The lab is taking new cases · 9am–5:30pm, Mon–Fri Urgent? Ring 0800 6890668
SDR Southampton Data Recovery 0800 6890668 Open a case
SDR / Harder cases / VMware & virtual machine recovery

Specialist · virtual machines & VMware

VMware data recovery, Southampton. One dead VM is really four faults sharing a single disguise.

A failed virtual machine (VM) can be broken at any one of four layers: the RAID layer beneath the VM, the VMFS datastore where the VM's files reside, the VMDK files themselves, or the file system within the guest operating system. The only sequence that works runs upward through all four layers, beginning at the RAID layer and rebuilding each one on images before touching the layer above it. And no, we need neither your ESXi host, nor vCenter, nor any other matching items.

Only pay if we recover your data, on most jobs Free diagnosis & quote in writing Postal intake across all of Hampshire

Speak to an engineer about it
0800 6890668

What VMware errors actually mean.

Not listed? Try the fault finder →
The complaintWhat that tells youFirst step
A VM removed from the datastoreVMDK (virtual machine disk) blocks are stored on the VMFS (VM File System) volume and remain there until something else reuses them.Freeze all new provisioning now
Cannot open the disk: the parent virtual disk has been modifiedThere is a problem with the CIDs in the snapshot chain: the CIDs of the child and base do not match.Never force it; chains snap
Datastore unreachable following a RAID failureStorage collapsed underneath, but as a general rule the VMFS layer above it is repairable.Start at the lowest layer
A backup run has left a locked ghost snapshot behindConsolidation stopped and orphaned the delta filesFrequent, and usually fixable
Hyper-V guest dead once a checkpoint mergedThe AVHDX chain is broken, and all of its structures are binary, not textCheck every parent link first
VMDK recovered, yet Windows inside it will not bootSo we're at layer four: the guest's file system needs workPartly finished is not finished
Posting it: first, get a price quotation from us, by phone or with the form. Then send your item to us via tracked, fully insured post to our secure intake laboratory where it will be processed (we pay for return postage). Alternatively, you may also bring it to our Guildford intake laboratory Monday to Friday, 9am to 5:30pm, no appointment needed. Ring us and we would be happy to assist you with packaging, and there are postage details available on the postal details webpage.

Four VMware layers, working upwards.

Layer 1 — physical storageThe disks or RAID that everything else rides on. We run them through our server and RAID 5 procedures before anything else, because no datastore repair will hold up over a block device that lies.
Layer 2 — the VMFS datastoreVMware's own clustered file system, read from the image at its fixed structures: small file blocks of one megabyte (on current versions), large ones of 512MB. Deleted machines turn up here: they do not get "wiped", they just get "unlinked".
Layer 3 — descriptor and extentA virtual disk appears as a tiny text descriptor along with a gigantic flat or delta extent. It doesn't take much; one bad line in that descriptor and your hypervisor will reject a perfectly sound disk, so we always edit descriptors and never touch flat files.
Layer 4 — inside the guestThe recovered VMDK has a typical NTFS, ext4 or XFS file system that has typical failures. It's treated like an "ordinary" file system on a copy. Only after that are we able to say that "recovered" means bootable.

How the recovery runs, step by step.

01

Logged in, checked at no cost Free

Your item will have a unique case number assigned to it immediately after we receive it. An engineer will diagnose the issue, tell you what can genuinely be recovered and then provide you with one fixed price in writing. The diagnosis itself is completely at no cost to you. You are under no obligation to allow us to do any work that would incur a charge.

No-cost diagnosisQuote fixed in writingNothing to commit to
02

Settle the layer underneath

The physical disk (array) is imaged and, if the array requires reconstruction, we do so as per the procedures outlined in our server pages; everything above then works only from these read-only images.

Storage imaged at the outsetArray reassembled in software
03

Read the structures, lift the disks

From an image, we parse the VMFS or NTFS datastore. We identify all of the VMDKs (live and deleted) in the image. Before anything is stitched together, we map the snapshot chains by validating CIDs and parent links.

VMDKs lifted outChains checked rather than forced
04

Mounting first, booting after

Once descriptors have been put right, chains are consolidated properly and finally the guest's file system is addressed; the job is complete when the machine's data can be mounted successfully, not once files simply exist.

Descriptors put rightGuest confirmed to mount
05

Checked, returned, signed off

Once you have the complete listing of all the recoverable files back from us, you can "sign-off" and your recovery fee becomes due. Once you've signed-off, all the files will be posted out to you on new media (with postage included) and the case will remain open until you confirm that they can be opened properly at your location.

Sign-off on the file listNew media providedReturn postage on us

First checks on the bench

  • Deleting only unlinks: when you remove a VM on an idle datastore, it stays perfect; on a busy datastore, the next provisioning job lands right on the same space where your original VM was. Freeze it now and be sad later.
  • The descriptor governs: a couple of kilobytes of text determine whether terabytes will mount. We can fix those descriptors. The flat extents behind them are never touched.
  • Backup software can be considered a co-conspirator: jobs which stop halfway leave orphaned snapshot files as well as phantom locks behind them. They are a significant cause of what is commonly referred to as 'VMware corruption'.
  • VHDX stores failure records for itself: since Hyper-V has internal logging, in most cases (especially loss of power) you will have your answer by simply reading the log rather than guessing what happened.

Why bottom-up is not just being picky: the data of a virtual machine goes through four "owners" (physical blocks, VMFS, VMDK, and finally the guest file system) to get from raw disk data to usable files. If you repair the wrong owner, you will write fiction into all the layers on top of it. The only correct sequence of operations is dull and non-negotiable: image, rebuild, parse, extract, repair, mount. Always that way round, because it is the only order where nothing can be destroyed.

Out of the casebook.

EX · SDR-2026-0846SIGNED OFF ✓

A Friday deletion of two live VMs, with nothing written over them

An administrator ran a clean-up script that deleted the wrong directory. Nothing was written to the datastore over the weekend, so both VMDKs were left unlinked but intact. We retrieved the two VMDKs from the VMFS image on Monday, rebuilt the descriptor files, and had both systems up and running in a sandbox by Tuesday evening; one of them was payroll.

Both VMs started3 days on the bench

While it's still with you.

Do

  • Stop any additional data being written to this datastore.
  • Record which backup product was running and the snapshot history
  • Get a quote from us first, after which you can either post every disk behind the LUN or drop them off at the Guildford intake lab, labelled by server bay, or send them to us as images
  • Identify what the guests are running; that determines how we validate

Avoid

  • Put anything fresh onto the datastore involved
  • Edit a -flat.vmdk by hand, ever
  • Tidy up by doing a forced consolidation or removing snapshots
  • Take a recovered file to mean you have a working machine

Bench questions, straight answers.

Can a deleted virtual machine be restored?

Yes, often. When you remove a VM you detach its files from the datastore index, however the blocks remain in the datastore until they are provisioned again. Freeze the datastore and take an image of the storage, and whole VMDKs can normally be parsed back out. How quickly you freeze is the key.

'The parent virtual disk has been modified' — what does that mean?

Think of it like a VMware snapshot chain failing a paternity test: the parent ID in the child delta no longer matches the base disk, which usually happens after an operation such as a consolidation or restore only partially completes. The repair occurs at descriptor level, and simply making the parent IDs match again, without knowing why they stopped matching, is how a healthy chain of snapshots becomes lost data.

Must you have our ESXi host or vCenter?

No. All this takes place on our bench using images of the storage beneath: VMFS is read, the VMDKs are pulled out, guests repaired, and nothing is booted up on your hardware until you tell us to do so. A dead host makes no difference to what comes back.

Does Hyper-V work the same way?

The thinking is the same, but the syntax differs: VHDX contains internal binary logs, whereas AVHDX checkpoint chains contain parent links which need to be verified through PowerShell (not a text editor) before anything is merged. The "layers first, images only" rule still applies.

Whatever has gone wrong, keep it switched off.

Powering a damaged device up again costs you data. Start by opening a case instead; diagnosis is always free, regardless of your decision.

0800 6890668