Data disk corruption

Question

Data disk corruption

Steve Lee 0

I have had two data disks that are attached to multiple machines become corrupted within a couple of days of deploying them. It presents as users not able to access a folder, and when I run a chkdsk on them, it shows up as a RAW format.

The first drive I created lasted about 1 day before becoming corrupt, and the second lasted about 2 days.

Himanshu Shekhar 4,025 Reputation points Microsoft External Staff Moderator

2026-02-16T17:33:34.8266667+00:00
@Steve Lee

The same managed data disk is being attached directly to multiple virtual machines. This results in concurrent block-level access to the disk. Your geoscience application is not cluster-aware and does not coordinate write operations between systems. As a result, multiple operating systems attempt to manage the same filesystem independently, which leads to metadata corruption over time. When the filesystem becomes inconsistent, the disk appears as RAW.

Azure Shared Disks are supported only for specific clustered scenarios, such as Windows Server Failover Clustering, where both the operating system and the application are designed to coordinate disk access using supported cluster-aware filesystems. Outside of these supported configurations, attaching a single managed disk to multiple VMs for simultaneous read and write operations is not supported and will result in corruption.

Your alternate configuration works because only one virtual machine performs direct read and write operations against the disk. The remaining machines access the data over the network using SMB or NFS protocols. In that model, the filesystem is managed by a single operating system instance, which prevents corruption.

To move forward safely, you should redesign the storage approach using one of the following supported methods.

Use a dedicated file server and expose the storage over SMB or NFS.

Use Azure Files or Azure NetApp Files for shared storage access.

Assign each virtual machine its own high performance data disk and consolidate results at the application or data layer.

Directly attaching the same managed disk to multiple virtual machines should only be done when implementing a fully supported clustering solution with proper coordination.

The key point is that storage performance does not change the fundamental requirement for coordinated access. High performance disks do not make concurrent block-level attachment safe unless the workload is specifically designed and supported for shared block storage.

Microsoft References :

Azure Shared Disks overview: https://learn.microsoft.com/azure/virtual-machines/disks-shared

Supported scenarios and limitations: https://learn.microsoft.com/azure/virtual-machines/disks-shared#limitations

1 answer

Your answer

Answer 1

TP 151.7K Volunteer Moderator

Hi Steve,

I assume you are using the shared disk feature whereby you can mount a managed disk on multiple VMs. Please correct me if my assumption is wrong.

Please note that shared disks are typically for clustered workloads, for example, Windows Server Failover Cluster. The workload needs to be designed to handle writes from multiple hosts to single disk properly.

Which specific workload are you running on the VMs?

Please click Accept Answer and upvote if the above was helpful.

Thanks.

-TP

Steve Lee 0 Reputation points

2026-02-16T17:04:33.43+00:00

HI TP,

It is a setup for geoscience processing that has the disk attached to 4 VM's. We have another data disc that is attached to a file server and mapped to the individual machines through it, that we have not had any issues with. The data for this new drive needs to be high performance, so it is a higher tier drive and than the other data drive, and I thought attaching it directly would be a better performance option. That being said, my experience in Azure VM's environment setup is limited.
TP 151.7K Reputation points Volunteer Moderator

2026-02-16T18:07:01.6433333+00:00

Okay, it sounds like the geoscience software isn't designed for this scenario. Vast majority of software is not designed for this type of low-level disk sharing, so it is normal.

It sounds like you need to use similar technique as you are using with the other data disk. When you have a disk attached to a VM that is a file server and then other VM's use it via mapped drive, only the one VM (the file server) is actually reading/writing to the disk.

In your scenario now, you've got 4 different VMs trying to write to same disk. For that to work you need software designed specifically for it, which is part of what clustering solution provides.

If performance of using file server is not enough, would it be possible to give each VM their own high performance attached data disk? I don't know how the workflow works, but if each VM could take a "chunk" of data that needs to be processed, have it on its own disk, and then copy the results to file server? Just thinking out loud.
Himanshu Shekhar 4,025 Reputation points Microsoft External Staff Moderator

2026-02-18T19:30:26.29+00:00

@Steve Lee

Just checking if provided response was helpful as suggested by TP! please let me know if you have any queries.

Share via

Data disk corruption

1 answer

Your answer