Message ID | 20240122-reclaim-fix-v1-0-761234a6d005@wdc.com (mailing list archive) |
---|---|
Headers | show |
Series | btrfs: zoned: kick reclaim earlier on fast zoned devices | expand |
On 22.01.24 11:51, Johannes Thumshirn wrote: > We had a report from the field where filling a zoned drive with one file > 60% of the drive's capacity and then overwriting this file results in > ENOSPC. > > If said drive is fast and small enough, the problem can be easily > triggered, as both reclaim of dirty block-groups and deletion of unused > block-groups only happen at transaction commit time. But if the whole test > is faster than we're doing transaction commits we're unnecessarily running > out of usable space on a zoned drive. > > This can easily be reproduced by the following fio snippet: > fio --name=foo --filename=$TEST/foo --size=$60_PERCENT_OF_DRIVE --rw=write\ > --loops=2 > > A fstests testcase for this issue will be sent as well. Please disregard this series. I'm stupid and had lockdep active during testing, so the timing was messed up and reclaim had a chance to kick in.