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.
An incremental backup cannot be restored without the full backup it grows from. If each file is locked for the same number of days as your retention, the full unlocks first, while newer restore points that depend on it are still inside retention. The lock then protects the least important files and leaves the most important one exposed.
The lock on the full has to reach until the last restore point that depends on it has left retention. That is the full interval plus the retention, less one day. Or the backup product has to keep extending locks as new restore points arrive. Both designs cost something, and they fail in different ways.
Why does a lock equal to retention leave a gap?
A lock set at write time counts from the moment each object is written. In Amazon S3 the retain-until date for a bucket default is the object version's creation timestamp plus the duration, and each version carries its own date. Amazon states that "retention periods apply to individual object versions."
Retention counts from a different place: from the restore point you want to be able to reach. The oldest restore point you promise needs the full written weeks earlier, and every increment between them. The full was written first, so with equal lock lengths it expires first.
Take a retention of 30 days and a monthly full. On day 31 the full from day 0 is no longer locked. Restore points from days 2 to 29 of that chain are still inside retention, and each needs that full. Anyone who can delete the full on day 31 has made all of them unrestorable, although their own increments are still locked.
How long does the lock have to be?
For a forward incremental chain, a full written on day 0 and followed by a new full every F days is needed by every restore point up to day F − 1. The last of those points leaves retention at day F − 1 + R, where R is the retention in days. So the full needs a lock of F − 1 + R days from its write date, and every increment in the chain needs a lock that ends on the same date.
| Full every (F) | Retention (R) | Lock needed on the full | Lock needed on the increment written on day 5 of its chain |
|---|---|---|---|
| 7 days | 30 days | 7 − 1 + 30 = 36 days | 31 days |
| 14 days | 30 days | 43 days | 38 days |
| 30 days | 30 days | 59 days | 54 days |
| 90 days | 30 days | 119 days | 114 days |
The table assumes the product keeps a chain until its last restore point has left retention, and that the lock must end on that date. If your product deletes by a different rule, for example by deleting single points, the numbers change. Backup chain mechanics sit with the restore articles. This one covers only the lock.
Veeam documents the same shape for its hardened repository. The count of the immutability period "starts from the moment the last restore point in the active backup chain is created", and the period is "automatically extended for all backup files in the active chain." In its worked example the full was created on 12 January, increments followed on 13 and 14 January, and with a 10-day period every file stays immutable until 24 January, which is the last restore point plus 10 days. Veeam adds that it does not extend immutability for inactive chains. That is the same arithmetic as the table, done by the product instead of by you.
Veeam also documents what happens when a lock is not extended. For log backups the immutability period is not extended in the active chain, so if it expires the files become mutable, and "the application restore may fail as the log backup chain becomes incomplete." It advises that the period cover all backups in the active chain.
Example (illustrative): host fs01 takes a 1,000 GB full on day 0 of each 30-day cycle and a 20 GB increment each day after it. Retention is 30 days, kept by chain. A simple count of 30 restore points suggests 1,000 + 29 × 20 = 1,580 GB. But a chain cannot go until its last point, written on day 29, leaves retention on day 59. At day 59 the store holds two chains: 2 × 1,580 = 3,160 GB. With a 30-day lock on every file, the full written on day 0 unlocks on day 30. Between day 30 and day 59, anyone with delete rights can remove that full and take out up to 29 restore points that were still in retention. With the lock sized at F − 1 + R = 59 days, the full stays locked to day 59. The extra storage is the 3,160 GB − 1,580 GB = 1,580 GB that chain-based retention keeps anyway. A lock longer than 59 days would be paid for on top of it, at the provider's rate per GB-month.
What changes when the store keeps only data that is new to it?
Some products do not write a chain of files. They store each block once and let many restore points refer to it. A block first written on day 1 can be referred to by a restore point written on day 29. A lock set at write time on that block runs out on day 31 whatever refers to it.
Products with this design extend the lock on blocks that are reused. Veeam documents that it "continues to keep reused or dependent blocks of data locked by continuously assigning them to new generations, thereby extending their immutability expiration period." It also adds a "Block Generation" period: 30 days for Amazon S3, IBM Cloud, Google Cloud and 11:11 Cloud object storage, and 10 days for all other types. This is added to the immutability expiration "to reduce I/O operations and associated costs", so that blocks written within the same generation share one expiry date and need no extension request in between.
Kopia documents the same dependence in plain terms. By default Kopia "does not renew Object Lock retention dates"; renewal is switched on with kopia maintenance set --extend-object-locks true, and Kopia states that full maintenance must run at least as often as the lock period, otherwise "the Retention date will pass, and your data will no longer be protected" by Object Lock. It adds that the full-interval must be at least one day shorter than the retention-period.
Static lock or rolling extension: which costs what?
| Long lock at write time | Lock extended as data is reused | |
|---|---|---|
| Lock length | F − 1 + R for chains; as long as the longest dependency for shared blocks | Shorter at first, then extended |
| Extra storage | Whatever the lock holds beyond your own retention schedule | Close to what retention needs, plus the block-generation margin where used |
| Credentials | Backup account only needs to write | The account that extends needs retention-changing rights |
| Request cost | None beyond writing | A request per extension. Veeam says block generation exists to cut them |
| Silent failure | A mistaken period is permanent in compliance mode | If maintenance stops, locks lapse while the backup still reports success |
| Where limits bite | Amazon S3 compliance mode cannot be shortened | Azure allows at most five increases on a locked container-level policy; Microsoft states there is no limit on increases for a policy set on a blob version |
Two details deserve attention. In S3, extending is a per-version call: you submit a new request with a later retain-until date, and the older date is replaced. Anyone with s3:PutObjectRetention can do it. In Azure, the effective retention period is calculated from the most recent value of the interval, so extending a container-level policy lengthens it for every blob in the container, including data you no longer need. Microsoft also caps container-level increases on a locked policy at five.
How do I check this on my own system or with my provider?
- Find the full interval and the retention in the backup product's schedule, in days.
- Work out F − 1 + R. That is the lock the oldest full needs if locks are set once.
- On S3, read the lock on a recent full and a recent increment. These calls only read:
aws s3api head-object --bucket backup01-archive --key <key-of-full> --version-id <version-id>
aws s3api get-object-retention --bucket backup01-archive --key <key-of-increment> --version-id <version-id>
head-object returns the lock mode and the retain-until date if the caller holds s3:GetObjectRetention. Subtract the write date from the retain-until date. If the full's lock is shorter than F − 1 + R and your product does not extend locks, there is a gap.
- If the product extends locks, ask the vendor in writing:
- Which credential extends them, and where does it live?
- What happens if maintenance fails or is switched off for a week? Does the console warn?
- By how much beyond retention does it keep a block locked?
- Is the extension done per object, and what does it add to the request bill?
- For a provider-run store, ask: "For a restore point at the end of my retention window, which objects does it need, and is each of them locked until at least that restore point leaves retention?"
Risk: if you decide to switch on lock renewal in a product, test it on a non-production repository first. In Kopia the setting is kopia maintenance set --extend-object-locks true, and it will not accept the setting unless the full-maintenance interval is at least one day shorter than the retention period. Switching it on lengthens locks on data you may have meant to remove, and compliance-mode locks cannot be shortened afterwards.
Read next
Related articles you may want next.
- Backup & RecoveryLocked backup storage that only grows: find the retention or legal hold that is keeping the dataBackups in an Object Lock bucket cannot be deleted and the bill grows. Find the retention date or legal hold behind it, and what you can change.Read article
- CloudS3 Object Lock and versioning: designing backups that an admin cannot deleteWhat versioning, MFA delete, legal holds, governance and compliance mode each stop, who can undo them, and a safe build order for a backup bucket.Read article
- Backup & RecoveryGrandfather-father-son retention: counting the restore points a policy really keepsHow to count the restore points a 30 daily, 12 monthly backup policy keeps, where the gaps fall, and what to ask a provider to confirm in writing.Read article
Risks and limits
- These commands have not been run or lab-tested. They are taken from the vendors' documentation.
- The table of lock lengths assumes a forward incremental chain, one full every F days and retention by whole chain. Reverse incremental, synthetic full and merge-based products have different dependencies. Ask the product's vendor for the rule.
- Veeam's pages describe Veeam behaviour and are used here only to show how a product handles the problem. They say nothing about other products.
- S3 date arithmetic assumes retention is applied at write time, as Amazon documents for bucket defaults. A product that sets its own date per object can choose any date.
- The Kopia page is dated 11 June 2025 and may lag the current release.
- A lock stops deletion through the storage service. It does not stop a restore that depends on a block that was never written correctly, and it does not replace a restore test.
Sources
- How Immutability Works (hardened repository), Veeam Backup & Replication 13 User Guide, page updated 8 September 2026, accessed 17 November 2026
- Block Generation, Veeam Backup & Replication 13 User Guide, page updated 6 February 2026, accessed 17 November 2026
- Immutability for Backup Files, Veeam Backup & Replication 13 User Guide, page updated 4 August 2026, accessed 17 November 2026
- Locking objects with Object Lock, Amazon S3 User Guide, accessed 17 November 2026
- Object Lock considerations, Amazon S3 User Guide, accessed 17 November 2026
- head-object, AWS CLI reference, accessed 17 November 2026
- get-object-retention, AWS CLI reference, accessed 17 November 2026
- Ransomware Protection, Kopia documentation, last modified 11 June 2025, accessed 17 November 2026
- Immutable storage for blob data overview, Microsoft Learn, last updated 25 August 2026, accessed 17 November 2026
This article was originally published on .