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

NetLeverage
NetLeverage

Hyper-V backup reads the whole disk every time: resilient change tracking (RCT) reset or unavailable

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.

An incremental Hyper-V backup works only when the backup tool can ask Hyper-V which blocks of a virtual disk changed since the previous backup. Hyper-V answers from resilient change tracking (RCT) and a reference point left by that previous backup. If tracking is not available for a disk, the tool cannot get a list of changes from Hyper-V, and what it does next depends on the product. Veeam says that with change tracking disabled it reads all VM data to detect what changed. A missing reference point is product-specific behaviour: neither Microsoft nor Veeam documents it. "My incremental runs as a full" describes that outcome.

The conditions named below sit on the host, the cluster, the virtual machine or its disks.

What the error means

Microsoft's backup guidance for Hyper-V says that from Windows Server 2016, Hyper-V supports backup through its WMI interface, where "a combination of reference points and resilient change tracking (RCT) is used to allow developers to access the information about backed up virtual machines in an efficient manner". Backup software reads changes with the QueryChangesVirtualDisk call, which "retrieves information about changes to the specified areas of a virtual hard disk (VHD) that are tracked by resilient change tracking (RCT)", measured from a change tracking identifier that marks the earlier state. The WMI class that creates and removes reference points has the methods CreateReferencePoint and DestroyReferencePoint.

Veeam's Hyper-V guide describes the flow in the same terms. The product creates a checkpoint, and before the checkpoint is merged into the base disk Hyper-V converts it to a reference point. At the next incremental, it "queries Microsoft Hyper-V to get incremental changes between the reference point created during the previous job session and the checkpoint created during the current job session".

To keep tracking data after interruptions, Veeam's guide says Hyper-V keeps three bitmaps: one in memory (most detailed), an RCT file used in normal events such as a VM move, and an MRT file (coarsest) used after abnormal events such as a power loss or host crash. "RCT and MRT files are created for every VM disk and stored at the VM disk level."

Two points matter for your diagnosis. First, Veeam says the RCT file is used in normal events such as a VM move and the coarser MRT file after abnormal events such as a power loss or host crash. Neither Veeam nor Microsoft states how large the next backup is after such an event, so read your product's job log. Second, "full every time" can mean two things: the tool read the whole disk to find changes and sent only those, or it sent everything. Veeam says that when change tracking is disabled, "the backup window may increase dramatically, as Veeam Backup & Replication will read all VM data to detect what blocks have changed". Check which of the two your product does before you decide how serious it is.

Likely causes

Cause What you see How to confirm
The environment does not meet RCT requirements Change tracking is never used; every run may read the whole disk Host and cluster at 2016 or later, cluster functional level raised, VM configuration version 8.x or later
TRIM or optimise inside the guest Backup size jumps to full size after the guest runs it Microsoft documents it for Azure Backup Server; check whether the guest runs trim
The tracking files were deleted The first backup after the deletion reads everything Look for .rct and .mrt files beside each disk; check when they were created
The reference point from the last backup no longer exists Product-specific: no source documents a message for a missing base Product log; another backup product or script may have removed it
Tracking disabled in the backup product Job log says change tracking is off at host or job level Job and host settings
A disk was swapped or a VM restored The disk is a different file with no history Disk path and creation time in the VM settings

Prove the cause before changing anything

All of these are read-only.

Version and level. Microsoft's "Upgrade virtual machine version" page gives the command to list each machine's configuration version:

Get-VM * | Format-Table Name, Version

Veeam requires VMs on Hyper-V Server 2016 or later, and a VM configuration version of 8.x or later. For a cluster, Veeam says that if even one node is older than 2016, change tracking is not used, and the cluster functional level must be raised. The cluster rolling upgrade page gives:

Get-Cluster | Select ClusterFunctionalLevel

Compare the host's supported versions with Get-VMHostSupportedVersion, which that page documents.

Tracking files. On the host, look in the folder of each virtual hard disk for .rct and .mrt files with the disk's name. Microsoft's MPIO troubleshooting page names these files as "associated with" the virtual hard disks. If the files have a recent creation time and the first slow run followed it, tracking was restarted then. That is a reasoned reading; no source states it.

Guest TRIM. Microsoft's Azure Backup Server page says that "when the trim is performed within the guest OS, the tracking of incremental blocks is reset, resulting in a full backup", because it "changes the SequenceNumber of the tracked incremental blocks". A reset that repeats every night points to a scheduled trim or optimise job inside the guest.

Product settings and log. Read the backup job's log for the line that describes the transfer: whether tracking was used, whether the product reports a missing reference point, and how much data was read compared with how much was sent.

Fix

Requirements not met. Upgrade the old node, raise the cluster functional level, and upgrade the VM configuration version only when you are sure you will not move the VM back to an older host.

Disruptive step: Update-ClusterFunctionalLevel cannot be undone. Microsoft says that after you update, "you can't go back to an earlier functional level... and you can't add nodes running earlier versions of Windows Server to the cluster". The page names a System State backup of the cluster database as the way to recover the original configuration.

Disruptive step: Update-VMVersion <vmname> needs the VM shut down (the page tells you to shut it down) and cannot be reversed: "You can't downgrade a virtual machine configuration version after you've upgraded it". Microsoft says you cannot import a virtual machine created on a newer version of Hyper-V to an older version of Hyper-V.

Guest TRIM. Microsoft recommends stopping the guest trim "unless the purpose is to improve the efficiency of storage on the Hyper-V host side", to avoid the increase in backup size. Decide which matters more for this machine. Turning trim off means the VHDX file will not shrink on the host.

Tracking files deleted or reference point missing. Microsoft and Veeam document no repair step for the tracking files, so check your product's guide for how it starts a new base. Allow for one run that may read the whole disk, schedule it for a quiet window, and make sure only one backup product is managing reference points for that VM.

Tracking turned off in the product. Turn it back on in the job or host setting. Veeam says it can be disabled "for troubleshooting purposes".

Files deleted by the Microsoft workaround. Microsoft's MPIO page, in its section "Windows Server 2022 or 2019: VM IO performance degrades when Hyper-V uses resilient change tracking (RCT)", tells you to delete the .rct and .mrt files if VMs previously backed up by a host-level application cannot start after the October 23, 2025 update (KB5070884, OS Build 20348.4297) and subsequent updates. Microsoft does not say what happens to the files afterwards. The expectation, not documented, is that they are recreated and the next backup has no base, so it reads the whole disk.

Disruptive step: deleting .rct and .mrt files discards change tracking history for that disk. Do it only when the VM cannot start and Microsoft's instructions apply to you, and expect one whole-disk backup afterwards. As a safer way back, move the files to another folder instead of deleting them, and put them back only if Microsoft Support tells you to; Microsoft's page says "If the issue persists, contact Microsoft Support." Once deleted, the tracking data cannot be restored. If Microsoft's own instructions do not apply to you, do not delete these files to "reset" a backup.

Example (illustrative): VM app01 has a 500 GB virtual disk and about 10 GB changes a day. A normal incremental reads and sends about 10 GB. Over a 1 Gbit/s link, which is at best 125 MB/s, 10 GB (10,000 MB) takes 10,000 ÷ 125 = 80 seconds. A run that has to read all 500 GB (500,000 MB) takes 500,000 ÷ 125 = 4,000 seconds, or about 67 minutes, and reads 500 ÷ 10 = 50 times as much. Real rates are lower than the line rate. If your window is 60 minutes, a nightly reset makes the job overrun every night.

Confirm the fix

  • The next incremental after the base run reads and sends about the daily change, not the disk size. Compare the job's data-read figure with the previous day's.
  • The job log shows change tracking in use for each disk, with no message about a missing reference point.
  • After a full week with the guest trim schedule in place, the sizes stay stable.
  • Run a restore test of a point after the fix. Incremental speed says nothing about whether the data restores.

What to watch afterwards

  • Daily data read against disk size, per VM. A sudden step to the full size is the signal to read the log.
  • Anything that touches .rct, .mrt or checkpoints: updates, storage migrations, and second backup tools on the same host.
  • Cluster maintenance. In Veeam's description, one node left on a version older than 2016 means change tracking is not used for the cluster.
  • Guest scheduled tasks that trim or optimise the disk.

Risks and limits

  • Microsoft's pages describe RCT as a feature of Windows Server 2016 and later. Hyper-V on earlier versions used other methods that are not covered here.
  • Microsoft's documentation does not list every condition that resets tracking. Those named here come from Microsoft's Azure Backup Server and MPIO pages and from Veeam's guide. Other backup products may react differently.
  • The reference to a competing backup product removing a reference point is a possibility from how reference points work, not a documented case.
  • Veeam's description of reading all data without tracking is for Veeam. Another product may transfer the whole disk.
  • Version upgrades and functional level changes cannot be undone. Plan them in a maintenance window with a tested backup.
  • These commands have not been run or lab-tested.

Sources