A scenario worth planning for
Consider a hypothetical case management system that fails to load one morning. The database server is running, but its storage has been encrypted by something other than the IT department. From that point the day turns on two questions: how much data can be recovered, and how long will recovery take?
Both questions can be answered in advance, which is far better than answering them during an incident. This article introduces how to do that on AWS.
Agree two numbers before choosing any technology
The recovery point objective is how much data the service can afford to lose, expressed as time. Suppose the only backup runs at midnight and nothing else captures changes during the day. If the live data becomes unrecoverable at 17:00, everything entered since midnight is lost. The recovery time objective is how long the service can be unavailable before the consequences become unacceptable for the people who rely on it.
Both are business decisions, so the service owner should make them, with engineering explaining what each option costs. Costs rise steeply as the numbers shrink, which is why it pays to set them service by service instead of declaring one target for the whole estate.
Know which part of the problem is yours
AWS runs the infrastructure beneath your services, while protecting the data and configuration you place there is your organisation's responsibility. Several services include their own protection, such as automated database backups with point-in-time restore and object versioning in Amazon S3, and these are worth enabling.
AWS Backup then gives you a single place to express policy across many services. A backup plan sets the schedule, the retention period and any copies to other accounts or Regions, and resources can be assigned to a plan by tag. Authorised teams can then inspect backup coverage and policy in one place, which makes gaps much easier to find.
Assume someone has your credentials
Attackers who encrypt systems for ransom have every reason to find and destroy backups first. Keep backup vaults in a separate account from the workloads they protect, administered by different people.
AWS Backup Vault Lock adds another barrier. In governance mode, users with sufficient permissions can still change or remove the lock. In compliance mode, once a grace period of at least three days has passed, nobody can remove or alter the lock, including the account's root user and AWS, and recovery points cannot be deleted before their retention ends. That protection is valuable, and it also turns retention into a cost commitment, so set it deliberately.
There is one important exception. If the account that owns a locked vault is closed, AWS deletes the vault's contents after 90 days. Treat account closure as a privileged operation, keep the vault-owning account under clear organisational ownership, and restrict who can close it.
A logically air-gapped vault goes further. It is always locked in compliance mode, its backups are held in an account owned by the AWS Backup service, and it can be shared with a separate recovery account so that restoration remains possible if the original account is compromised. AWS recommends keeping cross-Region copies in these vaults for greater resilience.
A backup proves little until it has been restored
A successful backup job shows that data was written somewhere. AWS Backup restore testing goes a step further by running scheduled restores, choosing the latest or a random recovery point for each resource, and recording how long each restore took. Restored resources can be retained for up to 168 hours so that you can check them, after which AWS Backup deletes them.
That validation has to be configured. Set the retention window and run your own checks against the restored resources, because the restore job alone shows that data came back and how long it took. Whether a caseworker can sign in, find a record and save a change is a separate question that only a test of that task answers. Our article on recovery test evidence covers what a useful exercise involves.
Where to start
Choose one service that matters. Write down its recovery objectives and who agreed them, check that its backups sit in an account the workload's administrators cannot reach, then restore it once, time it, and record what surprised you.
If a second view would help, our complimentary recovery readiness session looks at one service and the evidence you already hold.
Working document: a recovery exercise record
Each exercise should leave a record that a service owner, an auditor or the next engineer can understand without having been present.
| Field | What to record |
|---|---|
| Service and owner | The service tested and the person accountable for accepting its recovery. |
| Scenario | What was simulated, for example encrypted storage requiring restoration into a separate recovery account. |
| Recovery point | The recovery point used, its timestamp, vault and account. |
| Objectives | The agreed recovery point and recovery time objectives, and who agreed them. |
| Timings | When the incident was declared, when restoration started, when data was available and when the service was accepted. |
| Service acceptance | The user task tested, who performed it and the result. |
| Gaps and actions | Each problem found, with an owner and a date for resolution. |
Further reading
AWS Backup: logically air-gapped vaults
