Bcachefs Merges Support For Btrfs-Like Snapshots

[S.Pam]

The buggy storage stack problem described in birdie's link happened a lot of times already in my current desktop. Very similar. Probably triggered by write commands of Firefox closing not finished yet when the computer is shutdown. Risk lowered if I remember to wait some time between closing Firefox and shutting down the computer. Rescued everytime by fsck to the ext4 drive.

The computer is running ASUS B450 motherboard + Curcial SATA SSD. Not sure which one is the culprit, maybe both.

Try disabling write cache. 'hdparm -W0'.

[billyswong]

Try disabling write cache. 'hdparm -W0'.

Thanks for this idea! I will give it a try. Any many people that care data integrity more than speed may also want it set as default too.

[kreijack]

raidz works well without a ssd or any kind of tiered storage. it doesnt write raidz data twice

EDIT: it's not even possible to do that. you can only cache sync writes. normal writes never use any kind of caching

Is it the purpose of ZIL/SLOG ?

The point is that the RMW cycle of the parity requires two condition:
1) all the stripe is empty
2) or the RMW cycle is protected by a LOG

ZFS, thanks to the variable stripe size, does 1; however in case of small changes there is a fragmentation problem. Anyway changing in the middle of a stripe requires the rewriting of the full stripe.
So, ZFS is the best incarnation of the RAID5/6/7; however the RMW cycle requires some technical compromises which lead to some performance penalty, which is mitigated by the ZIL/SLOG/L2ARC

[Eumaios]

What this file system really needs--and the developer could provide it in seconds--is strategic capitalization: BCacheFS. I can't be the only one who instinctively scans the name as BCAChefs.

[alcalde]

btrfs: raid5 is a joke

This is an urban legend. Btrfs RAID 5 is just fine. Proof:

https://www.unixsheikh.com/articles/battle-testing-zfs-btrfs-and-mdadm-dm.html

[F.Ultra]

Again, completely different situation and Oracle was just finishing what Sun wanted to do. Remember the case with .net implementing Java?

Completely different situation, Microsoft added proprietary extensions to their implementation of Java creating incompatible binaries while still marketing it as Java. Google did no such thing with Dalvik.

[F.Ultra]

i use integritysetup in a raid6 mdadm setup atm. this can also detect and repair such failes

Painfully slow though if you have somewhat larger drives since it will kick a full reconstruct on every single no clean shutdown.

[flower]

Painfully slow though if you have somewhat larger drives since it will kick a full reconstruct on every single no clean shutdown.

no it doesnt. just use bitmap mode and place the crc on an external ssd drive

[maffblaster]

I don't have all the exact commands ready, but here is the gist of it. Very quick and dirty hacked together guide, you'll need to RTFM to fill in the gaps
...
BTW, I use disk -> bcache -> luks -> xfs -> mergerfs myself, but this guide is for disk -> luks -> bcache -> xfs -> mergerfs, which is how I would do it if I were to start over. Also make sure that the root filesystem where you put /etc/luks.pwd is encrypted!

With your permission I'd like to take the content you wrote up and convert it for use on the Gentoo wiki. Do I have your sign-off? Thank you for the excellent guide!

[Tuxie]

With your permission I'd like to take the content you wrote up and convert it for use on the Gentoo wiki. Do I have your sign-off? Thank you for the excellent guide!

Please do! Regarding the fstab line, I realized now that it uses cache.files=off, which doesn't work with mmap. A truly generic version with mmap support would use cache.files=partial but that gives a significant performance hit when dealing with many small files (iirc). You can however have multiple versions of the pool mounted simultaneously with different mount options for different applications so you can have a separate /pool_mmap mounted with cache.files=partial where you point things that require mmap, such as rtorrent and sqlite3.