Are nested pools possible?

[Pitfrr]

Couldn't find any similar thread so here it is (not sure I have the good keywords though).

Context: On my backup system, I have a pool composed of 8 x 2TB drives in RAIDz2 (using FreeNAS 9.10.x).
3 drives present issues and are failing so that leaves me with 5 x 2TB.
I use this system for backup and with 5 drives remaining I might get away with a RAIDz1 pool (I need about 7TB space for the backup).
Note: new drives are available for the backup system I just need to have a temporary backup while they are burnt in.

Question:
I also have some 1TB drive lying around and I was thinking........
Would it be possible to do something like using 2 x 1TB drive instead of one of the 2TB drive?
Having 2x 1TB striped used as one of the 2TB drive in the RAIDz1 pool? I'd call that nested pools...
Note: I don't want to extend the initial pool, I want to replace one of the drives with several disks.

Initial pool configuration:

"Nested" pool configuration:

Then the pool would still have one drive parity even if the stripe fails (if one of the 1TB drive fails the complete stripe would be lost but not the complete pool).

Not sure if this is possible though... :) At least probably not through the web interface.
And I'm not saying I would do that on a production system... :)
It's more of a theoretical question (and maybe for testing purposes or in my case temporary backup).

 

[sretalla]

if one of the 1TB drive fails the complete stripe would be lost but not the complete pool

If one of the members of the stripe is lost, the entire stripe is lost, so that theory doesn't work.

If you use the CLI, ZFS will take whatever partition you present it by GPTID (assuming the size is sufficient), so you could do all kinds of trickery to make more than one disk look like one disk.

In any case it is absolutely not possible to do that in any supported way... feel free to play with options using a RAID controller or other options to present 2 disks as a single disk to the OS to make what you want, just be prepared to lose all your data and have no help from this forum to recover.

 

[Pitfrr]

Thanks for the feedback!

In any case it is absolutely not possible to do that in any supported way...

I though so... :-O

I don't have experience configuring ZFS from the CLI...
What commands should I be looking at? Would you have any pointers?

I might experience with it a bit, just for fun and to gain a bit of experience (later when I'm finished with my backups! ). Therefore I wouldn't mind if I lose the data.

 

[sretalla]

I'm not going to spend a lot of time writing out instructions for you to do something that's dangerous and should never be used in any context for storing data with ZFS (and FreeNAS).

You can look into gpart (gpart list shows the rawuuid of partitions) to make partitions on whatever disk you can create (perhaps with GEOM: https://www.freebsd.org/doc/handbook/geom-striping.html) then using zpool create with the rawuuids to put the pool together.

As a final note, I think your exercise is folly and something nobody should be interested in doing, but I can appreciate the nature of experimental thought, so I haven't just ignored the post.

DO NOT DO THIS ON ANY SYSTEM HOLDING REAL DATA THAT YOU CARE ABOUT!

 

[Pitfrr]

I'm not planning on using it actively on FreeNAS to store data.

It's only out of personal interest to get more in-depth understanding of ZFS, toying, messing around, experimenting, this helps quite a lot in understanding. And it's nice to have some pointer to know where to start to look...

I'm trying slowly to build up knowledge on FreeNAS, that is the idea behind.
I'll check out the link. Thanks.

 

[sretalla]

I want to put a copyright on the term Frankenpool and ask that you use this name rather than nested pool... It's a monster destined for disaster from the moment it was made.

 

[Patrick M. Hausen]

You could use gstripe to create a striped block device and use that in your pool ...

 

[sretalla]

Just as additional thought...

In creating the Frankenpool, the striped member actually doubles the chance of failure (assuming both 1TB disks are equally likely to fail and one failure kills that entire member).

See this: https://www.edcollins.com/backgammon/diceprob.htm and look at the difference between one and 2 dice (if you assume "likely to fail" is equated with rolling a 6).

 

[Yorick]

If one of the members of the stripe is lost, the entire stripe is lost, so that theory doesn't work.

As horrid as it is, it kinda does work - it’s akin to a single drive failure, which raidz1 can survive. The Frankenstripe would need to be replaced with an actual 2TB drive and that drive would need to be resilvered in.

Yeah. Scary. I’m also not clear how “8x2TB raidz2 and 3 drives are failing” works - presumably not failing so badly they just offline out of the vdev and, holy cow. “I can lose 2 drives and I am losing 3 - it’s fine. Just fine. This is fine.” Lol.
Well you are moving to a 5x2TB config, so that’s a good move. That or replacing the three failing drives.

 

[Yorick]

It's a monster destined for disaster from the moment it was made.

With feverish intensity, Frankenstein stitched two drives together using ‘gstripe’, reveling in his power to make one from two. Just as he was about to graft this limb onto his monster, inspiration struck him. He ripped apart what he had just created, took two more from the bench, and placed all four into a raidz1. Then, he created a zvol on it and, cackling maniacally, grafted THAT into his monster pool. “A hahahahah!!! Behold!!!”

 

[Pitfrr]

There is temptation like Frankenstein to bring your creation to life of course...
Still I would find the experiment interesting and it could probably bring some knowledge.
But then I wouldn't bring it to life (i.e. in production) it would only stay in the lab....