You can’t do this - the minimum number of drives (or files) for a RAIDZ1 is 3.
The concept is right - you create a pool using 2 physical drives and a sparse file acting as a 3rd pseudo-drive. Then you delete the sparse file making the pool degraded. If you do a zpool status you will see that the pool is a degraded RAIDZ1 - which means that when you have a 3rd drive available, then you can replace the broken pseudo-drive with a real physical drive and let it resilver, and then you will have a proper RAIDZ1 pool.
I have NOT done this myself, so wait until someone else can verify these instructions, but I think you need to:
Use replication to copy the old-pool to the new-pool (do not set read-only).
Once you are happy that all the data is on the new pool, you can switch over your SMB shares etc. to the new location.
When you are completely happy that your data is in the new-pool and that you have access to it, destroy the old-pool.
Use the now spare disk to Replace the offline pseudo-disk in the new-pool and resilver the redundancy data to it.
Once the resilver is complete you should be done.
I believe that pools are created as auto-expand, so once the resilver has completed the pool should auto-expand to fill the drives, but if not we can help you achieve this through the UI or from a shell.
As I said, this is my guess based on research and experience - I have not done this myself, so get a 2nd opinion.
We wouldn’t generally recommend this; the risk of data loss on a resilver is significant. But if you want to do this, currently, this is the only way to do it:
Alternatively, wait for the release of 24.10, which will include both RAIDZ expansion and the ability to create a two-disk RAIDZ1 vdev.
This is a good point. During a resilver ALL the drives in the vDev are put under heavy workload to do the resilver, increasing the risk of another drive failing during the resilver process, and the longer the resilver takes the greater the risk of failure during the process. The length of time for a resilver depends on the number of drives in the vDev and the size of those drives (or the amount of data in the pool), and even for 3x wide vDevs if you have large drives (and 20TB drives are LARGE) then RAIDZ2 is recommended.
First, for consistency with the way the TrueNAS middleware creates pools. Second, because particularly in Linux, drive identifiers change a fair bit, and if sdc is listed as a pool member, and it becomes sdd, that can confuse ZFS. Partition UUIDs are the Linux flavor of gptids and avoid that problem, because they’re static and unique for a given partition.
And I can’t get disk UUIDs using e.g. lsblk or ls -l /dev/disk/by-x.
So yes, you need to create GPT ZFS partitions on the new drives as @Dan says, and when one of us works out how to do this, they can repost the above instructions amended as appropriate.
I don’t have a test environment handy at the moment, but adapting from Manual disk replacement in TrueNAS SCALE | Dan's Wiki, I believe it would be parted /dev/sda mklabel gpt followed by parted /dev/sda mkpart zfs 1 100%. You can then get the partition UUID by running lsblk -o NAME,SIZE,PARTUUID.
What happens when you have a partition of 20TB and a sparse file which is smaller. Can ZFS create a file system that uses only part of a partition? (I can’t see any technical reason why not. In Linux to grow a file system, you first grow the partition - and the existing file system runs just fine - and then you grow the file system as a separate step.
Actually, this might indeed be a pretty good idea. It does seem best to use the UI to create vDevs exactly as you might wish, and this would be the way to do so.
RAIDZ vdevs are based on the size of the smallest device, so this isn’t a problem. It’d be the same as if you created a vdev with two 6 TB disks and one 4 TB disk–you’d get 8 TB capacity (less overhead etc.), and if you replace the 4 TB disk with a 6 TB one, it expands to 12 TB.
Interesting thought. Would the GUI even let you import such a pool, though? Let’s try. I can create the pool:
pool: testpool
state: ONLINE
config:
NAME STATE READ WRITE CKSUM
testpool ONLINE 0 0 0
raidz1-0 ONLINE 0 0 0
/mnt/software/sparsefile1 ONLINE 0 0 0
/mnt/software/sparsefile2 ONLINE 0 0 0
/mnt/software/sparsefile3 ONLINE 0 0 0
errors: No known data errors
I’ve since deleted the sparsefiles, but I’d expect I could still have imported it from the CLI using the -d flag to tell it to look for the pool devices in the directory where I created the sparsefiles. Of course, the GUI doesn’t have any way to deal with that (nor should it, really).
Edit: yep:
root@nas[/mnt/software]# truncate -s 4T sparsefile1
root@nas[/mnt/software]# truncate -s 4T sparsefile2
root@nas[/mnt/software]# truncate -s 4T sparsefile3
root@nas[/mnt/software]# zpool create -o altroot=/mnt testpool raidz1 /mnt/software/sparsefile1 /mnt/software/sparsefile2 /mnt/software/sparsefile3
root@nas[/mnt/software]# zpool export testpool
root@nas[/mnt/software]# zpool import -d /mnt/software testpool
cannot mount '/testpool': failed to create mountpoint: Read-only file system
Import was successful, but unable to mount some datasets
root@nas[/mnt/software]# zpool export testpool
root@nas[/mnt/software]# zpool import -d /mnt/software -o altroot=/mnt testpool
root@nas[/mnt/software]# zpool status testpool
pool: testpool
state: ONLINE
config:
NAME STATE READ WRITE CKSUM
testpool ONLINE 0 0 0
raidz1-0 ONLINE 0 0 0
/mnt/software/sparsefile1 ONLINE 0 0 0
/mnt/software/sparsefile2 ONLINE 0 0 0
/mnt/software/sparsefile3 ONLINE 0 0 0
errors: No known data errors
