https://evilpiepirate.org/git/bcachefs-tools.git/tree/Changelog.mdwn
All kinds of stuff; we're looking pretty good on stability and now it's mostly people just people turning up all kinds of interesting corner cases.
Lots of FUSE patches too - we now have fuse + xfstests running through the CI. The goal there is if there's another Arch hiccup or anything like that, fuse should be a solid and stable (just slow) fallback.
here's the changes: https://evilpiepirate.org/git/bcachefs-tools.git/tree/Changelog.mdwn
dear god this is a big one
last night, on an epyc 9454, md raid0 of 2x enterprise pcie gen4 ssds, dbench 48 clients:
xfs - 16269.8 mb/sec bcachefs - 19029.9 mb/sec
compare to https://www.phoronix.com/review/linux-70-filesystems/3
for reference - phoronix's setup has a CPU with slower cores, but the solidigm SSD they're using is nuts, it outperforms the two enterprise SSDs combined in the hetzner box I'm currently testing on
Starting to focus more on performance and crank through performance issues - this release fixes all the ones I've seen concrete data on
cycle detector/SIX locks improvements - these fix most of the "srcu lock held too long" warnings we've been seeing, and make performance under heavy contention much more robust
btree write buffer multithreading: after the journal pipelining improvements that landed recently, this was the next big bottleneck on the really big systems, and it's fixed thoroughly now
btree node merge attempts thrashing the btree cache are fixed, and 3:2 btree node merging and other improvements are coming in .3
Full changelog - https://evilpiepirate.org/git/bcachefs-tools.git/tree/Changelog.mdwn
So now, if you've got a workload where we're slower than btrfs/zfs, let me know :)