On Wed, Jan 23, 2019 at 4:04 PM Michal Soltys <soltys@xxxxxxxx> wrote: > > On 19/01/23 18:14, Song Liu wrote: > > On Wed, Jan 23, 2019 at 8:53 AM Michal Soltys <soltys@xxxxxxxx> wrote: > >> > >> Hi, > >> > >> From kernel documentation: > >> > >> "In write-back mode, MD reports IO completion to upper layer (usually > >> filesystems) right after the data hits cache disk." > >> > >> Does "data hits cache disk" mean the actual flush of the journal device as well ? > >> Or is it staged for later moment when the full-stripe write(s) to the actual raid > >> device are to be performed ? > >> > > MD layer will write the data to the cache disk, but not flush it. Flush happens > > when upper layer (file system or application) issues flush (fsync, etc.). > > > > Does this answer the question? > > Yes, thanks. > > For the record, my question came after checking 2016 presentation ( > https://events.static.linuxfound.org/sites/events/files/slides/2016_vault_write_journal_cache_v2.pdf > ). The slide titled "RAID - 456 Write Cache: Write Path" kind of > suggested that the flush might be happening, which was a bit suspicious > (or I misunderstood that slide). Yeah, that was a little confusing. The actual bio will have PREFLUSH or FUA flag if it contains upper layer bios with PREFLUSH or FUA. You can see the logic in function r5l_do_submit_io(). Thanks, Song > > > > > Thanks, > > Song > > >
