read. | Use a dedicated key per vault in the backup account, protect key deletion with SCPs, and include the key in restore tests. |
Alternatives
What to compare it with
| Alternative | Pick the alternative when | Pick AWS Backup when |
|---|---|---|
| Native service backups (RDS snapshots, DynamoDB PITR) | One or two databases with simple needs and no cross-account or audit requirement. | You need one policy across many services and accounts, locked vaults and audit reports. |
| S3 Versioning, Replication and Object Lock | You protect lakehouse buckets: keep old versions, copy to another region, and lock where the law requires. | You need scheduled backups of databases and volumes. |
| Delta time travel and deep clone | You need to restore a Delta table to an earlier version, or keep a consistent copy in another region. | The thing to protect is not a Delta table. |
| Azure Backup | You are on Azure and protect VMs, SQL and storage there. | You are on AWS. |
| Google Cloud Backup and DR | You are on Google Cloud. | You are on AWS. |
Cost
How you pay
AWS Backup charges for backup storage per GB per month, with a lower rate for cold storage, and for data restored per GB for some resource types. It also charges for cross-region data transfer when copying, for restore testing, and for Backup Audit Manager evaluations. Many backups are incremental, so after the first full copy only changes are stored, but not every resource type supports that. Check the AWS Backup pricing page for your region.
What drives the bill
- Total backup storage, which grows with retention length and change rate of the source.
- Cross-region and cross-account copies, which store a second copy and pay transfer.
- Long monthly or yearly retention, especially under Vault Lock where it cannot be cut later.
- Restore tests and real restores, charged per GB for some resource types.
Cost traps
- Backing up large S3 buckets that are already versioned and replicated pays for the same data three times.
- Cold storage has a minimum stay, so deleting early still charges for the full period.
- Resource types without incremental support store a full copy every time.
- Compliance-mode Vault Lock makes a retention mistake permanent.
Prices change; check the AWS pricing page for your region before deciding.
Security and governance
Keeping it safe
- Keep vaults in a separate backup account, ideally in another region, so that someone with admin rights in the workload account cannot delete them.
- Encrypt each vault with its own customer managed KMS key, and limit who can use and schedule deletion of that key.
- Use Vault Lock or logically air-gapped vaults for ransomware protection, and set vault access policies that deny deletion.
- Record backup and restore API calls in CloudTrail, and send Backup events to EventBridge for alerts on failed jobs or policy changes.
- Exclude personal-data stores that are subject to erasure, and write that exclusion down in the privacy design.
- Restrict who can start a restore, because a restore can create a copy of production data in a less protected place.
Operating it
What to watch, and the limits
| Watch | Why |
|---|---|
| NumberOfBackupJobsFailed | CloudWatch metric for failed backup jobs; alarm on any failure for tier-1 resources. |
| NumberOfCopyJobsFailed | Failed cross-region or cross-account copies, which leave you with only one copy. |
| NumberOfRestoreJobsFailed | Failed restores, including those run by restore testing. |
| Restore testing duration | How long each tested restore took, compared with the recovery-time target. |
| Backup Audit Manager control compliance | Resources not covered by a plan, or with retention shorter than policy. |
| Vault and key changes | CloudTrail events that change vault policies, Vault Lock or vault KMS keys. |
| Limit | Value and notes |
|---|---|
| Point-in-time recovery window | Continuous backups support restore to any point within up to 35 days |
| S3 backups | The bucket must have S3 Versioning turned on |
| Cold storage | Backups moved to cold storage must stay there at least 90 days |
| Vault Lock compliance mode | After the grace period ends, the lock cannot be removed by anyone, including AWS |
Limits and quotas checked 2026-11-11; many can be raised through Service Quotas.
Decisions
Situations and verdicts
| Situation | Choice | Verdict |
|---|---|---|
| Aurora order database that DMS reads for CDC | AWS Backup plan by tag, hourly with 35-day point-in-time recovery, copy to a locked vault in the backup account | Pursue: an outage here stops data flowing in, and restores must be proven. |
| Token vault and personal-data gold tables | Deep clone to the second region, S3 replication, 30-day version expiry and a tested restore drill; no AWS Backup | Drop AWS Backup here: long-kept copies would defeat erasure. |
| DynamoDB live state in the fast lane | Point-in-time recovery on the table; AWS Backup only if audit asks for locked copies | Keep if you have it: state can be rebuilt from the stream within the retention window. |
| Landing buckets deleted after one day | No backup; the source can resend | Drop: backing up landing undoes its short-life design. |
FAQ
Questions people ask
Does AWS Backup back up S3?
Yes. It backs up S3 buckets, with continuous backups for point-in-time recovery up to 35 days and periodic backups for longer retention. Versioning must be on. It treats the bucket as objects, not as Delta or Iceberg tables.
Should I use AWS Backup for my Delta Lake tables?
Usually not. Delta time travel, deep clone and S3 replication give consistent table-level recovery. On this platform AWS Backup is also kept off personal-data tables so erasure requests reach every copy.
What is AWS Backup Vault Lock?
A setting on a backup vault that prevents anyone from deleting backups or shortening retention. Governance mode can be changed by users with special rights; compliance mode cannot be changed after a grace period.
How do I prove backups actually work?
Use restore testing to restore backups on a schedule and record how long it took, and use Backup Audit Manager reports to show coverage. Keep both as evidence for auditors.
Is AWS Audit Manager still available for backup evidence?
AWS Audit Manager moved to maintenance in 2026 and is closed to new customers. AWS Backup Audit Manager, the reporting inside AWS Backup, is a separate feature and keeps working.
Sources
Checked against
- AWS docs: What is AWS Backup
- AWS docs: AWS Backup Vault Lock
- AWS docs: Services in maintenance (AWS Audit Manager)
- AWS: AWS Backup pricing
Checked 2026-11-11.
Related
Where it connects
S3 Versioning, Replication and Object Lock · AWS and Databricks reference platform · Table maintenance · Amazon Aurora PostgreSQL · AWS KMS · All AWS data services