Bcachefs Sees Improved Journal Pipelining & More Efficient Discard With Linux 6.9

[phoronix]

Phoronix: Bcachefs Sees Improved Journal Pipelining & More Efficient Discard With Linux 6.9

Earlier this week with the original Bcachefs pull request for Linux 6.9 Linus Torvalds wasn't happy with some of the code pertaining to spinning out a new library code so that it could be re-used by at least the XFS file-system. A revised pull request was since submitted without that library spin-out and Torvalds today went ahead and merged that updated file-system driver...

https://www.phoronix.com/news/Bcachefs-Linux-6.9-Pulled

[Mathias]

Note: If you mounted a multi-device with 6.8, don't go back to 6.7!

bcachefs: do not run 6.7: upgrade to 6.8 immediately if you have a multi device fs
there's a bug in 6.7 with filesystems that are mid upgrade and then get
downgraded not getting marked in the superblock as downgraded, and this
translates to a really horrific bug in splitbrain detection when the old
version isn't updating member sequence nmubers and you go back to the
new version - this results in every device being kicked out of the fs.

No need to recreate and repopulate - you just don't want to be going
back to 6.7 from a newer version.

source

[dralley]

I wonder if the formatting tool defaults are also hurting performance on some devices. You can see that in Michael's benchmarks bcachefs is choosing 512 byte blocks on devices that use 4096 byte blocks on other filesystems.

Was brought up on reddit a while back https://www.reddit.com/r/bcachefs/co...time_defaults/

[fitzie]

Note: If you mounted a multi-device with 6.8, don't go back to 6.7!


source

he's happy to try to throw the stable team under the bus, and attack them with all sorts of unreasonable requests and lies, but won't put an notice on bcachefs.org, won't tell people how to patch 6.7.

if there is bcachefs bug in stable, before he starts scream at people and demand they immediately cut a release, he should put a notice on bcachefs.org with a link to the patch/git repo. then he can submit a backport request.

can someone explain how he's this oblivious?

[mdedetrich]

he's happy to try to throw the stable team under the bus, and attack them with all sorts of unreasonable requests and lies, but won't put an notice on bcachefs.org, won't tell people how to patch 6.7.

if there is bcachefs bug in stable, before he starts scream at people and demand they immediately cut a release, he should put a notice on bcachefs.org with a link to the patch/git repo. then he can submit a backport request.

can someone explain how he's this oblivious?

It took btrfs ~8 years to put a warning about RAID5/6 possibly causing loss of data due to write hole in mkfs (and yes it was known for that entire time) and your complaining about this?

[curfew]

Bcachefs. Sometimes won't eat your data.

[Daktyl198]

It took btrfs ~8 years to put a warning about RAID5/6 possibly causing loss of data due to write hole in mkfs (and yes it was known for that entire time) and your complaining about this?

That's a whataboutism. There's a reason distros didn't use btrfs for 8+ years and it's because it wasn't stable either. You're allowed to be mad at both things.

[cend]

Note: If you mounted a multi-device with 6.8, don't go back to 6.7!


source

Is that bug found late or...? Where's the fix commit?

[varikonniemi]

- Improved journal pipelining, significantly improving performance on high iodepth write workloads

nearly 300% performance increase:

This patch reworks the journalling code to allow more than one (non-flush) write to be in flight at a time. With this patch, doing 4k random writes and an iodepth of 128, we are now able to hit 560k iops to a Samsung 970 EVO Plus - previously, we were stuck in the ~200k range.https://evilpiepirate.org/git/bcache...43f727f045a468

[janos666]

Is there any ETA on online defragmentation?