As another anecdote:

I've been working with systems for a long time, too. I've screwed things up.

I once somehow decided that using an a.out kernel would be a good match for a Slackware diskset that used elf binaries. (It didn't go well.)

In terms of filesystems: I've had issues with FAT, FAT32, HPFS, NTFS, EXT2, ReiserFS, EXT3, UFS, EXT4, and exFAT. Most of those filesystems are very old now, but some of of these issues have trashed parts of systems beyond comprehension and those issues are part of my background in life whether I like it or not.

I've also had issues with ZFS. I've only been using ZFS in any form at all for about 9 years so far, but in that time I've always able to wrest the system back into order even on the seemingly most-unlikely, least-resilient, garbage-tier hardware -- including after experiencing unlikely problems that I introduced myself by dicking around with stuff in unusual ways.

Can you elaborate upon the two particular unrecoverable issues you experienced?

(And yeah, Google is/was/has been poisoned for a long time as it relates to ZFS. There was a very long streak of people proffering bad mojo about ZFS under an air of presumed authority, and this hasn't been helpful to anyone. The sheer perversity of the popular myths that have popularly surrounded ZFS are profoundly bizarre, and do not help with finding actual solutions to real-world problems.

The timeline is corrupt.)

> Over 22 years of maintaining servers, I have had serious issues exclusively with ZFS.

I've been using ZFS since it initially debuted in Solaris 10 6/06 (also: zones and DTrace), before then using it on FreeBSD and Linux, and I've never had issues with it. ¯\_(ツ)_/¯

Not to be deliberately argumentative but still no concrete examples of zfs failures are shown, just hand wavey "I had issues I couldn't google my way out of". I've never heard of a healthy pool not mounting and I've never heard of a pool being unhealthy without a hardware failure of some sort. To the contrary, zfs has perfectly preserved my bytes for over a decade now in the face of shit failing hardware, from memory that throws errors when clocked faster than stock JEDEC speeds to brand new hard drives that just return garbage after reporting successful writes.

From my understand of ZFS:

When it is treated as just a filesystem, then it works about like any other modern filesystem does.

ZFS features like scrubs aren't necessary. Multiple datasets aren't necessary -- using the one created by default is fine. RAIDZ, mirrors, slog, l2arc: None of that is necessary. Snapshots, transparent compression? Nope, those aren't necessary functions for proper use, either.

There's a lot of features that a person may elect to use, but it is no worse than, say, ext4 or FFS2 is when those features are ignored completely.

(It can be tricky to get Linux booting properly with a ZFS root filesystem. But that difficulty is not shared at all with FreeBSD, wherein ZFS is a native built-in.)

Linux takes more skill to manage than Windows or macOS, yet we all know Linux. zfs is _the_ one true filesystem, and the last one you need to know. Besides that, to know zfs is to have a deeper understanding of what a filesystem is and does.

I will admit though, to truly get zfs you need to change how you think about filesystems.