Hi Brian, On Tue, Nov 10, 2026 at 11:11 PM Brian Brombacher <[email protected]> wrote:
> I’m curious how many of you have actually experienced a disk failure due > to not having enough spare blocks for SSDs manufactured after 2020. I > write a LOT of data to SSDs and I have yet to experience this issue and I > take no precautions such as saving space at the end of the disk with an > unused partition. Now that I just said that, I’ll have 10 disks fail > tonight, karma… > actually in datacenter we got different outcomes, but MOSTLY with different producers. Must be said that on production servers the write rates for some customers might be quite high on a daily basis; 95%+ of the underlay storage is ZFS based, hence demand on disks reliability is high. Please consider this data kinda "subjective" as we're not keeping detailes statistics, especially of the lifetime of the substituted drives. -old models SSDs (intel, sandisk,etc): all retired but kept in drawer; intels have shown robust, but sample is to small -tried some "solid" and they reavealed to be anything but solid... this choice didn't last a month -also crucial MX series die like flies (we don't buy them anymore after 4-5 years "el cheapo" experience) -samsung evo have a relatively low failure rate 2-4 disks per month over about 20 racks -samsung pro are die hard, changing a pro is an event that might take place once or twice a year -samsung PCIe HHHL PM1735 or PM1725 cards (unfortunately not produced anymore, blame on samsung!) never seen one die!!! :O -samsung PM983, PM9A3, etc. oh, they die too, but somewhat in between evo and pro -samsung NVMe evo/pro: die mostly only when used for long periods for zfs caching, could be maybe the same for AI caching -samsung NVMe server grade: not yet dead, but we got very few and are not being used for caching -high end kioxia: we got some at work, never seen one die but sample is also too small -in general we tend too avoid too "consumerish" brands, they're a cost not a save I hope this answer meets your curiosity!

