Comment by compsciphd

anecdote: I've had very bad experience with these OS white label drives, even when marked as new. I've had much better luck shucking USB drives.

4+ years ago I bought 20 "new" (can't validate), "seagate manufactured" (can't validate) "OS" SAS drives, and 2 started throwing errors in truenas quickly (sadly after I had the ability to return them). Had another 20 WD and Segate drives I shucked at the same time (was going into 3 12x SAS/SATA machines and 1 4x SATA NAS). The NAS got sidelined as had to use the SATA drives were meant for and no longer trusted the SAS drives so wanted to keep the 2 extra drives as backup. Which was a good idea, as over the next 4 years another 2 of SAS drives started throwing similar errors.

so 20% of the white label drives didn't really last, while 100% of the shucked drives have. What was even worse, the firmware on the "OS" drives was crap, while it "technically" had smart data, it didn't provide any stats, just passed/not passed. (main lesson learned from this, don't accept

Another anecdote: For a long time I wasn't sure what to do with the SAS drives as in the past I used unused drives for this for cold offline storage, but SAS docks were very expensive ($200+). Recently it seems they have come down in price to under $50 so I bought and was able to fill the drives up (albeit very slowly, it seems they did have problems (was only getting 10-20MB/s), but at least I was able to validate their contents a few times after that, a bit less slow (80MB/s).

Aside: 3 weeks ago I had multiple power outages that I thought created problems in one of the shucked drives (was getting uncorrectable reads, though ZFS handled it ok) and a smart long test show pending sectors. But after force writing all the pending sectors with hdparm, none of the sectors were reallocated. I now think it just had bad partial writes when the power outage hit, so the sectors literally had bad data as the error correcting code didn't match up, also explains why they were all in blocks of 8), and multiple smart long tests later and "fingers crossed", everything seems fine.