Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article7 min readViews0

Backups Stored for 3, 7, or 30 Days: How to Choose Backup Retention Depth

We show how to calculate backup retention time based on detection delay, decision time for restoration, and snapshot interval—without guessing.

Comments 0

Three sealed backup capsules at different depths under an orange rescue buoy
In this article

Backup retention depth is chosen not by the principle "the more, the calmer," but by the worst-case plausible time to detect an error. A practical guideline: retention time must exceed the sum of detection delay, time to confirm the cause and decide on restoration, plus one snapshot interval. Otherwise, the last valid point may disappear just when it becomes clear what needs to be restored.

Frequency and depth are different parameters. Daily backups answer how often a new point appears. Storing for seven days means how long an old point remains available. Seven daily backups do not always mean seven full days of history: exact creation and deletion times, missed jobs, and provider rules matter.

First, find the last safe point

Let's consider a store training scenario. A snapshot is created daily at 02:00. An import error entered the database on Monday at 13:00, but it was noticed only on Friday at 10:00. By 16:00, the team confirmed the cause and selected recovery. The last known good copy was created on Monday at 02:00.

Event

Moment relative to the error

What this changes

Last safe backup

T−11 h

By the time the decision is made, its age will be 110 hours

The error entered the data

T0

All subsequent snapshots may contain the defect

The issue was discovered

T+93 h

A three-day history is too short

The decision to restore has been made

T+99 h

A working copy aged 110 hours is needed — four days and fourteen hours

Timeline: safe copy 11 hours before the error, detection after 93 hours, and decision after 99 hours
In the training scenario, the safe copy at the time of the decision is 110 hours old: a three-day history will no longer preserve it.

In this example, three days are insufficient, seven days preserve the needed point, and thirty days offer no additional benefit for this specific scenario. Here is an interesting point: if the error occurred one minute before the next snapshot, the last safe copy will be almost a full day older than the error itself. Therefore, the full backup interval is added to the detection time, and the result is not rounded down.

  1. Assess the maximum detection delay: not the average, but the one possible for a rarely checked function.

  2. Add time for verifying the cause, agreeing on a rollback, and obtaining access to copies.

  3. Add one interval between snapshots: it covers the unfavorable error position between two jobs.

  4. Verify the result against the exact time of copy deletion and leave a margin at the deadline.

Such a calculation sets a business requirement, not a universal standard. NIST Recovery Planning Guide recommends defining requirements and priorities for a specific system; it does not assign the same 3, 7, or 30 days to all sites.

What 3, 7, and 30 days provide in practice

Depth

When it may suffice

Where the risk lies

3 days

Errors are visible on the same day, checks are performed daily, and decisions are made quickly

A defect occurring before the weekend or in a rarely used feature may survive the entire history

7 days

Checks occur several times a week; the sum of detection delay, decision time, and copy interval plus a buffer is less than seven days

Monthly exports, period closure, and slow customer complaints may reveal the problem later

30 days

An error may appear with a significant delay, but the calculation, together with decision time, copy interval, and buffer, fits within thirty days

A longer history is useless without a clear understanding of the copy's composition, access, and recovery test

The table does not automatically select a tariff. If warehouse stock levels are reconciled once a week while payment statuses are checked daily, they may require different protection methods. Order databases benefit from more frequent points or change logs, while rarely changing files need a longer history. A single retention period for the entire project is convenient in a contract but does not always reflect the actual risk.

Why a daily backup may not leave a safe copy

  • The error remains unnoticed for a long time and gets included in all new snapshots.

  • The task is formally started, but the database, user files, or configuration were not included in the copy.

  • Snapshots are stored in the same account and are deleted along with the working data.

  • The provider counts the term from the exact creation time, while the team counts calendar days mentally.

  • A copy failure reduced the number of points, but no one processed the notification.

  • Recovery was not tested after updating the CMS, database, or integrations.

In the event of a ransomware attack that encrypts data, long-term storage of copies alone is insufficient. CISA's guide to ransomware protection recommends keeping critical copies offline and encrypted, regularly verifying their availability, integrity, and recovery procedures. A writable copy can be corrupted by the same credentials used for the main server.

Multiple backup groups are more useful than a single retention period

For a project with varying detection speeds, you can separate the history. For example, this is a training model, not a finalized policy: seven daily snapshots cover recent errors, five weekly ones cover issues found during weekly reconciliation, and twelve monthly ones cover rare period-closing errors. The composition and independence of each layer must still be verified separately.

Layer

Schedule example

Task addressed

Short

Daily, keep 7 backups

Quick rollback of recent changes

Medium

Weekly, keep 5 backups

Error detected during the next verification

Long

Monthly, keep 12 backups

Seasonal or reporting comparison

The volume is calculated with open assumptions. If a full copy takes 80 GiB, seven independent full copies require up to 560 GiB before accounting for compression and deduplication: 80 × 7 = 560. An incremental scheme may take less, but it cannot be estimated without the volume of changes and chain-building rules. Space savings must not turn a single corrupted base copy into a failure of the entire history.

What to clarify with the hosting provider or contractor

  • What exactly is included in the copy: database, files, email, configuration, tasks, and user uploads.

  • Which time zone is used to create points and when they are deleted.

  • Whether depth is measured in calendar days, full days, or the number of points.

  • What happens when a task is skipped and who receives the notification.

  • Whether a server and copies can be deleted with a single account; whether separate or immutable storage exists.

  • How long it takes to retrieve a copy and who has the right to initiate recovery.

  • Whether a project can be restored in isolation without emails, payments, 1C integration, or calls to active APIs.

When comparing support options, look beyond the number of days: on the comprehensive support page, variants differ in storage depth, but before choosing, you must still agree on the data composition, copy locations, and recovery sequence.

Acceptance criterion for the selected depth

  • For each critical function, the maximum detection latency is recorded, not a convenient average.

  • The calculation includes the time to decide on recovery and the full interval between snapshots.

  • The panel displays actual points with dates; the count and age match the storage conditions.

  • There is at least one safe point for the selected control scenario.

  • The last isolated recovery was completed with checks for orders, payments, files, and integrations.

  • Copies are protected from deletion by the same event or access that would damage the main site.

  • After changes to the CMS, database, schedule, or data composition, the calculation and test are repeated.

A separate backup verification via RPO, RTO, and test restoration answers the question of whether the store can be restored to operation. Material on copies stored outside the main hosting helps verify the independence of the storage location. Depth is the third separate check: will the required valid point remain available by the time the decision is made?

The final criterion is simple: choose not the longest retention period offered by the provider, but the one that survives your worst plausible scenario for error detection and decision-making on restoration. If the calculation yields five days, a three-day option is already unsuitable, and a seven-day option must be verified against actual creation and deletion timestamps. Unfortunately, the attractive word 'daily' does not replace a calendar here.

Discussion 0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.