Is it possible to have a folder in cache drive that is not part of the array?

[TheFreemancer]

I replaced my 240GB SATA SSD with a 1TB M.2 NVME drive and I use it as cache drive. 

This cache drive is used for appdata, system, isos, VM img, docker img, all good part of the unraid system. 

 

But I also used it for partial downloads (incomplete folder) and for rclone cache. 

Now here's the problem:

 

root@Tower:/tmp# dd if=/dev/zero of=/mnt/user/INCOMPLETE/test1.img bs=3G count=1 oflag=dsync
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB, 2.0 GiB) copied, 4.36276 s, 492 MB/s

 

and if it's outside of the array:

 

root@Tower:/tmp# dd if=/dev/zero of=/mnt/cache/INCOMPLETE/test1.img bs=3G count=1 oflag=dsync
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB, 2.0 GiB) copied, 1.68525 s, 1.3 GB/s

 

 

How can I still have the INCOMPLETE folder but don't have it inside the array?

 

[Kilrah]

Set the INCOMPLETE share to be Cache: Only.

Edited October 19, 2022 by Kilrah

[TheFreemancer]

5 minutes ago, Kilrah said:

Set the INCOMPLETE share to be Cache: Only.

 

It is set to only. Even tried prefer. 

 

As I understand this only changes how the mover will treat the folder. 

I want it to be outside of the array but still on the cache drive so I don't have to buy another NVME SSD just for this.

[Kilrah]

Only will not move anything but it will also put anything new on the cache. There is nothing that will end up on the array that isn't already there if a share is set to cache only and you write to /mnt/user/[share].

Edited October 19, 2022 by Kilrah

[JorgeB]

Cache is always outside the array, you can use the disk share for better performance, map to /mnt/cache inside the server or enable disk shares for SMB.

[TheFreemancer]

6 minutes ago, JorgeB said:

Cache is always outside the array, you can use the disk share for better performance, map to /mnt/cache inside the server or enable disk shares for SMB.

 

Disk share would be the case I need better performance from another device on my LAN. That's a good info there.

 

Mapping to /mnt/cache seems to be the answer. 

On Sabnzbd I've mapped to /mnt/cache/INCOMPLETE and the most performance difference I'm seeing is instead of using /mnt/user/rclone_cache is using /mnt/cache/rclone_cache/

 

Is there any complication that can come from using direct disk access instead of the user folder in the cache drive?

I'm aware this is a problem when writting files to disks that are indeed part of the array.

[Kilrah]

Writing directly to /mnt/cache is fine BUT AFAIK you won't benefit from the "min free space" protection to avoid filling it up to the brim, which should really be avoided.

[TheFreemancer]

That's fine, and thanks for the info. 

 

As I use it mostly for incomplete downloads from sabnzbd and for the rclone cache, which is already limited in size, this is fine. 

[TheFreemancer]

Sorry to revive this topic but I need to ask one more question.

 

I have replaced my kingston nv1 snvs/1000g for a adata XPG gammix s11 Pro that I was using in another PC.

The XPG is much faster compared to the Kingston, I was getting 3GB/s with it on Windows.

 

The m.2 connection on this motherboard is PCIE 3.0 X4 so it should max out at 32gbit/s and from previews tests with the kingston I was getting 1.3GB/s if writting directly to the cache.

 

After replacing it:

 

root@Tower:~# dd if=/dev/zero of=/mnt/cache/INCOMPLETE/test1.img bs=3G count=1 oflag=dsync
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB, 2.0 GiB) copied, 2.32709 s, 923 MB/s

 

Much slower. 

 

Before I try anything else, this cache drive is formated as BTRFS and I don't remember if the previews one was XFS. 

 

Should I go XFS for this drive to max out it's performance? Or file system matters little in this case?

[TheFreemancer]

Answering my own question XFS is faster compared to BTRFS:

 

root@Tower:/mnt/cache# dd if=/dev/zero of=/mnt/cache/test2.img bs=3G count=1 oflag=dsync
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB, 2.0 GiB) copied, 1.59658 s, 1.3 GB/s

 

and doing a different test:

root@Tower:/mnt/cache# dd if=/dev/zero of=/mnt/cache/test.img bs=1M count=5120
5120+0 records in
5120+0 records out
5368709120 bytes (5.4 GB, 5.0 GiB) copied, 2.34594 s, 2.3 GB/s

 

Almost the same speed I was getting on windows.