create a mirror out of two zfs storage pools

[Johnny Fartpants]

Has anyone got any idea where you would start with this or better still practical knowledge? Two independent pools joined together to make one for a different form of HA?

 

[Chris Moore]

Has anyone got any idea where you would start with this or better still practical knowledge? Two independent pools joined together to make one for a different form of HA?

The only way that even begins to make sense is if you are using the wrong terminology.
It is not something that would be supported by FreeNAS alone.
You could use something like GlusterFS to span storage across multiple pools, but configuring that is outside the scope of FreeNAS and it might not even be possible using FreeNAS even though it is possible to use ZFS on Linux with GlusterFS.

Why don't you try to give more information about what it is you are trying to accomplish, and someone might be able to tell you the right way to do it.

 

[Johnny Fartpants]

The only way that even begins to make sense is if you are using the wrong terminology.
It is not something that would be supported by FreeNAS alone.
You could use something like GlusterFS to span storage across multiple pools, but configuring that is outside the scope of FreeNAS and it might not even be possible using FreeNAS even though it is possible to use ZFS on Linux with GlusterFS.

Why don't you try to give more information about what it is you are trying to accomplish, and someone might be able to tell you the right way to do it.

Thanks Chris. Im not trying to accomplish anything in particular just something that I've been thinking about for a while. Starting thinking about HA and the typical dual head scenario and then started thinking about the cool things that you could gain if you could join two or more pools together. Yeah Im familiar with GlusterFS. I was thinking perhaps you could create one big zvol on each pool and offer that out to FreeNAS VM on a cluster and then make a mirror out of the two pools perhaps. I know a few of people in the industry that use Hardware RAID devices export them and bind them into ZFS pools so this would be a similar idea but losing the Hardware RAID element.

 

[Nick2253]

I was thinking perhaps you could create one big zvol on each pool and offer that out to FreeNAS VM on a cluster and then make a mirror out of the two pools perhaps.

If you did this, you would NOT want to use FreeNAS/ZFS as your merging layer. You would want something with a relatively simple filesystem, because you can leave all the fancy stuff to ZFS on the physical hardware.

I know a few of people in the industry that use Hardware RAID devices export them and bind them into ZFS pools

There are so many things wrong with doing it this way that I would not trust anything that these "few people" are doing. Segregating your levels of abstraction is one thing, using the completely wrong tools for each level is something entirely different.

 

[Johnny Fartpants]

If you did this, you would NOT want to use FreeNAS/ZFS as your merging layer. You would want something with a relatively simple filesystem, because you can leave all the fancy stuff to ZFS on the physical hardware.



There are so many things wrong with doing it this way that I would not trust anything that these "few people" are doing. Segregating your levels of abstraction is one thing, using the completely wrong tools for each level is something entirely different.

Thanks Nick, this makes sense. Yeah its not something that sits great with me either but the overall idea of joining pools got me interested. Perhaps with the work iX is doing with FreeNAS and Minio this will be irrelevant in the future.

 

[Chris Moore]

I know a few of people in the industry that use Hardware RAID devices export them and bind them into ZFS pools so this would be a similar idea but losing the Hardware RAID element

They should definitely loose the hardware RAID. That defeats the entire purpose of ZFS and I will leave it at that even though I wanted to say more. But that is just wrong.

 

[Johnny Fartpants]

They should definitely loose the hardware RAID. That defeats the entire purpose of ZFS and I will leave it at that even though I wanted to say more. But that is just wrong.

Agreed.

 

[Johnny Fartpants]

This is the sort of thing I was thinking of.
Z-RAID SSF

"But what about a solution that allows all data on the main server and the backup server to be really in sync? This would mean that you do not use async technologies like backup or replication but sync technologies like mirroring or Raid-Z between your storagervers where ZFS protecs against a whole server or pool failure. This is Z-RAID SSF where you build a ZFS Pool not from disks but from independent storageservers each with a local datapool and a manual or automatic Z-POOL and Service Failover in case of problems with the primary server."
https://napp-it.org/doc/downloads/napp-it.pdf

 

[Nick2253]

Again, the issue is that ZFS is the wrong solution for the top-level. ZFS has a ton of incredible features, like CoW and scrubbing, that only make sense when ZFS is working directly with the hardware. Once you abstract the hardware away, ZFS is completely inappropriate. Putting a CoW filesystem in a volume on top of a CoW filesystem is at best a performance hit, and at worst a performance nightmare.

Now, it's possible that Napp-It has created some hybrid filesystem that is ZFS with all those features stripped out. However, once you get to that point, you might as well just use mdadm to create a simple mirror on Linux.

 

[Chris Moore]

This is the sort of thing I was thinking of.
Z-RAID SSF

"But what about a solution that allows all data on the main server and the backup server to be really in sync? This would mean that you do not use async technologies like backup or replication but sync technologies like mirroring or Raid-Z between your storagervers where ZFS protecs against a whole server or pool failure. This is Z-RAID SSF where you build a ZFS Pool not from disks but from independent storageservers each with a local datapool and a manual or automatic Z-POOL and Service Failover in case of problems with the primary server."
https://napp-it.org/doc/downloads/napp-it.pdf

The key feature to understand is that what you are seeking is not part of FreeNAS. A solution may be found elsewhere, but the closest thing you might be able to cobble together using FreeNAS is to have several FreeNAS systems working as iSCSI targets with another FreeNAS system being the initiator with those targets mapped into a pool on the "super host".
This is purely a thought experiment and I don't know if it could be made to work.

You would need to really consider the "RAID" level of the cascaded FreeNAS systems because you could end up with a lot of redundant data.