All checklists

You have a backup. Have you tried restoring it?

Check what is saved, where it lives, whether data and uploads fit together, and how to restore or roll back. Eight illustrated checks for a small website or app.

Last updated

#Backup#Recovery#Deployment

Official sources checked:

A backup menu is easy to ignore while a site works. Delete the wrong record or break a deployment, though, and the useful question becomes: what can I recover, and how? These checks give a small website or app a practical recovery plan. GitHub, PostgreSQL and Cloudflare are examples; your provider determines availability, retention and cost. Numbers in the illustrations are examples, not promised recovery times. Rehearse in an isolated temporary environment using synthetic or suitably protected data.

CHECK 01 / 8Do you know what lives outside the code repository?

Different locations need separate coverage.
Different locations need separate coverage.

List code, article originals, database records, uploaded originals, thumbnails and hosting configuration. Put a storage location and export method beside each. A repository backup covers what is in that repository; it does not automatically collect a separate database or file service. This inventory is a suggested operating practice, so a small static site can mark unused components as not applicable with a reason.

Where to look
Repository files · database console · file storage · hosting and domain settings
When to tick it
Every required component has a location, backup method and owner; missing coverage is recorded.

Read the official source ↗

CHECK 02 / 8How much recent data could you afford to lose?

Choose data-loss and downtime targets first.
Choose data-loss and downtime targets first.

Choose two targets: how far back the restored data may be, and how long reopening may take. For example, yesterday evening and half a day are planning examples, not provider guarantees. A quiet blog and an app receiving new reservations need different targets. Compare your choice with the actual backup schedule and available restore points. PostgreSQL documents several backup approaches; it does not choose these business targets for you.

Where to look
Latest backup timestamp · available restore points · your recovery targets
When to tick it
You have written data-loss and downtime targets and compared them with the available recovery method.

Read the official source ↗

CHECK 03 / 8Have you inspected the latest result, not just the schedule?

A schedule needs a recent, readable result.
A schedule needs a recent, readable result.

Check the latest run, timestamp, file size and a readable sample. Decide whether you need repository history as well as current files: a downloaded ZIP and a Git mirror have different scope. GitHub also documents separate considerations such as LFS data. Keep a simple failure alert or regular review so an old artifact cannot quietly pass as a current backup.

Where to look
Backup run history · actual artifacts · repository and LFS coverage
When to tick it
The latest artifact has an inspected timestamp, size and sample, plus a way to detect stale or failed backups.

Read the official source ↗

CHECK 04 / 8Could you reach the backup if the original server were unavailable?

Separate storage, restricted access.
Separate storage, restricted access.

A copy on the same temporary disk can disappear with the application. Choose a separate storage location and check access from the recovery operator’s account. Protect the backup from public download, and keep a recoverable route to encryption keys where encryption is used. OWASP’s secrets guidance informs access and key handling; the separate-copy check is our practical recommendation. Never paste passwords or API keys into a public repository or audit report.

Where to look
Backup location · access controls · encryption and key recovery path
When to tick it
An authorized operator can read the copy without the original server, and public access and key exposure are prevented.

Read the official source ↗

CHECK 05 / 8Will the database and uploaded files fit together?

Restore a matching set of components.
Restore a matching set of components.

Restoring yesterday’s records beside an unrelated set of files can leave broken references. Record the data timestamp, file inventory and compatible app version as a recovery set. PostgreSQL’s pg_dump produces an internally consistent database export; it does not copy a separate upload service. Check roles, extensions, configuration and externally managed identity data separately. Treat missing components as gaps, rather than assuming the database archive contains the whole application.

Where to look
Database restore point · upload inventory · app version · identity-service coverage
When to tick it
Records and files match, and separate account, permission and configuration requirements are documented.

Read the official source ↗

CHECK 06 / 8Have you restored into an isolated temporary environment?

Rehearse without overwriting production.
Rehearse without overwriting production.

Load a backup into an empty test environment, then open a sample record and attachment through the app. Archive readability and application usability are different checks. Disable email, payments, social publishing and scheduled jobs before starting, and verify that the test app cannot write to production. Prefer synthetic data where possible. Record failures and remove a temporary environment containing sensitive data when the rehearsal is finished.

Where to look
Temporary database and test URL · restore log · sample screens · disabled outbound integrations
When to tick it
The rehearsal left production untouched, opened sample records and attachments, and sent no external actions.

Read the official source ↗

CHECK 07 / 8Does rolling back code also roll back data?

Code history is not database history.
Code history is not database history.

Treat deployment rollback and data restoration as separate operations. Cloudflare Workers versions do not track changes inside associated storage resources. A previous app version may also be incompatible with a newer database schema. Rehearse that combination before choosing an old release; a compatible forward fix may be preferable. Write down which code version works with which data state, rather than relying on the presence of a Rollback button.

Where to look
Deployment history · database schema changes · previous-version compatibility test
When to tick it
A rollback candidate is tested against the intended data state, with a separate restore or forward-fix route where needed.

Read the official source ↗

CHECK 08 / 8Could someone follow your recovery instructions on a bad day?

Record observed results and remaining gaps.
Record observed results and remaining gaps.

Write a short runbook with the operator, access route, chosen restore point, ordered steps and stop conditions. Our suggested order is to verify records, files and permissions at a test URL before an authorized production switch. Record the rehearsal date, actual elapsed time and unresolved gaps. After any real switch, check the public URL and core actions again, and account for recent changes that the selected backup could not recover.

Where to look
Recovery runbook · latest rehearsal record · before-and-after verification list
When to tick it
Another operator can follow the steps; stop conditions, observed rehearsal results and remaining data loss are recorded.

Read the official source ↗

Copy this prompt and paste it into your AI

Replace the bracketed fields with your URL, tools and test environment. Copy the whole prompt to request all eight checks, observed results and remaining account-side work.

Sources & further reading