On Fri, 29 Jan 2016, Ilya Dryomov wrote: > On Fri, Jan 29, 2016 at 6:43 PM, Sage Weil <sage@xxxxxxxxxxxx> wrote: > > On Fri, 29 Jan 2016, Guang Yang wrote: > >> Hi Sage, > >> I found a thread discussing whether we could turn on disk cache with > >> new kernel or not - > >> http://www.spinics.net/lists/ceph-devel/msg06173.html > >> > >> From within that thread, you mentioned that since the fsync/fdatasync > >> would flush the disk cache as well, we could actually turn on the > >> cache without worrying about the data safety. > > > > Yeah, any vaguely modern kernel will do a proper flush to the device; no > > need to worry about the disk cache. > > > >> Is that true if journal_dio is being used (by default), in which case > >> O_DIRECT | O_DSYNC is being used to write journal rather than > >> fsync(2)? > > > > Right. O_DSYNC triggers a flush at the SCSI (or virtio) level, so these > > journal [A]IOs are fully synchronous all the way to the platter. > > The last time I looked, O_DSYNC was nothing but an implicit fdatasync > after each write, triggered at the VFS layer - not that much different > from the explicit fdatasync. Right, and that (implicit) fdatasync triggers a flush. sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
