Immutable backups: what the lock actually protects

Share
Laptop screen displaying software code
Contents

At a glance

How S3 Object Lock and Borg append-only storage change deletion rights, and why recovery still depends on versions, encryption keys and maintenance access.

Follow the right that changes

The useful question about an immutable backup is concrete: which identity can destroy which recovery point, and until when? A storage label alone does not answer it. A team may have a backup writer, a cloud administrator, a person holding encryption keys and a maintenance job. If one compromised account controls all four, the design has different failure paths from a system that separates them.

Amazon S3 Object Lock and Borg's append-only mode illustrate two ways of restricting change. They operate at different layers and should not be treated as equivalent products. The following analysis uses their documentation checked on 16 November 2026. It examines the permissions and recovery dependencies rather than prescribing a vendor or a legal retention period. For a small team, this distinction matters when buying a backup plan: ask for the configuration and recovery procedure behind the word immutable. [1][2][3]

A protected version can disappear from the ordinary view

AWS documents Object Lock as protection attached to individual object versions in a versioned bucket. A new version or a delete marker can still be created. Consequently, an ordinary view that appears to show a deleted object does not settle whether a protected version remains recoverable. The recovery process must identify the right version, not merely request the familiar filename.

Consider a hypothetical project archive uploaded each evening under the same name. An operator looking only at the latest result may miss an older retained version. The practical requirement is an inventory that connects a business recovery point to its stored version and date. That mapping is part of the backup design. A dashboard saying storage is protected is less useful during an incident than knowing which version contains the last usable archive and which person can retrieve it. [1]

The privileged route decides the protection

S3's governance and compliance modes assign different powers. AWS documents a specific permission and request header for bypassing governance retention. Compliance mode prevents ordinary object deletion or shortening the protected period even by the account's root user; AWS separately notes account deletion as an exception. A governance label therefore needs a permissions review, while a compliance setting needs deliberate retention planning.

For example, granting the everyday backup process the same bypass capability as the maintenance administrator changes what a stolen backup credential can accomplish. Separating those roles creates an opportunity to limit that credential's reach. It is still necessary to check the actual effective permissions, including inherited roles and operational access. A diagram with two differently named accounts is not enough if the same unattended machine can obtain both credentials. This is an architectural question, not a reason to assume either mode is unsuitable. [1]

Encryption adds a separate recovery dependency

AWS explicitly distinguishes preserving an object from preserving access to its encryption keys. An object protected against deletion can become unreadable if the required KMS key is deleted. Retention and encryption therefore need separate recovery owners and checks. A storage administrator may demonstrate that an old version exists while the person responsible for decryption has no working recovery route.

In a tabletop exercise, start with the people and devices you assume unavailable. Ask the remaining operator to locate the version, obtain the approved credentials and decrypt a harmless sample. Record where that sequence stops. Do not put production secrets into a shared test document. The result is useful even before a full disaster drill: it reveals whether recovery depends on the same laptop or personal account whose loss the backup was meant to survive. Existence, permission to read and ability to decrypt are three distinct requirements. [2]

Append-only storage has a maintenance boundary

Borg documents append-only mode as preventing compaction of old repository data. Logical changes and the recovery of an earlier transaction require their own understanding; this is not simply a second name for S3's timed object retention. The documentation also distinguishes access through an append-only server from access able to modify repository files directly. That boundary determines who can enforce the restriction.

Maintenance deserves special attention because removing obsolete data is a legitimate operation. A system that never frees storage eventually faces a capacity problem, but allowing the routine writer to erase the retained history weakens the intended separation. Assign a maintenance path and check its inputs before using it. Keep the recovery instructions alongside the repository's operating notes. A team that can enable append-only mode but cannot identify a recoverable transaction has only completed part of the design. [3]

Three separate recovery questions

QuestionEvidence to retainResponsible role
Does the required version remain?Version or recovery-point identifier and retention stateStorage owner
Can the emergency operator retrieve it?Tested access path and effective permissionsAccess administrator
Can the content be decrypted and used?Separate recovery test and key-access procedureRecovery owner

Buy and test a recovery window

The commercial comparison is a recovery window, not merely a feature checkbox. Write down how long a useful version should survive, what could hide it from normal listings, who can change retention and how the decryption material remains available. Include the storage growth and maintenance work that this choice creates. The shortest workable window depends on how quickly the team detects mistakes and can start recovery.

Then test a small representative recovery with the intended emergency identity. Locate the exact version or recovery point, restore it separately and check the content. Keep the outcome, date and configuration used. Repeat after a meaningful change to credentials, retention or the storage backend. This produces a decision a team can maintain: which deletion route has been restricted, which dependencies remain and who can complete the recovery. A lock is valuable when that whole sequence is understood, not when its name replaces the sequence. [1][2][3]

Evidence & sources

Source record: · Checked:

Record limits · Evergreen comparison of deletion permissions and recovery dependencies using AWS and Borg 1.4.5 documentation checked on 16 November 2026. Operational scenarios are hypothetical editorial examples.

  1. [1]Source reported

    S3 Object Lock applies retention to individual object versions with distinct governance and compliance permissions.

    Supporting excerpt: S3 Object Lock versions and retention modes

    Original quotation

    individual object versions
    S3 Object Lock versions and retention modes ↗

    Source language: EN · Captured Nov 16, 2026

  2. [2]Source reported

    Object Lock does not preserve access to a deleted encryption key.

    Supporting excerpt: S3 Object Lock encryption and operational considerations

    Original quotation

    encryption keys
    S3 Object Lock encryption and operational considerations ↗

    Source language: EN · Captured Nov 16, 2026

  3. [3]Source reported

    Borg append-only protection concerns repository data and requires controlled server access.

    Supporting excerpt: Borg append-only mode and transaction recovery

    Original quotation

    append-only
    Borg append-only mode and transaction recovery ↗

    Source language: EN · Captured Nov 16, 2026