On Wed, Jun 18, 2014 at 08:10:03PM +0200, Goffredo Baroncelli wrote: > Hi Fink > > On 06/13/2014 04:41 PM, Werner Fink wrote: > > That is: set NOATIME, NOCOW, and NOCOMP for the journal directory > > If I read correctly, you want set UN-conditionally the NOCOW behavior. > Please, please, please DON'T DO that. > > The NOCOW behavior is not without disadvantage: yes it increase the > performance but > the file also lost the btrfs checksum protection; when BTRFS manage the disks > in RAID mode and a corruption happens, it uses the checksum to select the > correct mirror during the reading. If you set UN-conditionally the NOCOW > behavior you lost this capability even if the user _want it_ (and if they > spend moneys in two or more disks, it is likely they _want it_). > > Moreover the NOCOW flags has some "strange" behavior when a NOCOW file is > snapshotted (it lost the NOCOW property); this may lead to irregular > performance. > > If you want it, it must be configurable at least with a sane default (which > IMHO should be "do nothing", following the "least surprise" rule). > > If you are looking to something like that, I suggest also to defrag the > journal file before the open (but still as configurable option, and > considering the "least surprise" rule).
Yes, this is too much of a hack. Various defragging approaches seem better. Zbyszek _______________________________________________ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
