From: Qu WenRuo <wqu@suse.com>
To: "linux-btrfs@vger.kernel.org" <linux-btrfs@vger.kernel.org>,
David Sterba <dsterba@suse.cz>
Subject: Re: [PATCH v6 0/4] Introduce per-profile available space array to avoid over-confident can_overcommit()
Date: Wed, 29 Jan 2020 05:19:37 +0000 [thread overview]
Message-ID: <55af2fa3-0762-3893-90b4-e0b90a225d27@suse.com> (raw)
In-Reply-To: <20200116060404.95200-1-wqu@suse.com>
[-- Attachment #1.1: Type: text/plain, Size: 5882 bytes --]
Hi David,
Mind to merge the patchset for misc-next?
As patches 1~4 are reviewed, and the last patch is not that complex to
grasp, I guess it's time to finish the long existing df 0 space bug,
with better RAID5/6 estimation for df.
For the long discussion about whether we should set available space back
to 0 when metadata is exhausting, I still tend not to do so, since
metadata and data are separate resources.
Thanks,
Qu
On 2020/1/16 下午2:03, Qu Wenruo wrote:
> There are several bug reports of ENOSPC error in
> btrfs_run_delalloc_range().
>
> With some extra info from one reporter, it turns out that
> can_overcommit() is using a wrong way to calculate allocatable metadata
> space.
>
> The most typical case would look like:
> devid 1 unallocated: 1G
> devid 2 unallocated: 10G
> metadata profile: RAID1
>
> In above case, we can at most allocate 1G chunk for metadata, due to
> unbalanced disk free space.
> But current can_overcommit() uses factor based calculation, which never
> consider the disk free space balance.
>
>
> To address this problem, here comes the per-profile available space
> array, which gets updated every time a chunk get allocated/removed or a
> device get grown or shrunk.
>
> This provides a quick way for hotter place like can_overcommit() to grab
> an estimation on how many bytes it can over-commit.
>
> The per-profile available space calculation tries to keep the behavior
> of chunk allocator, thus it can handle uneven disks pretty well.
>
> And statfs() can also grab that pre-calculated value for instance usage.
> For metadata over-commit, statfs() falls back to factor based educated
> guess method.
> Since over-commit can only happen when we have unallocated space, the
> problem caused by over-commit should only be a first world problem.
>
> Since this patch introduced a new failure pattern, some new error
> handling are introduced:
> - __btrfs_alloc_chunk()
> At the end of that function where calc_per_profile_avail() get called,
> if it failed due to -ENOMEM, we will revert device used space, and
> remove the allocated chunk.
> This is the only new error handling added by patch 5.
>
> - btrfs_remove_chunk()
> There is no good way to revert the change. So here we abort
> transaction, just like what the old error handling does.
>
> - btrfs_grow_device()
> This function has its problem by not reverting device used space from
> the very beginning.
> This patchset will enhance it in patch 4.
>
> - btrfs_shrink_device()
> This function already has good error handling, reuse it.
>
> - btrfs_verify_dev_extents()
> Mount time error will lead to mount failure, nothing to worry about.
>
> Contents of the patchset:
> Patch 1: Core facility, with basic (not perfect) error handling
> Patch 2: Fix for over-confident can_overcommit()
> Patch 3: Make statfs() more accurate
> Patch 4: Better error handling for btrfs_grow_device()
> Patch 5: Better error handling for __btrfs_alloc_chunk()
>
> If needed, patch 4 and patch 5 can be merged into patch 1.
>
> Changelog:
> v1:
> - Fix a bug where we forgot to update per-profile array after allocating
> a chunk.
> To avoid ABBA deadlock, this introduce a small windows at the end
> __btrfs_alloc_chunk(), it's not elegant but should be good enough
> before we rework chunk and device list mutex.
>
> - Make statfs() to use virtual chunk allocator to do better estimation
> Now statfs() can report not only more accurate result, but can also
> handle RAID5/6 better.
>
> v2:
> - Fix a deadlock caused by acquiring device_list_mutex under
> __btrfs_alloc_chunk()
> There is no need to acquire device_list_mutex when holding
> chunk_mutex.
> Fix it and remove the lockdep assert.
>
> v3:
> - Use proper chunk_mutex instead of device_list_mutex
> Since they are protecting two different things, and we only care about
> alloc_list, we should only use chunk_mutex.
> With improved lock situation, it's easier to fold
> calc_per_profile_available() calls into the first patch.
>
> - Add performance benchmark for statfs() modification
> As Facebook seems to run into some problems with statfs() calls, add
> some basic ftrace results.
>
> v4:
> - Keep the lock-free design for statfs()
> As extra sleeping in statfs() may not be a good idea, keep the old
> lock-free design, and use factor based calculation as fall back.
>
> v5:
> - Enhance btrfs_update_device() error handling in btrfs_grow_device()
> - Ensure all failure caused by calc_per_profile_available() is the same
> with existing error handling
> - Fix a bug where chunk_mutex is not released in btrfs_shrink_device()
>
> v6:
> - Don't update the array if we hit any error.
> To avoid calling calc_per_profile_avail() in error handling path.
>
> - Re-order the patchset
> Make the core facility the first patch.
> Error handling improvement in later patches.
>
> - Add better error handling
> Improve one existing bad error handling, and provide a better solution
> for __btrfs_alloc_chunk()
>
> Qu Wenruo (5):
> btrfs: Introduce per-profile available space facility
> btrfs: space-info: Use per-profile available space in can_overcommit()
> btrfs: statfs: Use pre-calculated per-profile available space
> btrfs: Reset device size when btrfs_update_device() failed in
> btrfs_grow_device()
> btrfs: volumes: Revert device used bytes when calc_per_profile_avail()
> failed
>
> fs/btrfs/space-info.c | 15 ++-
> fs/btrfs/super.c | 182 +++++++++----------------------
> fs/btrfs/volumes.c | 245 ++++++++++++++++++++++++++++++++++++++----
> fs/btrfs/volumes.h | 11 ++
> 4 files changed, 290 insertions(+), 163 deletions(-)
>
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]
next prev parent reply other threads:[~2020-01-29 5:22 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-01-16 6:03 [PATCH v6 0/4] Introduce per-profile available space array to avoid over-confident can_overcommit() Qu Wenruo
2020-01-16 6:04 ` [PATCH v6 1/5] btrfs: Introduce per-profile available space facility Qu Wenruo
2020-01-16 16:14 ` Josef Bacik
2020-01-17 0:55 ` Qu Wenruo
2020-01-17 1:50 ` Josef Bacik
2020-01-17 1:54 ` Qu Wenruo
2020-01-17 2:00 ` Josef Bacik
2020-01-16 6:04 ` [PATCH v6 2/5] btrfs: space-info: Use per-profile available space in can_overcommit() Qu Wenruo
2020-01-16 6:04 ` [PATCH v6 3/5] btrfs: statfs: Use pre-calculated per-profile available space Qu Wenruo
2020-01-16 16:21 ` Josef Bacik
2020-01-16 6:04 ` [PATCH v6 4/5] btrfs: Reset device size when btrfs_update_device() failed in btrfs_grow_device() Qu Wenruo
2020-01-16 6:04 ` [PATCH v6 5/5] btrfs: volumes: Revert device used bytes when calc_per_profile_avail() failed Qu Wenruo
2020-01-29 5:19 ` Qu WenRuo [this message]
2020-01-29 9:23 ` [PATCH v6 0/4] Introduce per-profile available space array to avoid over-confident can_overcommit() Nikolay Borisov
2020-01-29 10:51 ` Qu Wenruo
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=55af2fa3-0762-3893-90b4-e0b90a225d27@suse.com \
--to=wqu@suse.com \
--cc=dsterba@suse.cz \
--cc=linux-btrfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).