So to preface. I'm not going to do this. I'm not running zfs even (maybe yet , maybe ever).
I was going trough the basics and I have questions my regular hangout couldn't answer in a way that wasn't "that's just dumb" . So you're supposed to size your pools and vdevs according to your loss tolerance and all . That I got.But then why would you use multiple vdevs for anything but more performance? So imagine you have 12 disks . 2 vdevs of 6 raidz2 was explained to me as an ok setup for that amount of disks.
In that config I could lose 2 disks per vdev and not lose any data. So I could lose 4 total disks and still have all my data (again as long as it was 2 per vdev).I pretty sure I'm right up to here.
Now for the dumb. Why wouldn't you just make 2 pools with one vdev each and then just symlink pool 2 to pool 1 to make it so even if you lost more than 3 disks per vdev you wouldn't lose all your data (or use mergefs instead of symlinks)? What makes this a bad idea (besides performance , assuming performance is totally secondary to data resilience)? Some data is better than no data right ?
Again I am not implementing this , this is from a place of ignorance and I'm trying to "get" zfs and this just hasn't computed.
I was going trough the basics and I have questions my regular hangout couldn't answer in a way that wasn't "that's just dumb" . So you're supposed to size your pools and vdevs according to your loss tolerance and all . That I got.But then why would you use multiple vdevs for anything but more performance? So imagine you have 12 disks . 2 vdevs of 6 raidz2 was explained to me as an ok setup for that amount of disks.
In that config I could lose 2 disks per vdev and not lose any data. So I could lose 4 total disks and still have all my data (again as long as it was 2 per vdev).I pretty sure I'm right up to here.
Now for the dumb. Why wouldn't you just make 2 pools with one vdev each and then just symlink pool 2 to pool 1 to make it so even if you lost more than 3 disks per vdev you wouldn't lose all your data (or use mergefs instead of symlinks)? What makes this a bad idea (besides performance , assuming performance is totally secondary to data resilience)? Some data is better than no data right ?
Again I am not implementing this , this is from a place of ignorance and I'm trying to "get" zfs and this just hasn't computed.