It is possible in SCALE 24.10 (Electric Eel) which is currently in beta. It is not possible in the current production version of Scale. You can either run beta and have the feature now or run production without the feature and update once Electric Eel is out of beta.

This is not true. You can add any sort of vdev to an existing pool. There is no "balancing"; ZFS doesn't relocate existing data, so whatever you've written to the old vdev just stays there. New writes will be distributed in proportion to the speed and occupied ratio of each vdev. Unless you have high turnover on the pool, there will always be a perpetual imbalance regardless of vdev geometries. It would be somewhat foolish to reduce the redundancy level when adding another vdev, but aside from that, nothing stops you from adding a vdev with more or fewer disks than any other vdev.

Your understanding is incorrect, or mine is - dealer's choice, really. "Old" data does keep the same data:parity ratio for its lifetime, but it gets redistributed across all the drives. There's some neat animations in the presentations explaining how this works. No. This is a very special case, and it does clever things to avoid needing to rewrite the world to get away with it. AFAIK no, you can only do one at a time, but as you say, you can do it again once it finishes. (I can't immediately think of a reason you inherently couldn't do more than one at once, except if somehow it violated the redundancy properties you needed, but I can't think of a way that would be true...) It's not even in a release yet, much less aggressively tested widely, and I don't know that it'll make it into the next stable branch, so you may be waiting a while if you want to make use of it. My usual advice is, don't plan around a feature that's not released yet being released on any timetable, no matter how confident you are, because many things can happen.

Raidz expansion was finalized and has been in master for a while now, so if you really want it, you can simply compile ZFS from master yourself. It will make it into 2. I was really excited for raidz expansion a few years ago, but I realized I don’t need it, and it’s always going to be better to simply build a new pool doing a zfs send from a backup pool. This way, you achieve better space utilization and start from an a state with minimal fragmentation. EDIT: The first release candidate for 2.3 was just released, and it includes raidz expansion, DirectIO, and fast dedup. 2.3 should release as a final version in 3 months or so.

This feature will be available in the OpenZFS 2.3 release, which is probably about a year out.

I can't confirm any of the above, but I did bookmark this script for rebalancing ZFS pools: https://github.com/markusressel/zfs-inplace-rebalancing