Cache pool disk replacement went wrong?

[hawihoney]

My Cache pool consists of two NVMe M.2 SSDs. I need to increase them. So I thought stop the Server, replace one of the two disks and start the Server.

 

To my surprise the array comes up (different from replacing array drives). It shows a missing disk but I can't stop the array. Balance running is the message. What is balancing what? There's only one disk of the two-disk cache pool left. The new cache disk shows up under Unassigned devices.

 

What went wrong?

 

Any help is highly appreciated.

 

***EDIT*** Screenshot attached, values of Cache pool are complete nonsense now. 250 GB disk shows 500 GB. Empty space is now 64 GB, it was 10 GB. Is balancing deleting the Cache disk that is left?

 

 

tower-diagnostics-20200902-1011.zip

Edited September 2, 2020 by hawihoney

[JorgeB]

It's balancing the pool to a single device, you can then add the new one, but the best way of doing this would be a direct replacement, i.e., stop array, select new cache device in the place of the old one, start array.

[hawihoney]

I can't stop. BTRFS operation in progress. Stop button unavailable.

 

What did I wrong? I stopped the server, replaced the disk and started the server. Cache pool started with missing disk and is pondering the remaining cache disk.

 

I don't understand that. Consider a cache pool with a failing disk, and the first thing that BTRFS does is hammering the remaining good disk? This disk is at 62 degree currently during that balance operation.

 

I fear the good disk will fail too shortly.

 

 

Edited September 2, 2020 by hawihoney

[hawihoney]

One hour later. The balance of the remaining disk took an hour.

 

Stopped array, set new disk in second slot, and started array. What happens now? BTRFS is again writing at high speed to the remaining disk plus writing to the new disk.

 

Sorry BTRFS, this is a good laugh. During replacement of a (failing) disk BTRFS _writes_ at high speed to the remaining good disk _twice_. I would expect to read from the remaining disk and write to the new disk as the array does. I have no FS knowledge but this can't be serious.

 

[ChatNoir]

I don't understand what this is such a big issue for you.

 

I understand what you say but this is the same thing that happens to a RAID or unRAID array when a drive is failing. Whether simulating or rebuilding a failed drive, every other drive is solicitated heavily with obvious risks.

[hawihoney]

There's a difference.

 

Unraid array, disk fails:

Stop server, replace array disk, start server

Array will not be started, set new disk in place, start array

All disks except new disk are read, the new disk is written

 

Unraid cache pool, disk fails:

Stop server, replace cache pool disk, start server

Array/pool will start up with missing cache disk

Remaining cache disk is written, stop not possible

When balance finished, stop array

Set new disk in place, start array

All disks are written

 

Don't you see the difference?

 

The array reads all remaining disks once and writes to the new disk once.

The cache pool reads all remainig disks twice and writes to the new disk twice.

 

Edited September 2, 2020 by hawihoney

[JorgeB]

38 minutes ago, hawihoney said:

I can't stop. BTRFS operation in progress. Stop button unavailable.

You can only stop after the balance finishes.

[JorgeB]

Just now, hawihoney said:

The cache pool reads all remainig disks twice and writes to the new disk twice.

Not if you do like I mentioned above.

[hawihoney]

I don't get that. What would you do exactly?

 

I stopped the server, replaced the cache pool disk and started server. The array starts immediately and cache pool starts immediately with balance of remaining disk. No chance to select the new cache pool disk. New disk is added to Unassigned devices instead.

 

What's wrong, what should I do?

 

***EDIT*** I see. What you suggest is not possible. I'm talking about two NVMe M.2 SSDs, for two available slots. I can't add a third one.

 

Edited September 2, 2020 by hawihoney

[JorgeB]

2 minutes ago, hawihoney said:

What you suggest is not possible. I'm talking about two NVMe M.2 SSDs, for two available slots. I can't add a third one.

Correct, direct replacement is only possible if you can have both the old and new devices connected, more info in the FAQ.