Skip to contentSkip to conversation

Btrfs progs release 6.19

DISCUSSION David Sterba 4 participants 7 messages Feb 13 – Mar 26 Btrfs file system

lore
Share & tools
lore

David Sterba

author maintainer
#0 lore raw

Hi,

btrfs-progs version 6.19 have been released (ther version 6.18 has been skipped).

There's change in mkfs.btrfs defaults, the block group tree is now enabled. A note is printed (which was a last minute change and the release was redone, the top commit is 9ac9a97ee3ebcc71).

Changelog:

* mkfs:
* make block-group-tree default (support since linux 6.1), use -O ^bgt to unset it for backward compatibility
* speed up initial device discard by procesing the ranges in order
* disable block-grooup-tree feature if a dependent feature is explicitly unselected (like disabling no-holes), instead of erroring out
* check:
* add ability to detect and fix missing orphan items in deleted subvolumes
* add ability to fix inode refs from directory items
* enhance detection on unknown inode keys
* libbtrfsutil:
* minor version update to 1.4.0
* add missing aliases for API updates done in 0.1.3, C and python
* libbtrfs:
* patchlevel version update 0.1.5
* error handling updates
* fixes:
* with DUP profile and mixed sequential and conventional zoned make sure to track the right write pointers
* scrub: fix ETA wraparound calculations, when many files get deleted during the operation bytes_scrubbed and bytes_total get too much out of sync, the ETA will be 0
* corrupt-block: add ability to specify key value when corrupting item keys
* experimental features:
* initial remap tree support (new logical-to-logical mapping layer), coming in linux 7.0
* other:
* documentation updates
* CI updates, new and updated tests
* code cleanups and refactoring

Tarballs: https://www.kernel.org/pub/linux/kernel/people/kdave/btrfs-progs/
Git: git://git.kernel.org/pub/scm/linux/kernel/git/kdave/btrfs-progs.git
Release: https://github.com/kdave/btrfs-progs/releases/tag/v6.19
Python: https://pypi.org/project/btrfsutil/

#1 lore raw
Quoted 3 lines · David Sterba

  * make block-group-tree default (support since linux 6.1), use -O ^bgt to
    unset it for backward compatibility

I guess that also mean that this is now considered stable and has been very well tested?

How about migrating an existing fs to use the block-group-tree (with btrfstune --convert-to-block-group-tree)... has that also been well tested and is considered safe to be used on filesystems with precious data?

Thanks,
Chris.

#2 lore raw
Quoted 7 lines · Christoph Anton Mitterer
Quoted 3 lines · David Sterba

  * make block-group-tree default (support since linux 6.1), use -O ^bgt to
    unset it for backward compatibility

I guess that also mean that this is now considered stable and has been very well tested?

Yes, the bgt feature has been part of the fstests runs.

Quoted 4 lines

How about migrating an existing fs to use the block-group-tree (with btrfstune --convert-to-block-group-tree)... has that also been well tested and is considered safe to be used on filesystems with precious data?

The in-place conversions are a bit more tricky as they start from an existing filesystem and there's a possibility of finding some corner case. Last time this was in 6.15.

It should be doable to test the conversion using device mapper snapshot devices on an existing valuable filesystem without destroying it.

#3 lore raw
Quoted 10 lines · David Sterba
Quoted 6 lines

How about migrating an existing fs to use the block-group-tree (with
btrfstune --convert-to-block-group-tree)... has that also been well tested and is considered safe to be used on filesystems with precious
data?

The in-place conversions are a bit more tricky as they start from an existing filesystem and there's a possibility of finding some corner case. Last time this was in 6.15.

I see. Thanks for the information.

Quoted 3 lines

It should be doable to test the conversion using device mapper snapshot
devices on an existing valuable filesystem without destroying it.

But that would only tell me if the bug actually happens (and is noticed) during conversion... not if something is simply stored in a wrong way which I may not notice until much later (when it might already be too late).

Thanks anyway :-)
Chris.

#4 lore raw

For what it’s worth, (not much) I used --convert-to-block-group-tree on a large fs with precious data (~10 drives, ~100Tb, Raid1 data, Raid1C3 metadata) and had no issues.

Convert took a LONG time, but afterwards, all my data integrity checks passed, scrub passed, and mount times are approximately 1 second now, compared to minutes previously.

—Joshua Villwock

Quoted 29 lines · Christoph Anton Mitterer
Quoted 10 lines · David Sterba
Quoted 6 lines

How about migrating an existing fs to use the block-group-tree (with
btrfstune --convert-to-block-group-tree)... has that also been well tested and is considered safe to be used on filesystems with precious
data?

The in-place conversions are a bit more tricky as they start from an existing filesystem and there's a possibility of finding some corner case. Last time this was in 6.15.

I see. Thanks for the information.

Quoted 3 lines

It should be doable to test the conversion using device mapper snapshot
devices on an existing valuable filesystem without destroying it.

But that would only tell me if the bug actually happens (and is noticed) during conversion... not if something is simply stored in a wrong way which I may not notice until much later (when it might already be too late).

Thanks anyway :-)
Chris.

#5 lore raw
Quoted 7 lines · joshua

For what it’s worth, (not much) I used --convert-to-block-group-tree on a large fs with precious data (~10 drives, ~100Tb, Raid1 data, Raid1C3 metadata) and had no issues.

Convert took a LONG time, but afterwards, all my data integrity checks passed, scrub passed, and mount times are approximately 1 second now, compared to minutes previously.

Thanks for the information. How long have you been running with it?

Cheers,
Chris.

#6 lore raw
Quoted 10 lines · Christoph Anton Mitterer
Quoted 7 lines · joshua

For what it’s worth, (not much) I used --convert-to-block-group-tree on a large fs with precious data (~10 drives, ~100Tb, Raid1 data, Raid1C3 metadata) and had no issues.

Convert took a LONG time, but afterwards, all my data integrity checks passed, scrub passed, and mount times are approximately 1 second now, compared to minutes previously.

Thanks for the information. How long have you been running with it?

Don't know exactly, but it was shortly after the new Debian stable release. So probably around mid-to-late August.

I'd say a good 4-8 months at least since then, with no issues so far.

Signature · 1 line
--Joshua Villwock