I wouldn’t count on that. Bit flips due to cosmic rays can happen even underground. Hence the need for periodic scrubs. I use slender two 2TB USB-C SSDs for image backups during travel. These m.2-based Gumsticks are slender enough to fit in a pocket. Having two should allow enough probability that all images will make it through.
1 Like
Unlucky Is really only theorical 
1 Like
Paper already looks like a safer bet for centuries in a dry underground vault. The British parliament still relies on parchment for archival records. And Babylonian clay tablets have stood the test of time for some millennia. Take THAT, ZFS! 
3 Likes
The closest thing to long-term storage like paper but with a somewhat higher media density is M-Disks. But even they will fall prey to the issue of how to read the data as formats disappear, drive innards turn to dust, etc. At least DVDs are common enough that readers are likely to be around for a while.
1 Like
I have bought MANY 8TB SMR drives from seagate, mostly via shucking of USB drives. They were originally used as single drives, then in RAID5, and eventually RAIDZ2. I am working to eliminate them; at this point they only exist in a RAIDZ2 media backup array (which is itself a redundant backup).
Do they suck: yes.
Do they die a little bit more often than I’d like: also yes.
Do they die a lot more often under ZFS: very yes.
Would I do the same thing again: still yes.
I could not have afforded a set of non-SMR HDDs. Even now, the 16TB CMR drives I’ve switched to are really quite expensive considering I have 5+spare. Prior to going ZFS, they were an effective solution considering my usage pattern (even if their write performance did sometimes fall off a cliff so hard it caused software to break when I was moving data on to them).
But here’s a thought:
What if you had a ZFS array where each record was exactly sized and aligned with a shingle zone on the SMR drive? Then the fact that it’s SMR would be irrelevant because we would never try to write to only part of a shingle zone. I’d be interested in such a system.
1 Like
That is what HA-SMR, (Host Aware SMR), is supposed to allow.
That is so unlikely to happen with OpenZFS in the next few years, as to be a hard no.
I’ve heard that some SMR disks use shingled zone sizes of 256MBs, which would not be too unreasonable. Far larger than the 1MB block size of OpenZFS today. But, if someone was only storing large files that might make sense.
The real kicker, is that at present ZFS is completely un-suited for this role. ZFS has no block allocation scheme that would support such, even if the shingled zones were nicely laid out in 256MB sizes, from the beginning of the disk. Unless a pool had a Special vDev for metadata & small files, several of those 256MB zones would have to be used for that purpose.
The only way I see HA-SMR support being added to ZFS, is by creating a new vDev type for HA-SMR HDDs. But, that concept breaks down because of redundancy. Whence redundancy is needed, RAID-Zx or Mirroring, THOSE have to be made HA-SMR HDD compatible.
But, I could be wrong.
Someone might be working on HA-SMR HDD support for OpenZFS in stealth and is simply waiting until the code is working enough to make it public.
I’m interpreting that as a suggestion that I should buy some HA-SMR drives and get on it 
(which isn’t going to happen, my ADHD is far too extreme to support a multi-year development process).