Bdtyktl

Currently I have a Ubuntu server setup with 2 4tb disks each with ext4 in a RAID 1 (not using zfs). I want to have complete redundancy in case of a drive failure.

I am adding another 2 4tb drives to the machine. I'll probably do this again in the future so I'm thinking I should switch to zfs to support this.

I read on some guides that you can do both striping and mirroring with zfs, which I think is the way I want to go.

Normally I would stripe the drives into two pairs, and then raid 1 the two pairs to get a fast redundant "8tb" drive.

How would I accomplish this in zfs? And would that allow for an easy addition of more drives down the line?

It's super easy to configure that (line-splitting to make the structure easier to see):

# zpool create <poolname> 

    mirror <disk1> <disk2> 

    mirror <disk3> <disk4>

This will create a pool with two stripes, each of which is a mirror between two disks. This is a nice flexible setup (easy to add new drives, easy to understand). To add another mirror in the future, you simply do:

# zpool add -n <poolname> 

    mirror <disk5> <disk6>

would update '<poolname>' to the following configuration:

  <poolname>

    mirror

        <disk1>

        <disk2>

    mirror

        <disk3>

        <disk4>

    mirror

        <disk5>

        <disk6>

(Using option -n does a dry run first and prints the pool configuration for you to inspect, which I would always recommend to avoid misconfigurations that are hard / impossible to back out of after the fact!)

However, note that this setup is "inefficient" in that it only guarantees data safety for one disk failure: you're only safe for two failures when they happen on different stripes, which happens in ~67% of two-disk failure scenarios. Also, the redundant data still uses 50% of your pool.

There is a less-flexible ZFS-specific RAID format called RAID-Z that allows you to pool all your disks and get efficient data protection. (I say less flexible because it is difficult to add more disks to the pool in the future, although that feature is being worked on currently.) There are raidz1 and raidz2 variants, which allow you to lose up to 1 or up to 2 disks respectively, no matter which disks fail. For 4 equal-size disks, the raidz2 variant will still use 50% of your storage for redundancy but will guarantee you survive two disk failures; the raidz1 variant will reduce your redundancy overhead from 50% to only 25%, but only guarantee you survive one disk failure.