I'd expect that to be permanent.
Sopel97
joined 1 year ago
the 131072 ones would probably go away without actual reallocation on next write, that seems to be an intermittent issue on HGSTs, losing whole tracks randomly (possibly caused by incorrect disconnect, had that happen once. that thing has quite a lot of power-on counts, q-sense error rate is also quite high). Ofc that's 64kB of data lost.
I'd be more concerned about the other 16 honestly
I have a drive that was decomissioned, presumably, due to 2 errors during long smart tests (within 2k hours, at roughly 30k). Other than that it has perfect SMART parameters, surface tests without issue, and has been running for a year 24/7 no problem now. Other than excessive vibration or a power issue I have no idea what could have caused that.
EDIT: was thinking Firecuda as they are SMR rather than the CMR barracuda, so read that SMR drivers are considered to be better.
read again
because they are different products, they have different amounts of them, there is different amount of them on the market at different prices, and the demand is different
this budget might be enough if you live in the US and buy cheapest used drives there are
Disregard the advice about SSDs. Start with 1 NVMe for boot and only add more if you have a need. Either for running VMs or if additional cache is actually needed.
I trust MB SATA more in terms of reliability. HBAs tend to overheat too.
However, if RAID topology allows, I'd try to spread the drives such that either one of MB or HBA failing completely would not bring the array down (RAID10 with 1 HBA, or RAID5/6 with 2 HBAs).
That's expected for 4k synchronous writes. They need to complete one by one and you're writing to a hard drive.
One ZFS drive per pool, with unraid on top? Should be mostly fine. As you say these drives are terrible for writes, but they should be pretty reliable and performant for your workload.