Re: [linux-next:master] [btrfs] 9a8eec6165: fio.write_iops 22.6% regression

Hi Oliver,

Thanks for the report. I'm aware of this regression and the cause is
understood: the flagged commit adds a per-COW xarray store on every
COW'd metadata block, and on this single-task 4k randwrite ftruncate
workload that store is what costs the write_iops.

A fix has been posted to linux-btrfs that replaces the xarray with a
non-allocating fixed inline buffer and recovers the lost throughput:

  "btrfs: replace writeback inhibition xarray with a fixed inline buffer"

It carries the Reported-by and Closes tags pointing at this report.

Thanks,
Leo