Agentless multi-disk crash-consistent incremental snapshots: alternatives to VM restore points?

cbt2 0 Reputation points
2026-11-20T02:59:49.39+00:00

We build a VM backup product on Azure. Today we take an incremental snapshot of each managed disk one by one,

then read the changed blocks with list_page_ranges(prev_snapshot_url=...). Because the disks are

snapshotted at different times, a multi-disk VM (for example a database with data and logs on separate disks)

is not consistent across disks.

What we need

  • Point-in-time crash consistency across all (or selected) disks of one VM
  • API only: no VM agent, no extension, no in-guest scripts
  • Incremental: changed blocks between backups, ideally on the same chain as regular incremental disk

snapshots

  • No more restrictions than regular incremental disk snapshots: any VM size, disk type, encryption setting,

or shared disk that supports incremental snapshots should work

What we tried with CrashConsistent VM restore points

They give us consistency, and diffs between disk restore points work. But they don't meet the requirements

above:

  • 3 restore points per VM per hour
  • Rejected on NVMe / Azure Boost VMs
  • Not supported on Confidential VMs or Write Accelerator disks; the whole VM is rejected when a shared disk

is attached

  • Source disk encryption (CMK / ADE) is not carried over
  • Separate incremental chain: we can't diff against regular disk snapshots

Questions

  1. Is there any other Azure mechanism, GA or preview, that creates crash-consistent snapshots of several

managed disks at the same point in time without an agent?

  1. Is there a supported way to make a disk restore point part of the disk's regular incremental snapshot

chain, or to diff a disk restore point against a regular incremental snapshot?

  1. What does Microsoft recommend for a backup solution that needs multi-disk crash consistency on

Azure?

Azure Virtual Machines
Azure Virtual Machines

An Azure service that is used to provision Windows and Linux virtual machines.

0 comments No comments

1 answer

Sort by: Oldest
  1. AI answer

    2026-11-20T03:00:55.5066667+00:00

    There is no other documented Azure mechanism that provides online, agentless, point-in-time consistency across multiple managed disks with the same applicability and incremental chain as ordinary managed-disk snapshots.

    1. Available multi-disk mechanism

    Crash-consistent VM restore points are Azure’s platform primitive for this operation. They capture a write-order-consistent point across the VM’s included disks, and individual disks can be excluded through excludeDisks.^1^

    Azure Backup’s GA agentless multidisk crash-consistent backup is not a separate snapshot primitive that removes the restore-point restrictions. It uses multidisk crash-consistent restore points under an Enhanced VM backup policy.^2^ Consequently, it does not solve the cases already ruled out:

    • Azure Boost and the listed newer VM generations
    • Ultra Disk and Premium SSD v2
    • Ephemeral OS, shared, and write-accelerated disks
    • VMs that are not premium-storage capable
    • Workloads requiring cached writes to be included reliably

    The same documentation warns that read/write host caching prevents a multi-disk consistency guarantee because cached writes might not have reached Azure Storage.^2^

    Azure Disk Backup is agentless and uses incremental snapshots, including for shared disks, but its consistency unit is one managed disk. It does not coordinate one point in time across several disks.^3^

    Ordinary managed-disk snapshots also operate per disk and do not coordinate multiple disks. Microsoft explicitly marks them unsuitable for coordinated scenarios such as disk striping.^4^

    2. Incremental-chain interoperability

    There is no documented API for inserting a disk restore point into an ordinary managed-disk incremental-snapshot chain or merging those lineages.

    The relevant API contract is based on the incremental snapshot family:

    • Ordinary incremental snapshots expose incrementalSnapshotFamilyId.
    • Get Page Range Diff can compare incremental snapshots only when they have the same family ID.^5^
    • A disk restore point separately exposes familyId, described as the backing snapshot’s MIS family.^6^

    Get Page Ranges with x-ms-previous-snapshot-url returns changes between managed-disk snapshots, but that operation does not provide a way to re-parent a snapshot, change its family, or bridge two families.^7^

    Therefore:

    1. Compare two ordinary incremental snapshots only when their incrementalSnapshotFamilyId values match.
    2. Compare disk restore points within their own supported restore-point lineage.
    3. Treat a regular snapshot chain and a restore-point chain as separate backup lineages.
    4. Do not infer interoperability merely because both resources expose a family identifier. No documented operation establishes cross-resource-family comparison or chain membership.

    3. Supported architecture choices

    Microsoft’s documented choices reduce to these:

    1. Supported online VM + agentless requirement:
      Use crash-consistent VM restore points directly, or Azure Backup Enhanced Policy configured for crash consistency. Azure Backup provides policy, retention, and recovery management, but retains the underlying support restrictions.^8^
    2. Unsupported VM/disk configuration + strict online requirement:
      No documented Azure platform API meets all the stated constraints. Per-disk snapshots or Azure Disk Backup remain incremental and agentless, but cannot guarantee a common multi-disk point.
    3. Unsupported configuration where downtime is acceptable:
      Stop/deallocate the VM, then snapshot every selected disk while writes are stopped. Microsoft documents shutdown as the straightforward way to obtain consistent disk snapshots without coordinating a running guest.^9^
    4. Online consistency where guest coordination is acceptable:
      Freeze and flush all disks before taking incremental snapshots, using mechanisms such as VSS or fsfreeze. This violates the no-agent/no-in-guest constraint and may provide filesystem rather than application consistency, depending on the mechanism.^9^
    5. Database recovery requirements:
      Use an application-specific backup or replication mechanism when transaction consistency or point-in-time recovery is required. Crash consistency is equivalent to recovery after a sudden power loss; it is not transaction-aware.^10^

    The resulting product design needs two paths: restore-point-based multi-disk protection for supported VMs, and either an explicitly crash-inconsistent per-disk mode, planned downtime, or guest/application coordination for configurations excluded from restore points.


    References

    1. Overview of Virtual Machine restore points
    2. What is agentless multidisk crash-consistent backup for Azure VMs?
    3. Overview of Azure Disk Backup
    4. Azure managed disks overview
    5. SnapshotData.IncrementalSnapshotFamilyId Property-Definition
    6. DiskRestorePointProperties interface-Property Details
    7. Get Page Ranges
    8. Back up Azure VM with agentless multidisk crash-consistent backup
    9. Backup and disaster recovery for Azure managed disks
    10. Reliability in Azure Disk Storage
    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.