Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Grandfather-father-son retention: counting the restore points a policy really keeps

Technical article
Backup & Recovery

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

A retention line such as 30 daily, 12 monthly does not tell you how many restore points you will hold on any given day. The count depends on a convention that the line rarely states: whether a month-end backup that is still inside the daily window counts once or twice. On our invented example the count on 15 November 2026 is 41 points under one reading and 42 under another, and the oldest point differs by about a month.

The gaps matter more than the total. A tiered schedule keeps every day for a month, then roughly one day per month. A file that was created and deleted between two month-end points is not in any restore point at all. This guide shows how to count the points, where the gaps fall, and what to ask a provider to put in writing. It is general information about retention; it does not describe any particular product.

Why it matters

A restore point is a saved state of a system or data set at one moment, which you can recover from. Retention is how long each one is kept before it is deleted. Buyers may compare retention by reading the number of copies, but the useful question is different: if something went wrong on a particular date and is noticed weeks or months later, is there a point that predates it?

Two situations make the count matter:

  • A fault, deletion or malicious change is discovered late. Slow discovery is common (our reading), so older, sparser points are the ones you will actually need.
  • A contract or quote promises a number of copies. If the counting rule is vague, a dispute after an incident is about arithmetic nobody wrote down.

Retention schedules describe backups. Replication keeps a second, current copy and normally has little history, and failover is the act of switching to that copy. A deletion or corruption can be replicated along with everything else (our understanding), so replication does not substitute for the older points that a retention schedule is designed to keep. Disaster recovery is the wider plan that decides which of these you need.

The grandfather-father-son idea

Grandfather-father-son is an older name for tiered retention. The son tier is frequent and short-lived (daily points), the father tier is less frequent and kept longer (weekly or monthly points), and the grandfather tier is the least frequent and kept longest (monthly or yearly points). The idea is to buy depth of history without keeping every point for the full period.

The names vary by product. What matters is that each tier has two numbers: how often a point is taken or promoted into that tier, and how long it is kept. Read each tier as its own rule.

How two major services express tiers

The two services below are documented examples, not recommendations. They show that the same idea is written in different ways.

Azure Backup uses one policy with a schedule and a retention part. Microsoft's documentation says that retention can be defined for daily, weekly, monthly and yearly backup points, and defines monthly as "a backup on a certain day of the month". For Azure VM backups, the support matrix says monthly or yearly backup frequency is not supported with the VM extension, but that you can retain daily or weekly backups for a monthly or yearly retention period. In other words, the monthly point is a daily or weekly point that is kept longer, not a separate backup job (our reading of that wording).

AWS Backup expresses the same idea with separate rules in one plan. Each rule has its own schedule and lifecycle, and the lifecycle includes a delete setting that is "The number of days after creation that a recovery point is deleted", and the create-plan page says "The total retention period is the number of days that you store your resource in AWS Backup". A plan that keeps daily points for 30 days and month-end points for a year therefore contains two rules. AWS also documents what happens when rules overlap: "AWS Backup retains the backup under the rule with the longer retention period" (plan options). The page states this as a general rule, not one limited to its hourly and 12-hourly example: when the start windows of different rules overlap, the longer-retention rule applies, and AWS adds that "the optimization takes into account the full start window, not just when the daily backup is taken". In a plan with a daily rule and a month-end rule, the month-end point is therefore kept under the longer rule when the two start windows overlap. What the documentation cannot tell you is how a particular provider has set the start windows, so ask the provider to confirm the schedules, the start windows and the list of points actually retained.

The practical lesson is that a policy summary such as 30 daily, 12 monthly is shorthand. The exact rules sit in the product settings, and the summary is only as good as the convention behind it.

Counting the restore points on a given day

Here is a counting method you can apply to any proposal. All dates are invented.

Take a policy with one backup every day, the newest 30 daily points kept, and the month-end point kept for 12 months. Today is 15 November 2026.

  1. List the daily points. 30 points, from 17 October to 15 November 2026.
  2. List the month-end points. The last 12 month-ends on or before today run from 30 November 2025 to 31 October 2026.
  3. Remove duplicates. 31 October 2026 is both a month-end point and a daily point, so it is one restore point, not two.
  4. Count. 30 + 12 - 1 = 41 unique restore points.

Our arithmetic (a short date script, not a provider tool) gives these results for the same policy on other days:

Date of count Daily points Month-end points Counted twice Unique points Oldest point
15 Nov 2026 30 12 1 41 30 Nov 2025
30 Nov 2026 30 12 1 41 31 Dec 2025
1 Mar 2027 30 12 2 40 31 Mar 2026

On 1 March 2027 the daily window reaches back to 31 January, so two month-ends (31 January and 28 February) sit inside it. The count is 40, not 41. The number of points you hold therefore changes from day to day under the same policy.

There is a second reading. If the 12 monthly points are meant to be in addition to the 30 daily ones, they must all be older than the daily window. On 15 November 2026 that means 12 month-ends from 30 September 2026 back to 31 October 2025, 42 points in total, with history reaching back about 380 days instead of about 350. Nothing in the sources we read says which convention a given product or quote uses. This is the first question to put to a provider.

Where the gaps are

With the invented policy above on 15 November 2026, the points held are one per day back to 17 October, and then month-ends only:

  • 30 September, 31 August, 31 July, 30 June, 31 May, 30 April, 31 March, 28 February, 31 January 2026
  • 31 December and 30 November 2025

The gap just beyond the daily window is 17 days, from 30 September to 17 October. Older month-end gaps in this example are 28, 30 or 31 days (31 January to 28 February 2026 is 28 days); 29-day February gaps arise only in leap-year examples.

Three consequences follow (our reading of the arithmetic):

  • You choose the nearest point, not the exact date. To recover the state of 10 October, the nearest earlier point is 30 September (ten days of change are missing) and the nearest later point is 17 October.
  • Short-lived data may never be captured. A file created on 3 October and deleted on 12 October exists in no restore point, because no point was kept between 30 September and 17 October.
  • The gaps grow with age. Recovering a change from last June means choosing between two month-end states.

Weekly points reduce these gaps. If the policy keeps weekly points for 8 weeks, the 17-day gap closes, at the cost of more points to store. Whether that is worth it depends on how long a problem usually goes unnoticed in your business, which only you can estimate.

Why the nominal numbers drift

The calendar above is the policy as written. The documentation shows several reasons the real count can differ.

Policy changes apply to existing points. Azure's documentation says: "the new retention duration is applied to the existing recovery points as well" when retention is changed. If retention is reduced, some points are cleaned up. It also says that changing from daily to weekly schedules removes the existing daily recovery points (Azure Backup architecture, updated 29 September 2026). A change meant to affect future backups can therefore shorten history immediately.

Pruning is not instant. For Azure VM backups the FAQ says pruning "takes 24 hours" after a policy change (Azure VM backup FAQ). AWS says it randomises deletions within a window after expiry: AWS Backup "randomizes its deletions within 8 hours following recovery point expiration" (backup deletion). A point you count on the last day of its life may or may not still exist.

The clock starts when the job starts. AWS states: "The creation date of a backup is the date the backup job started, not the date it completed." A job that starts at 23:30 and finishes after midnight belongs to the earlier date, which matters when you count days (AWS Backup, backup plan lifecycle).

Some points outlive the policy. Azure says that for protected machines whose source no longer exists, "up to two of the most recent recovery points" may be retained beyond policy expiry (manage and monitor Azure VM backups). That is a billing and storage matter as much as a recovery one. Azure also keeps deleted backup data for 14 additional days under its soft delete feature, as described on the same page.

Floors and ceilings exist. Azure's FAQ says the VM backup policy supports a minimum retention range from seven days up to 9999 days. AWS requires cold-storage points to stay for at least 90 days ("Backups transitioned to cold storage must be stored in cold storage for a minimum of 90 days", per the AWS lifecycle reference). We cover storage-tier minimums separately; for counting, the point is that a long-term tier may be constrained by rules that are not visible in the summary.

None of these invalidates the policy. They are reasons to ask for the counting rules and to check the real list of points periodically rather than trust the summary.

Choosing and agreeing a schedule

The following is our recommendation, not provider guidance.

  1. Start from discovery time, not from copy count. Ask how long a problem could go unnoticed (a deletion by a departed employee, a slow data corruption, an unnoticed change by an attacker). Choose the oldest point you must be able to reach, then check the gap size at that age.
  2. Write the convention down. State whether monthly points are in addition to the daily window or overlap it, and which day counts as the month-end.
  3. Check the youngest tier first. In our view, many restores will be of recent data. Frequent points with a short window serve that need cheaply.
  4. Fill the largest gap you cannot accept. If a 17-day gap is too big, add weekly points rather than lengthening every tier.
  5. Agree what happens when the policy changes. Ask whether a change applies to existing points, how soon pruning occurs, and whether you are told before history is shortened.
  6. Keep retention separate from other protections. Retention keeps old states. A copy held away from your provider's control, or a lock that prevents early deletion, protects those states in a different way.

Questions to put to your provider

  1. For the schedule in the quote, do the monthly points overlap the daily window or sit in addition to it?
  2. Which day or backup is promoted to the monthly point, and what happens in months with fewer days?
  3. What is the exact number and date of the oldest restore point today, and how can I see the full list?
  4. Does a change to retention or schedule apply to existing points, and when is the pruning carried out?
  5. If a backup job runs past midnight or fails, which date does the point carry, and is the missed point replaced?
  6. Are there minimum retention periods or archive rules that affect the oldest points?
  7. When protection of a system is stopped or the system is deleted, which points remain and for how long?

A worked example (illustrative)

The following is invented. A small accounting firm is quoted 30 daily and 12 monthly backups for its file server. On 1 April 2027 an accountant realises that a client folder was emptied by a bad script on 9 February. The firm asks for the 8 February state. The daily window starts on 3 March, so the 8 February daily point has already been pruned, and the nearest month-end points are 31 January (before) and 28 February (after). The folder was created on 4 February, so it is in neither. The firm draws the calendar, sees the gap, and asks the provider to add weekly points for 12 weeks. It also asks for the counting convention in writing.

Risks and limits

  • The arithmetic assumes one backup every day and no failures. Real schedules have missed jobs, which widen gaps.
  • Product behaviour differs. We read AWS Backup and Azure Backup documentation only; other products may count differently.
  • The sources describe how those services work, not how a particular provider configures them.
  • Documentation changes. Re-check the pages before relying on a number.
  • Retention depth says how far back you can go. It does not say whether a given point is clean, restorable or complete. That needs testing.

Next step

Draw the restore-point calendar for the proposed policy: list the daily window, the month-ends and the largest gap, then ask the provider to confirm each assumption in writing. If you are reviewing a managed backup arrangement, the Backup and Recovery page is a starting point for that conversation; ask for the scope in writing.

Sources

All accessed 15 November 2026.

Learning notes

  • A restore point is a saved state at one moment. Retention is how long each one is kept.
  • 30 daily, 12 monthly is shorthand. Ask whether the monthly points overlap the daily window.
  • The older the point you need, the bigger the gap between the points available.
  • A policy change can delete existing points, not only change future ones.
  • Replication and failover are not a substitute for older restore points.