Azure Windows Server 2025 freezing with ReFS disk

Question

Hello,

I recently migrated a storage server in Azure from Windows Server 2019 to a new Server 2025 instance. The old VM had a large disk formated as ReFS which was disconnected and mounted to the 2025 server. With the disk attached the 2025 server will run for any where between a few minutes and a week before something consumes the processor and it stops responding. I can't tell what since task manager stops updating. The machine continues to respond to a ping but all services essentially stop functioning. If RDP is connected when this happens the connection continues but becomes unresponsive or exceptionally slow responding. I tried building serveral replacement server 2025 VMs but they all develop this issue. I built a 2022 server and mounted the disk but mounting it to server 2025 seems to have altered in some way that prevents it from working. The disk appears unformated to Windows Server 2022.

I'm at a loss as to what to do next as the disk is too large to download with an insane expense.

Anyone else run into this?

Thanks

Answer 1

Hello Brett Miller,

this behavior seems to be a known set of problems around ReFS volumes in Windows Server 2025, particularly when ReFS is used with large data volumes and migrated across OS versions. You're seeing CPU spike to 100%, Task Manager freeze, RDP remain active but unresponsive, and all services effectively hang with no clean shutdown or crash. This issue typically does not produce any event log entries or BSOD, making it even harder to diagnose.

The core issue lies with ReFS (Resilient File System) in Windows Server 2025. When you attach the volume from a Server 2019 instance, the OS may automatically upgrade the internal ReFS version, which introduces two key problems:

1. It becomes unreadable on Windows Server 2022 or earlier, as the updated volume format is not backward compatible. That’s why it appears unformatted when you mount it on Server 2022.
2. On Windows Server 2025, this upgraded ReFS version has known issues with memory management and metadata handling, which under load can lead to exactly the kind of system-wide hangs you're observing.

This is documented in multiple threads, including this one:

ReFS volume appears RAW – version mismatch after update

Additionally, Microsoft has published guidance to address memory usage problems caused by ReFS:

Fix heavy memory usage by ReFS

Please try the following registry-based tuning workaround, which helps limit the working set size and trim memory pressure from ReFS operations:

On the affected Windows Server 2025 VM, open Registry Editor (regedit) or use the following command in an elevated PowerShell or Command Prompt:

reg add "HKLM\SYSTEM\CurrentControlSet\Control\FileSystem" /v RefsEnableLargeWorkingSetTrim /t REG_DWORD /d 1 /f

reg add "HKLM\SYSTEM\CurrentControlSet\Control\FileSystem" /v RefsNumberOfChunksToTrim /t REG_DWORD /d 4 /f

Restart the VM to apply these changes. You can run the following command to check the ReFS version on the disk

fsutil fsinfo refsinfo <drive_letter>:

This will confirm if the volume has been upgraded to a newer format.

These settings help Windows actively manage memory usage related to ReFS metadata. They were introduced in earlier hotfixes but continue to be relevant for larger ReFS volumes, especially in Azure-hosted workloads.

If the issue persists, share the memory dump with me when it becomes unresponsive. I will have a look into it from backend. Thanks