At Wed, 08 May 2013 11:16:19 +0800, Liu Yuan wrote: > > On 05/08/2013 12:06 AM, Hitoshi Mitake wrote: > >> > On 05/06/2013 10:15 PM, Hitoshi Mitake wrote: > >>> > > Ah, I see. But the preallocated objects would consume disk > >>> > > space. Removing them physically would be benefitical for thin > >>> > > provisioning, I think. How do you think? > >> > > >> > This is just the very unusual case and even if happens, only one extra > >> > object is created, no? > > I'm not sure about that. A number of removed objects would not be > > 1. The number would depend on a way of using disks of guest OSes. How > > do you think? > > > >> > By the way, if we don't log the delete operation, > >> > the 'discard command' doesn't return to the guest. So after restarting, > >> > the guest will see the file is still there because inode isn't updated > >> > successfully yet. Then people can re-delete the file, no? > > Do you mean a failure during discard command? If so, I agree with your > > opinion. But failures _after_ discard commands would cause problems > > when the journaling mechanism is enabled. > > Okay, please rebase the patch onto the master.
OK, I'll send v2 later. Thanks, Hitoshi -- sheepdog mailing list [email protected] http://lists.wpkg.org/mailman/listinfo/sheepdog
