Skip to content

btrfs: commit the transaction unconditionally for ensopc

Josef Bacik linux-btrfs 2 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v2 Latest4 patches 1 replyIn review
  2. v13 patches 4 repliesSuperseded

Changes in v2 from the cover letter

- added "btrfs: remove FLUSH_DELAYED_REFS from data enospc flushing" to deal with me changing the docs and to reflect that we no longer need this step in data enospc flushing.
- Updated the 'rip out' patch to no longer include that particular part of the documentation update.

Patches in v2 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4btrfs: rip out may_commit_transactionfs/btrfs/ctree.h fs/btrfs/space-info.c include/trace/events/btrfs.h+12 -1330 replies
2/4btrfs: remove FLUSH_DELAYED_REFS from data enospc flushingfs/btrfs/space-info.c+0 -110 replies
3/4btrfs: rip the first_ticket_bytes logic from fail_all_ticketsfs/btrfs/space-info.c+0 -160 replies
4/4btrfs: rip out ->total_bytes_pinnedfs/btrfs/block-group.c fs/btrfs/delayed-ref.c fs/btrfs/disk-io.c +4 more+0 -970 replies

Reviews trailers per patch

2 of 4 patches have a Reviewed-by or Acked-by.

PatchNikolay Borisov
1/4
2/4
3/4R-b
4/4R-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v2 thread

Cover letter [PATCH 0/4][v2] btrfs: commit the transaction unconditionally for ensopc

v1->v2:
- added "btrfs: remove FLUSH_DELAYED_REFS from data enospc flushing" to deal with me changing the docs and to reflect that we no longer need this step in data enospc flushing.
- Updated the 'rip out' patch to no longer include that particular part of the documentation update.

--- Original email ---
Hello,

While debugging early ENOSPC issues in the Facebook fleet I hit a case where we weren't committing the transaction because of some patch that I hadn't backported to our kernel.

This made me think really hard about why we have may_commit_transaction, and realized that it doesn't make sense in it's current form anymore. By-in-large it just exists to have bugs in it and cause us pain. It served a purpose in the pre-ticketing days, but now just exists to be a giant pain in the ass.

So rip it out. Just commit the transaction. This also allows us to drop the logic for ->total_bytes_pinned, which Nikolay noticed a problem with earlier this week again. Thanks,

Josef Bacik (4):
btrfs: rip out may_commit_transaction
btrfs: remove FLUSH_DELAYED_REFS from data enospc flushing
btrfs: rip the first_ticket_bytes logic from fail_all_tickets
btrfs: rip out ->total_bytes_pinned

 fs/btrfs/block-group.c       |   3 -
 fs/btrfs/ctree.h             |   1 -
 fs/btrfs/delayed-ref.c       |  26 ------
 fs/btrfs/disk-io.c           |   3 -
 fs/btrfs/extent-tree.c       |  15 ---
 fs/btrfs/space-info.c        | 175 +++--------------------------------
 fs/btrfs/space-info.h        |  30 ------
 fs/btrfs/sysfs.c             |  13 ---
 include/trace/events/btrfs.h |   3 +-
 9 files changed, 12 insertions(+), 257 deletions(-)
Signature · 1 line
2.26.3