I would love to hear about your real experiences with these compact NAS solutions. Any unexpected issues or limitations you discovered after you have started using one of them?
Secondly don't skimp on the disk. Ssd or hdd buy the best you can. I ran with shucked WD sata portables and the failure rate over two years was high. I now run with patriot SSD but there's a sense I underspecced.
Hard Disks and ssd run hot. A lot hotter than you might think. Cooling is noisy. So if your dream is passive no fan be warned, things doing storage just run hotter than you might think even if no rotating disk. Hot hard disks will be happier if it's thermally stable. Shorter life, but better than if the temps cycle. Hot ssd seem just to be hot.
Pick a unit which can run truenas, and start on scale not core because core is dying out. If you want a BSD nas look at sylve. Scale does docker and VMs. I run core, I wanted BSD. I am now on an Intel architecture, the pi wasn't strong enough to do nas and virtuals.
Even if you don't want truenas pick one which can because it means it's fully generic. Packaged nas solutions from the hw vendor can lock you in.
I truly believe zfs beats the alternatives. Snapshots, good redundancy, monitoring, good tooling. That's why I went BSD and truenas. The Linux zfs story has got better.
It's a myth you can't run zfs without ecc but it's better with ecc.
It's a myth you can't run zfs with less than 4gb but if you want to run virtuals and avoid stalls under write you want more than 4gb.
It's a myth you can run dedupe on low end hardware. It's a lot of work for less benefit. The default compression in zfs is good. I've never bothered with zlog disks at home, but for write intensive work they help. (We do at work, mirror ssd for log, hdd for the big space)
You still need 3-2-1 backup. Off-line zfs snapshots work for me and some cloud for the third leg.
The post earlier on today about Intel n100 based tiny nas looks interesting. They target ssd, 4 to 6, they look generic, they can do bigger memory models.
Building your own is for fun, not dependencies unless unavoidable. It's probably less reliable unless you really invest time and effort.
Tune zfs for your filesize. There's heaps of stuff online about optimal sizing for postgres and the like.
In the Jeff geerling nas thread, there didn't seem to be very much actual material support for warding / scaring people away from USB attached sata. It's worked fine for me for a decade, worked fine for people with vastly bigger setups than mine.
I largely regard it as decade old FUD myself, & totally decoupled from the state of things. Works more thanfine for a lot of people.
That said, I'd be willing to have some skepticism for some of the Pi hats that do usb-sata bridging, like what you were using. I'd feel much more confident using a more off the shelf dual drive bay product or two. $30 a pop.
What I had was a 4 device SATA hat, 2 USB ports bridged on each half of the divide to form the channel for data, GPIO pins were used solely for card control and power distribution. It was clever, but not sensible.
People who construct RAID sets over USB attached devices on discrete ports also wind up in odd places. The device enumeration can be semi random and so you have some logistical barriers to things (ZFS amongst other sane things, writes the ZVOL/DEV information into the structure on disk so which enumerated device it is becomes less important) as well as a power distribution problem. I ran an early version of this using an unpowered USB hub with 4 USB sticks and it was very badly behaved.
When I do my 3-2-1 the removable external ZFS drive is USB-3 connected.
It just works and comes with many additional packages
Now mini-NAS for backups of backups? That would make sense to me if you do not trust whomever is managing your administrative and backup servers. It might not even be full backups but one could at least back up the data critical to the business that your team is responsible for. A development team lead could automate more frequent backups to their mini-nas than the company has implemented for the wider audience use cases. That would not even need to be in the data-center unless your privacy, compliance and security teams say otherwise. Your home grown solution will still need really good physical security as random contract janitor can just walk out of the building with all of your intellectual property. The NAS just needs a very fast low latency connection to the data-center.
In my experience the revenue impacting data storage and flows should always be on the corporate maintained infrastructure unless one really wants to stand in front of the C-levels explaining why all the data was lost of a home grown implementation. Ideological and technical issues aside the optics will be awful. I've seen people walked out of companies for much less. Augmenting the corporate systems on the other hand can be a life saver especially if you know what data is critical to revenue flows and how many snapshots and full backups would keep your teams workflows uninterrupted. As an augmentation your systems could save the day and your team would look really sharp. As a side note, when your team does save the day by going above and beyond ensure that your management write up your team for awards. That can reduce management friction in the future and buy some leniency for mistakes that will inevitably occur.