Piracy: ꜱᴀɪʟ ᴛʜᴇ ʜɪɢʜ ꜱᴇᴀꜱ

Yes, what you're saying is the idea, and why I went with this setup.

I am running raidz2 on all my arrays, so I can pull any 2 disks from an array and my data is still there.

Currently I have 3 arrays of 8 disks each, organized into a single pool.

You can set similar up with any raid system, but so far Truenas has been rock solid and intuitive to me. My gripes are mostly around the (long) journey to "just Docker" for services. The parts of the UI / system that deals with storage seems to have a high focus on reliability / durability.

Latest version of Truenas supports Docker as "apps" where you can input all config through the UI. I prefer editing the config as yaml, so the only "app" I installed is Dockge. It lets me add Docker compose stacks, so I edit the compose files and run everything through Dockge. Useful as most arrs have example Docker compose files.

For hardware I went with just an off-the-shelf desktop motherboard, and a case with 8 hot swap bays. I also have an HBA expansion card connected via PCI, with two additional 8 bay enclosures on the backplane. You can start with what you need now (just the single case/drive bays), and expand later (raidz expansion makes this easier, since it's now possible to add disks to an existing array).

If I was going to start over, I might consider a proper rack with a disk tray enclosure.

You do want a good amount of RAM for zfs.

For boot, I recommend a mirror at least two of the cheapest SSD you can find each in an enclosure connected via USB. Boot doesn't need to be that fast. Do not use thumb drives unless you're fine with replacing them every few months.

For docker services, I recommend a mirror of two reasonable size SSDs. Jellyfin/Plex in particular benefit from an SSD for loading metadata. And back up the entire services partition (dataset) to your pool regularly. If you don't splurge for a mirror, at least do the backups. (Can you tell who previously had the single SSD running all of his services fail on him?)

For torrents I am considering a cache SSD that will simply exist for incoming, incomplete torrents. They will get moved to the pool upon completion. This reduces fragmentation in the pool, since ZFS cannot defragment. Currently I'm using the services mirror SSDs for that purpose. This is really a long-term concern. I've run my pool for almost 10 years now, and most of the time wrote incomplete torrents directly to the pool. Performance still seems fine.