1
0

Update to v1.34.0 #10

Open
holden wants to merge 12 commits from holden/bcachefs-tools:master into master
6 changed files with 131 additions and 20 deletions

View File

@@ -1,16 +0,0 @@
-----BEGIN PGP SIGNATURE-----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=06NA
-----END PGP SIGNATURE-----

Binary file not shown.

View File

@@ -0,0 +1,16 @@
-----BEGIN PGP SIGNATURE-----
iQIzBAABCgAdFiEEKnAFLkS8Qha+jvQrE6szbY3KbnYFAmmGnfIACgkQE6szbY3K
bna1Tg//cY9WR2+BrYrKkPMyFmQsolMjJlP8GWh5ZyFe8qCx28GyjXvqqQJN9U9x
UKtvc5sQ4E86Df18H9RINvhbvBeuqsA7o1kM7tPft6DjMGM7tz9LfbX4jdvbOdCb
NWTkQ4OjoYRQWyusv+5dGaa4CgSepI9TSxZ6G7p7+AqWOAv/cJDh3n+ZtYELrOPw
u84lMNxEcms0YX8qz1sFceRcPJ/Fw3VPVxInUnwpjh51IJLSGIVKN9hekPQWItCt
svDy/bOEW2HdsGMOjeCeBrff/FGUQrk/n0dphJTjRL0uUi1CW6SlQipLnqMJj2JG
+W720y++EWN4n9kiyQqAqa+DdOYTyiYTLcKIDe7qxRHSwuKt4MbAa3WjHxg7WFQU
tL9RM2rodx8cG+wD4rwsEGJ8C1OEnYelgIHVayfAnOTJ42t8HmtjbRGoESeoGxEc
lK90IyS+76LQ9RANGu/dGenElKF/Ehu5GDr/6xjXLCcuNr9oQuProviX7/WLS4rv
gcVVSGfZUSAXWJ77eH+mKFpy1UdvqkFHYd1KHc4MfZI+5KS0b4UFQjzcqrQghg5k
I2Oyq/xQmSJd5YlGfdbLMps6bo3nagGbGeXbUiLBD9vBozGlWKVWzFVkfeRzd4+q
AlohTC4UjQfYTDFO6VTGsD41zNIChbE3KPOkYCPZhIjsDS7G720=
=t2iF
-----END PGP SIGNATURE-----

Binary file not shown.

View File

@@ -1,3 +1,114 @@
-------------------------------------------------------------------
Sat Feb 7 21:14:00 UTC 2026 - Holden Fried <holden@opensuse.org>
- Update to release v1.36.1
* Interactive TUI for monitoring various filesystem internals,
slowpaths and device performance, with duration and frequency
tracking for various events. Helpful for diagnosing performance
issues.
* `encoded_extent_max` default bumped to 256k; new filesystems
now initialize `BCH_SB_EXTENT_BP_SHIFT` to 16, so higher
settings won't require rebuilding backpointers
* `--rotational` flag now works correctly during format
* Copygc now waits until a device is less than 20% free before
starting
* Improved `bcachefs reconcile status` output
* Large batch of erasure coding cleanup and hardening: better
error reporting for EC reconstruct reads, fix a race between EC
and data moves, and various other EC bugfixes
* Fix write buffer `move_keys_from_inc_to_flushing()` regression,
which was causing occasional oopses under load for some users
* Snapshot deletion is now much faster when deleting large
numbers of snapshos; we now use an eytzinger tree for the list
of nodes being deleted
* Fix sporadic superblock checksum failures during device scan
* Many smaller bugfixes
- Update to release v1.36.0
* bcachefs_metadata_version_no_sb_user_data_replicas
* This requires an incompatible upgrade to enable, and once
enabled we'll no longer store replicas entries in the
superblock for user data, which are used for deciding whether
we an do a degraded mount without data loss - instead, we defer
that and use the accounting btree to check, in early recovery.
* This is a performance/scalability fix: on filesystems with
large numbers of drives (a 50 device filesystem was the
original bug report), the superblock writes needed to add and
delete replicas entries become a bottleneck.
* Replicas entries for metadata (btree and journal) can still be
an issue, and another bug report indicated that these will have
to be addressed soon - a single slow (or dying) device in a
large multidevice will cause all superblock writes to slow to a
degree that can cause major problems. Metadata replicas entries
will however require a different approach to solve, so expect
that in a future update.
* Some fairly involved fixes for the data update path: it turns
out, the data update path was dropping replicas to devices
being evacuated (which are considered to have durability of 0)
before the extent was sufficiently replicated on other devices.
This caused data loss for a few users, unfortunately, but the
new code is much more rigorous when reconciling the exsiting
extent with newly written replicas and deciding which replicas
to keep and which can be dropped.
* Fix various codepaths that were (incorrectly) causing the
filesystem to go emergency read-only when finding a pointer to
an invalid device, instead of continuing so it could be
repaired or flagging the filesystem as needing repair. We now
should only go emergency read-only on pointer to invalid device
when that would indicate a runtime bug, not filesystem
corruption.
* Reconcile will now shut down correctly (when the filesystem is
going read-only or unmounting) when processing the
reconcile_*_phys btrees.
* Multiple other smaller reconcile fixes; various users report
that issues where reconcile did not seem to be finding pending
work seem to be resolved.
* Degraded btree nodes are no longer un-degraded synchronously;
now that we have reconcile this is no longer necessary, and
forcing them to be un-degraded synchronously was prone to
causing deadlocks on open_bucket allocation.
* The 'allocator stuck' log message now provides improved
information, and internally has been re-plumbed to have access
to the original 'struct alloc_request', so if necessary for
future debugging we can easily provide as much information
about how the allocation was attempted as required.
-------------------------------------------------------------------
Sun Jan 25 21:16:08 UTC 2026 - Holden Fried <holden@opensuse.org>
- Update to release 1.35.2
* Linux v6.19 is now supported
* Reconcile now considers the amount of durability we have
available among online devices when dropping extra replicas
(because the replicas setting was changed), and won't let the
online durability go below the replicas setting.
* Fix a race in the nocow write path when checking if we need to
fall back to a normal COW write
* Fix a livelock when walking btree roots in reconcile and
elsewhere
* Journal discards are now done asynchronously instead of being
done by the journal reclaim thread, and we try to keep more of
the journal discarded to avoid journal writes having to block
and do discards synchronously
* Fix several bugs with copygc <-> reconcile interaction, and
copygc should no longer spin when a device is completely full
with no fragmented buckets for it to evacuate.
* Fix propagating the incompressible bit in the data update
path: sometimes this would be lost, leading to spurious
"extent with bad/missing reconcile options" errors.
- Update to release 1.35.1
* Self healing for the new stripe refcount field in
`bch_alloc_v4`
* This fixes issues upgrading to 1.35 with (still experimental)
erasure coding feature.
* Major allocator refactoring, simplifying the central control
flow. Prep work for failure domains.
* Erasure coding can now delete stripes from triggers; this
gives better behaviour when data being deleted with no other
activity to cause stripes to be deleted.
* Fix a deadlock in device add when allocating journal on the
new device; this fixes a regression from the watermark cleanup.
-------------------------------------------------------------------
Tue Jan 13 19:52:59 UTC 2026 - Jan Engelhardt <jengelh@inai.de>

View File

@@ -17,7 +17,7 @@
Name: bcachefs-tools
Version: 1.35.0
Version: 1.36.1
Release: 0
Summary: Configuration utilities for bcachefs
License: Apache-2.0 AND (Apache-2.0 OR MIT) AND (Apache-2.0 OR Apache-2.0 WITH LLVM-exception OR MIT) AND MIT AND MPL-2.0 AND (MIT OR Unlicense) AND BSD-3-Clause AND GPL-2.0-only AND GPL-2.0-or-later AND LGPL-2.1-only