Cache config/workings: data loss possible?

[Mark12]

Hi folks,

 

I just finished configuring my Unraid server. I included a 256GB nvme cache drive.

That gives me the error that it is "unprotected". My first thought was: of course, why would I need to protect a cache/buffer.

 

But after some reading it seems "tier 1" storage is a better word for cache and "tier 2" for the normal data disks.

 

Am I correct in understanding that the cache disk is used for the actual storage of all kinds of data when set and that this data is not replicated to the array unless the mover moves it?

 

Can anyone tell my if I understand its function correctly:

 

1. A share with no cache: written to array, period. 
2. A share with cache set to "Yes": initial data is written to the cache for fast WRITING and then transferred to the array for reading. Data is vulnerable between writing and the mover action.
3. A share with cache set to "Prefer": initial data is written to they array and then transferred to cache for fast READING. I am not sure of this one: is it read from cache, but all edits are written to the array? Meaning the data is protected by the parity in the array? Or can there be data loss in between mover schedules still?
4. A share with cache set to "Only": data is only written to cache and never synced to the array. If the cache drive breaks down or gets corrupted your data is lost.

 

My array is an 8TB SSD, an 8TB HDD and a 4TB HDD. Would the cache even make sense in this scenario? Is the data written/read from the parity disk first? Or does it wait untill it is on either one of the HDD's also?

 

Thank you for your answers!

 

[itimpi]

Which drive is the parity drive?   Many people would not bother to have an SSD in the array if they also have HD as writing is slowed by the need to update both the HD and parity and the slower of the two will determine write performance.  Reading runs at the speed of the drive being read from.

 

You normally want the working files for VMs and/or Docker containers to be on a pool external to the main array for best performance.  Such pools can be made redundant by making them multi-drive.   There are also plugins available that will periodically backup VMs and or appdata (Docker files) to the array.

[JonathanM]

3 hours ago, Mark12 said:

A share with cache set to "Prefer": initial data is written to they array and then transferred to cache for fast READING. I am not sure of this one: is it read from cache, but all edits are written to the array? Meaning the data is protected by the parity in the array? Or can there be data loss in between mover schedules still?

No, cache prefer tries to keep all operations on the named pool, but can use the array for overflow if the pool runs below the minimum free space set, the mover will try to put that overflow data back on the pool when it runs, once again limited by open files and free space.

[Mark12]

Thank you for your replies. The cache prefer option is a bit clearer now.

 

I have to say: neither is an actual answer to the whole list of questions. Anyone else? Surely someone knows exactly how this works?

[itimpi]

Have you read the the online documentation accessible via the ‘Manual’ link at the bottom of the GUI or the DOCS link at the top of each forum page.?   The section covering Shares is here.   If there is something not answered there perhaps you can clarify exactly what it is that is still not clear.