Bcachefs Merges New On-Disk Format Version For Linux 6.11, Working Toward Defrag

[AndyChow]

I tried bacachefs on root last week with cachyos. Not even a day later got corruption. The moment I opened steam the whole fs would go read only, was working fine minutes before kernel update. Fsck showed nothing. I tried to remove the .steam folder from a live iso and immediately the fs goes RO. This was 100% reproducible. I renamed . steam and tested all was well, but the moment I tried even an ls on that folder got RO.

I had a snapshot from the previous day and rsyncd that over, got errors in .lost and found which also put fs to RO. This reminds me of the btrfs issues I used to get about a year back. I think I took some screeshots. They often say "file has vanished...and the path" and error looking up inum.

Probably too late for a bug report, but I was pushed for time. Any ideas on how to approach it in future to assist in getting bcachefs more reliable?

Rsyncd to xfs + zfs and all is well but I want to try bcachefs when I have a gap again.

I'm almost 100% sure this is the "famous" 32-bit problem. AFAIK right now bcachefs can only work in 64 bits, not 32 bit libraries like steam.

[szymon_g]

really now? thats what i hated the most. (not the result but the process)

He was young back then

[dragorth]

CCleaner - better do it manually if you know where to look.
Registry cleaners - all of them are snake oil and some actually manage to break Windows.
And I still use Vopt - probably the fastest defragmenter ever for Windows. I still have spinning rust where it's applicable.

A really long time ago, Registry Cleaners, mainly the compaction part, had uses to speed up Windows. It has been decades, however.

[voodoo20]

Probably too late for a bug report, but I was pushed for time. Any ideas on how to approach it in future to assist in getting bcachefs more reliable?

Which linux kernel you are used? I feel what 6.10+ got lot of fixes, and on any older kernel it did not worth to try at all. In my case i did not see any troubles related to bcachefs on 6.10 kernel, but on 6.9 i seen strange errors. Not clearly sure about them - also had some hardware issues.

[lyamc]

Did not manage to land root on it, and in fact at time of install bcachefs did not work from Arch install medium.

CachyOS has support for Bcachefs in the installer

[EvilPiePirate]

I wonder if bcachefs defragmentation will have the same caveat that it does on btrfs: that it effectively unwinds Copy-on-Write and dereferences/duplicates data? On btrfs, at least, it's one of those gotchas that makes it completely unusable for a lot of use cases (e.g. when you have lots of snapshots, suddenly each snapshot will occupy lots of disk space instead of merely being mostly references)

No, it will not

[EvilPiePirate]

I tried bacachefs on root last week with cachyos. Not even a day later got corruption. The moment I opened steam the whole fs would go read only, was working fine minutes before kernel update. Fsck showed nothing. I tried to remove the .steam folder from a live iso and immediately the fs goes RO. This was 100% reproducible. I renamed . steam and tested all was well, but the moment I tried even an ls on that folder got RO.

I had a snapshot from the previous day and rsyncd that over, got errors in .lost and found which also put fs to RO. This reminds me of the btrfs issues I used to get about a year back. I think I took some screeshots. They often say "file has vanished...and the path" and error looking up inum.

Probably too late for a bug report, but I was pushed for time. Any ideas on how to approach it in future to assist in getting bcachefs more reliable?

Rsyncd to xfs + zfs and all is well but I want to try bcachefs when I have a gap again.

Join the IRC channel, irc.oftc.net#bcache

We've got a bunch of debugging tools we can walk you through. Was it by chance the "dirent to nonexistent inode" bug?

[jokeyrhyme]

I wonder if bcachefs defragmentation will have the same caveat that it does on btrfs: that it effectively unwinds Copy-on-Write and dereferences/duplicates data? On btrfs, at least, it's one of those gotchas that makes it completely unusable for a lot of use cases (e.g. when you have lots of snapshots, suddenly each snapshot will occupy lots of disk space instead of merely being mostly references)

[dfyt]

I tried bacachefs on root last week with cachyos. Not even a day later got corruption. The moment I opened steam the whole fs would go read only, was working fine minutes before kernel update. Fsck showed nothing. I tried to remove the .steam folder from a live iso and immediately the fs goes RO. This was 100% reproducible. I renamed . steam and tested all was well, but the moment I tried even an ls on that folder got RO.

I had a snapshot from the previous day and rsyncd that over, got errors in .lost and found which also put fs to RO. This reminds me of the btrfs issues I used to get about a year back. I think I took some screeshots. They often say "file has vanished...and the path" and error looking up inum.

Probably too late for a bug report, but I was pushed for time. Any ideas on how to approach it in future to assist in getting bcachefs more reliable?

Rsyncd to xfs + zfs and all is well but I want to try bcachefs when I have a gap again.

[voodoo20]

Great to hear about bcachefs improvements. I already using it with 2 SATA SSDs and 4 HDD. Did not manage to land root on it, and in fact at time of install bcachefs did not work from Arch install medium. So i added 2 smalls SSD (they spare anyway) and install just on ext4 with manual system backup, and on fresh kernel it (bcachefs) going like charm. Nothing serious, just home file server with some add services, with lot of files migrated from NTFS. On main PC still use btrfs, but eventually want try bcachefs.