ReFS merge - "copy instead of clone", then nothing

[DanielJ]

I have a new problem with ReFS merges: The job starts merging normally, but after a while it just stops without failing or ever timing out (waited for 3 days). Looking in Resource Monitor on the repository, no process is using the backup files. I killed the job and retried, and it continued the interrupted process but soon stalled again. I was monitoring the repository at that time, and the backup files being worked on just disappeared out of view, like the agent crashed or something. In the agent log I found a message about "Trying to copy block instead clone", followed by nothing but interval counter entries. I think this is interesting as I couldn't find any other occurrences of this message. I suppose it means that VBR failed to use block cloning and tries to fall back to a regular copy, but that doesn't work either.

Is this the known problem with "stalling merges in v11"? It's the first time we have seen it. We have had many problems regarding ReFS and merging, but only with performance, not like this. And we have many jobs on this server, almost all in forever forward mode.

This is in v11 P20210525. I get nothing from support other than the standard recommendations of implementing DisableHtAsyncIo and UseUnbufferedAccess, or update to 11a. I will update the server when I can find a free weekend to do it on, but I wanted to know if this is the well-known issue that is actually solved in 11a, or something else.

Ticket: 05320280

[foggy]

Hi Daniel, I believe you're referring to this issue but to confirm this (or any other issue), our engineers need to analyze the logs, we can't do this based on the forum post. Thanks!

[DanielJ]

Well, support has the logs but I get canned responses instead of analysis.