What exactly happens when a drive in a BTRFS cache pool fails?

[drumstyx]

Long story short, I had a cache drive failure (ADATA nvme ssd, quite a surprise) and after recovery I'm looking to improve reliability. I've picked up a 2TB Samsung NVMe drive to complement the ADATA warranty replacement when it arrives (also 2tb)

 

Thing is, I know the process when an array drive fails, but I have no idea what happens when a BTRFS drive fails in a pool. I've read some horror stories about a failed drive causing an unrecoverable situation. Of course I back up important stuff to the array (and important stuff on the array is backed up offsite) but recovering from that is still a pain in the butt.

 

So, can anyone walk me through what a BTRFS cache pool drive failure looks like?

[JorgeB]

Assuming the pool is redundant when a device fails/drops offline the other one continues to work, it's important to monitor the pool to act as soon as possible, if the device failed you can replace it to rebuild the mirror, if it dropped offline you can bring it online then run a scrub to put it back in sync, note that btrfs can only repair the data if COW is enable, for any shares with COW disable, and this was the default for the system and domain shares before 6.10.0, it won't be able to sync the dropped device to due NOCOW also disabling data checksums.

 

There are some corner cases, especially when a device drops and comes back online that can cause some issues, but that can usually be solved, and that's also why it's important to monitor the pools, to minimize that risk.