This guide is for tech leads, operations leads and CTOs who own a live business application and its backups. It sits under Netbase's cloud platform engineering work, where backup policy, restore tests and a recovery plan are part of the platform, and the automation belongs in DevOps and platform engineering pipelines.
In this guide
- Why is a successful backup job not proof that you can restore?
- What does a restore test have to prove?
- Which test depth fits which cadence?
- The restore test runbook
- What should the restore evidence log contain?
- Can the cloud provider run restore tests for you?
- When is a full application restore not worth it?
- Where AI fits in restore testing
- What delivery record exists, and what does not
- Alternatives: who runs the restore tests?
- What are the limits of this guide?
- Frequently asked questions
- Next step
Why is a successful backup job not proof that you can restore?
A green backup job only proves that a copy was written. It says nothing about whether the copy is complete, whether it can be read back, or whether the application starts on it. Microsoft's reliability guidance warns that backups can be misconfigured to capture only part of the data and asks teams to validate integrity and completeness, not just whether a restore operation succeeds.
AWS lists the same traps as anti-patterns in its recovery testing practice: restoring a backup without querying any data, assuming the backup exists, and assuming the restore fits the recovery time objective. A restore test turns each assumption into a recorded fact with a date.
Two targets frame every test. Google Cloud defines the recovery time objective (RTO) as the maximum acceptable time an application can be offline, and the recovery point objective (RPO) as the maximum acceptable period of data loss. A restore test measures both: how long recovery took, and how old the newest restored record is.
What does a restore test have to prove?
A restore test passes only when four things are true, and each needs its own check:
-
Data
- Question it answers
- Is everything there and readable?
- Check that proves it
- Row counts, checksums or a sample of known records against the source
-
Freshness
- Question it answers
- Is data loss inside the RPO?
- Check that proves it
- Timestamp of the newest restored record against the incident time
-
Application
- Question it answers
- Does the product work on the restored data?
- Check that proves it
- The application starts, a test user logs in and completes one core action
-
Time
- Question it answers
- Is recovery inside the RTO?
- Check that proves it
- Clock from the decision to restore until the application check passes
Microsoft adds that recovery must be measured across the full flow, because one component can restore in time while the whole service still misses its target once dependencies also need restoring. Time the application, not only the database.
Which test depth fits which cadence?
Not every test needs a full rebuild. Pick the depth by what changed and what you need to prove:
| Criterion | Integrity check | Data restore | Full application restore |
|---|---|---|---|
| What it proves | The backup file or snapshot is readable and complete | One database or bucket restores and its data validates | The product runs on restored data inside the RTO |
| What it does not prove | That anything restores or starts | That the application, secrets and configuration come back | Recovery in another region or account |
| Typical trigger | After every backup run | On a fixed schedule and after a schema change | Before a major release, after an infrastructure change and on a fixed schedule |
| Who reviews the result | Automated alert | Engineer on duty | Tech lead and the business owner |
Microsoft's guidance ties restore cadence to how often the backup configuration, data schema or infrastructure changes: the more often they change, the more often restores need testing. Point-in-time recovery deserves its own test: PostgreSQL's continuous archiving needs an unbroken sequence of archived WAL files back to the base backup, and its documentation asks teams to set up and test WAL archiving before taking the first base backup.
The restore test runbook
-
Name the scope and the pass criteria
Write which system, which backup and the four checks above before you start, so nobody changes the bar during the test.
-
Pick the recovery point
Alternate the latest backup with an older or random one, since a recovery may need an older copy.
-
Start the clock at the decision
Record when the restore was declared, then who found the runbook, the credentials and the access.
-
Restore into an isolated environment
Use a separate account or project, never production, with its own network and no outbound mail or payment calls.
-
Validate data, freshness and the application
Run the data checks, read the newest record time and complete one core user action.
-
Stop the clock and compare
Note the elapsed time against the RTO and the data age against the RPO, and flag any gap.
-
Log, fix and tear down
Write the evidence log entry, open a task for every gap with an owner, and delete the restored copy.
A person other than the backup owner should run the runbook at least some of the time. The test then checks the written procedure, not one engineer's memory.
What should the restore evidence log contain?
The evidence log is the record a buyer, auditor or insurer asks for. Keep it outside the backup tool, so it survives if the tool or account is lost. One entry per test:
-
Test id and date
Restore test, month and system name
-
System and backup
Application, database or bucket, recovery point id and its timestamp
-
Depth
Integrity, data restore or full application restore
-
Operator and reviewer
Role of the person who ran it and the person who signed it off
-
Declared, restored, validated
Three timestamps: decision, restore finished, last check passed
-
Result per proof
Data, freshness, application and time: pass or fail with the evidence link
-
Gaps found
Missing secret, wrong permission, stale runbook step, missing dependency
-
Actions and owners
Task reference and owner for every gap, with a due date
A failed test is a useful entry, not an embarrassment. The log shows that gaps are found and closed, which is what a restore programme is for.
Can the cloud provider run restore tests for you?
Partly. AWS Backup restore testing runs scheduled restores of supported resources such as Amazon RDS, Aurora, DynamoDB, EBS, EFS and S3, picks either the latest or a random recovery point from a window you set, records how long each restore job took and then deletes the restored resources. You can retain the restored copy for 1 to 168 hours to run your own validation first, and AWS recommends a dedicated account for restore tests.
What the provider cannot do is decide whether your application works. A managed restore test proves that a resource comes back; the application check, the secrets and configuration, and the clock that starts at the decision remain your runbook's job. Use provider automation for the data restore depth and keep the full application restore as a human-led exercise.
When is a full application restore not worth it?
Do not run a full application restore every week for a small product with one database and a managed backup service. A scheduled data restore with automated checks covers most risk, and a full restore before major releases and after infrastructure changes covers the rest.
Do not build cross-region restore drills before single-region restores pass reliably. If the basic restore misses its RTO, a more complex test only adds noise. Fix the basic runbook first, then decide on a disaster recovery tier.
Where AI fits in restore testing
AI helps with the paperwork around a restore, not with deciding that it passed. An assistant can draft runbook steps from infrastructure code, compare restored row counts and schema against the source and summarise a failed restore job's logs into a first gap note. Netbase works with the major commercial and open-source AI tools and models, chosen per project.
Engineers still define the pass criteria, run the restore, sign the evidence log entry and own every fix; no model gets rights to delete backups, change retention or restore into production.
- In delivered work
MLOps pipeline
An MLOps pipeline Netbase delivered for a client that is not named; an AI delivery record, not a restore programme.
- Available capability
AI agents in operations
AI-assisted restore evidence review is a way Netbase approaches the work, not yet tied to a published case.
What delivery record exists, and what does not
Netbase names AWS, Google Cloud, DigitalOcean and Cloudflare as the cloud providers it uses. It also works as offshore development partner on a multi-tenant cloud ERP delivered as SaaS to small and mid-sized businesses for a US client, not named.
Delivery runs in a six-step lifecycle with weekly reviews and KPI dashboards, and most Netbase projects are delivered on fixed-price contracts. No published Netbase record describes a restore test programme in detail, and no restore duration, recovery target or result figure is given here.
Alternatives: who runs the restore tests?
| Model | Fits when | Selection criterion |
|---|---|---|
| Your own engineers on a calendar | A team with time reserved for upkeep | Restore tests survive busy feature sprints |
| Provider automation plus a quarterly human review | Supported managed databases and storage | Someone still runs the application check and reads the log |
| A managed service partner | No in-house operations capacity | The partner shows you its evidence log, not only a backup report |
Restore testing usually sits inside a wider care routine; Netbase's managed cloud services list restore tests as part of the scope, and the patch, scan and recover cadence shows how restores fit next to patching and scanning. To score restore practice against the rest of delivery, use the DevOps maturity scorecard. During a move, the final backup becomes a rollback point in the cutover plan.
What are the limits of this guide?
- It covers restore testing for one application, not choosing a disaster recovery tier or multi-region failover.
- Provider features and limits change; check the current documentation before you automate.
Plan the next step with a Netbase consultant
Frequently asked questions
Run an integrity check after every backup, a data restore on a fixed schedule and after schema changes, and a full application restore before major releases and after infrastructure changes. Teams whose configuration changes often need to test more often.
No. A restore test proves that one backup brings one application back within its targets. A disaster recovery test also covers failover to another region or account, traffic switching and failback, and it builds on restore tests that already pass.
They ask for proof that restores were tested, not only that backups ran. A dated evidence log with the recovery point, the elapsed time, the checks performed, the gaps found and their owners answers that question.
Next step
If you want your backups proven before an outage asks the question, book a solution review with Netbase.
Related services and solutions
AI-ready cloud platform engineering for SaaS and commerce workloads Netbase provides cloud platform engineering for SaaS and commerce teams that need a secure, repeatable home for their software and its AI features, on AWS, Google Cloud, DigitalOcean or Cloudflare. We design the architecture, write the environments as code, connect managed and AI services and hand over a platform your team can run, built the way we run ours.
Learn More
Managed cloud services with AI-assisted monitoring that keep your platform healthy Netbase provides managed cloud services for companies with a live platform to keep it secure, current and affordable after launch: AI-assisted monitoring and alert triage, patching, backups and regular cost reviews on AWS, Google Cloud, DigitalOcean or Cloudflare. Our team supports you Monday to Saturday, with Sunday off. SLAs are agreed on request; we publish no standard SLA figures.
Learn More
AI-assisted DevOps and platform engineering for routine releases Netbase provides DevOps and platform engineering for SaaS and commerce teams to make releases routine instead of risky. We set up CI/CD pipelines, repeatable environments and observability on AWS, Google Cloud, DigitalOcean or Cloudflare, with AI assistance for tests, reviews and incident triage, so every change ships the same way and problems surface before customers notice.
Learn More
SaaS product accelerator: launch an AI-ready SaaS on proven Netbase modules A SaaS product accelerator is a set of reusable Netbase modules for accounts, billing, roles and integrations that helps founders and product teams launch subscription software faster, with room for in-product AI from the first release. Reusing these modules can shorten development time, and the approach is proven on Printcart, Netbase's own web-to-print SaaS.
Learn More
Work with Netbase remotely
Every Netbase engagement is delivered remotely, so clients in any country and any industry can work with us; Netbase has worked with clients in 80 countries. Contracts and invoices are issued by Netbase JSC, and payment is made by bank transfer to Netbase JSC's company account or PayPal Business. Talk to the CEO, Huy Nguyen (David), on WhatsApp or by email.
Discuss a project
Netbase JSC helps organizations design, build, modernize, and operate digital products and AI-enabled business systems.project@netbasejsc.com
+84 937 869 689
91 Nguyen Chi Thanh, Dong Da, Hanoi, Vietnam
Get in touch
Tell us what you want to build, modernize, or operate.