This sounds great, but I would like them to see them fix bugs first, like this 6yr old one, with over 160 comments: https://github.com/openzfs/zfs/issues/9130 Every new feature, esp a complex one like this, creates a potentially huge amount of new bugs. Love ZFS, but would never run it without backups. Haven’t had data lose, but needed to restore from backups to get around constant panics.
Considering disks as chunks and mapping zvol structure over chunks so you can use asymmetric size disks. But what ensures zraid block coverage on more than one spindle? This probably is safe for a large enough set of drives, but what is the threshold?
> But what ensures zraid block coverage on more than one spindle? This probably is safe for a large enough set of drives, but what is the threshold?
Surely they'll use a similar allocation strategy to what ZFS currently uses to allocate disk sectors. And of course if you mix a 4TB and a 8TB drive in a AnyRaid-Mirror configuration, you can only ever have 4TB redundant, so like current ZFS I assume you'll see 4TB. But if you then add a 12TB drive later, you can get 12TB mirrored.
> Is it going to be accepted upstream?
Good question. The S3 storage backend was initially said to be upstreamed, but ended up staying private to maintain a competitive edge.
Haven't checked out HexOS before, but given they're partnering with TrueNAS[1] and hiring Klara, Inc for development, one can certainly hope so.
If you mean upstreamed to OpenZFS, then yes. The blog links to a openZFS leadership meeting so I would assume they are all in agreement that this will be in the mainline codebase.
This sounds great, but I would like them to see them fix bugs first, like this 6yr old one, with over 160 comments: https://github.com/openzfs/zfs/issues/9130 Every new feature, esp a complex one like this, creates a potentially huge amount of new bugs. Love ZFS, but would never run it without backups. Haven’t had data lose, but needed to restore from backups to get around constant panics.
Considering disks as chunks and mapping zvol structure over chunks so you can use asymmetric size disks. But what ensures zraid block coverage on more than one spindle? This probably is safe for a large enough set of drives, but what is the threshold?
Is it going to be accepted upstream?
> But what ensures zraid block coverage on more than one spindle? This probably is safe for a large enough set of drives, but what is the threshold?
Surely they'll use a similar allocation strategy to what ZFS currently uses to allocate disk sectors. And of course if you mix a 4TB and a 8TB drive in a AnyRaid-Mirror configuration, you can only ever have 4TB redundant, so like current ZFS I assume you'll see 4TB. But if you then add a 12TB drive later, you can get 12TB mirrored.
> Is it going to be accepted upstream?
Good question. The S3 storage backend was initially said to be upstreamed, but ended up staying private to maintain a competitive edge.
Haven't checked out HexOS before, but given they're partnering with TrueNAS[1] and hiring Klara, Inc for development, one can certainly hope so.
[1]: https://www.truenas.com/blog/powered-by-truenas-hexos/
Say you have a 2, 2x4s and an 8. You should be able to have almost everything mirrored or have some raidz2 and some raidz1.
You could mirror them, then you could use all the space.
Not sure I see how you'd do a raidz2+raidz1 combo using those drives, at least not without losing 4TB of the 8TB drive, but then I'm tired so.
If you mean upstreamed to OpenZFS, then yes. The blog links to a openZFS leadership meeting so I would assume they are all in agreement that this will be in the mainline codebase.
Super! thats what I like to see.
This is massive for home labbers