Expanding a RAIDZ with a set of different sized drives.

[alacaza]

Hi!

Yesterday, I expanded a RAIDZ volume composed by three drives of 10G each adding a vdev composed by another three drives. These drives have 20G each.
I was surprised when I get 60G of storage. I've just lost a 33% because of parity.

I've said, I was surprised because I thought that you cannot expand a RAIDZ with different sized drives, or better said, you can but I thought that in the concrete example I put, I would only get 40G of storage, because the original RAIDZ was composed by 10G drives.

So, assuming that you only lose storage necessary for assuring parity when you expand a RAIDZ, indepedently of the overall storage of the new vdev, I would like to ask you:

- What are the drawbacks of that scheme? Performance, maybe?
- How many drives can fail and where, before I lose data?

Thanks in advance

 

[Artion]

- How many drives can fail and where, before I lose data?

Thanks in advance

It's a per vdev property. If your vdevs are raidz1, you can loose one disk per vdev. If both are from the same vdev this vdev become unavailable and so the pool, and the data is gone.

(ZFS stripes data over vdevs, that means that pieces of data are spread over each vdev. If one fails, ZFS can not rebuild it. But each vdev keeps its data backed by metadata through its own member disks, so if one or more of them, depending on its zfs configuration, it can rebuild its data)

 

[danb35]

I thought that you cannot expand a RAIDZ with different sized drives,

You thought incorrectly. You've already been pointed to resources that would help you understand better how ZFS works. You really should read them; some of its characteristics can bite you if you aren't aware of them.