|
|
Subscribe / Log in / New account

When and why to deprecate filesystems

When and why to deprecate filesystems

Posted Mar 10, 2022 2:31 UTC (Thu) by bartoc (guest, #124262)
In reply to: When and why to deprecate filesystems by Wol
Parent article: When and why to deprecate filesystems

btrfs raid5 is arguably less "eat your data" than lvm raid5, since it does actually checksum and can, in fact, detect when drives are doing bogus stuff. You need to be quite careful with scrubbing regularly (and per drive) and with what you do if you need to rebuild the array. I don't really recommend running raid5/6 in any system using SATA drives, they are too willing to lie to you and try and cover up impending failures.

Tbh ZFS' raid5 also leaves something to be desired since it avoids the write hole by doing things that can have pretty bad performance consequences down the line (esp with how unwilling ZFS is to muck around with stuff on disk, in general). I'm hoping bcachefs' approach pays off (it does raid5 by initially writing the data in raid10 (or at least mirrored), committing that, then later rewriting that mirror to raid5 and atomically updating the filesystem metadata once it's done.


to post comments

When and why to deprecate filesystems

Posted Mar 14, 2022 15:10 UTC (Mon) by hkario (subscriber, #94864) [Link]

btrfs raid5 is still marked as experimental, while yes, HDDs lying about their ability to read valid data is an issue, that can be fixed by placing dm-integrtiy below MD-raid and then RAID-5 works fine

also, HDDs lying about bad sectors is not only a RAID-5 issue, it's just as likely to impact RAID-1 setups


Copyright © 2025, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds