JorgeB

If all 3 disks are exact copies and parity is valid yes, also disks need to be same capacity, can't be larger, then you'll need to do a new config and use the invalid slot command, if you get 3 drives working let us know and I'll post the instructions.

Disk is on a controller that only supports 2.2TB max, and it appears that because of that Unraid doesn't consider it a valid partition, best bet is to connect that disk to a controller that supports the full disk capacity, but you'll need to rebuild it again.

Not quite clear what you mean, please try to describe better what you're trying to do, also post the diagnostics: Tools -> Diagnostics

Everything VM/docker related is inside the same share on my cache drive with COW set to auto, that folder is actually a btrfs subvolume so it can get snapshotted and replicated to a different pool daily by a script, this reminds me that the COW setting can't be the only thing causing the write augmentation for the VMs, since all 3 vdisks are together and just the Windows Server has the high writes issues, both other Windows VMs have normal writes according to iotop.

Speed being the same suggests a problem with the source, do you have another computer you could test width? Ideally one running Windows 10 since there are some reports of lower SMB performance for some with Linux or OSX.

Doesn't look like it is, enable turbo write and transfer directly to array, is the speed better or same?

You need iperf3 to iperf3 or iperf2 to iperf2, it won't work with different versions.

Yes, mine are all set to auto, but I want them like that or else btrfs will also stop checksumming the data, and that's more important for me, even if it I have to live with the extra writes, because of this feature I detected silent corruption on an SSD I was using a few years ago with a VM, but you are correct, most users should have it off for vdisks, because of increased fragmentation and likely the additional writes.

Start by running a single stream iperf test to make sure network is working correctly.

Click on the share then check "delete" (it needs to be empty) That will do it.

Just one more screenshot from me on this, I left iotop running since earlier and you can see it's not even the VMs in general, for me it's mostly the Windows Server VM, which is the main one but it was mostly idle during this, loop2 has few writes comparatively, and libvirt (loop3) doesn't even appear on the list. I'm not sure if this is related to this topic, but I have been noticing the unusually large amount of writes to my cache device for some time, it's writing on average 1 or 2TBs per day, some days more, just never though too much of it.

Please post the diagnostics: Tools -> Diagnostics

libvirt.img is already on the cache device on my server.

Most of my writes are also from my 3 Windows VMs, I also have dockers, but there's not much writing going to loop2, at least comparatively, iotop accumulated writes after a couple of minutes:

Disk1 appears to be failing, please post a SMART report.

There's a little activity on disk1 but doesn't look like anything out of the ordinary considering system share exists there, but if you want install nerdpack and use iotop to see exactly whats reading/writing to the disk.

You are right, that behavior was changed not that long ago, but don't recall exactly when, tagged solved.

Should be fine, and if there are any lost/partial files xfs_repair usually places them on a lost+folder found on that disk.

Run again without -n or nothing will be done. If it asks for it use -L

Please post diags.

It's up to you, if you cancel the rebuild you'll need to start over, also depends what percentage is done.

Check filesystem on disk2: https://wiki.unraid.net/Check_Disk_Filesystems#Checking_and_fixing_drives_in_the_webGui

Please post the diagnostics: Tools -> Diagnostics

Note: Moved to lounge for now since this isn't a general support issue, I can later move it to the Guides sub-forum if you want to create a guide, or you can just create a new thread there.

Yes, but it's still the most logical cause of the problem, it certainly isn't software related.