On 2/27/19 4:00 PM, Max Reitz wrote: > On 18.02.19 16:36, Vladimir Sementsov-Ogievskiy wrote: >> 12.02.2019 15:35, Andrey Shinkevich wrote: >>> Clean QCOW2 image from bitmap obsolete directory when a new one >>> is allocated and stored. It slows down the image growth a little bit. >>> The flag QCOW2_DISCARD_ALWAYS allows a call to raw_co_pdiscard() >>> that does the actual cleaning of the image on disk. >>> With the flag QCOW2_DISCARD_OTHER, a reference count of the cluster >>> is updated only. >>> >>> Signed-off-by: Andrey Shinkevich <andrey.shinkev...@virtuozzo.com> >> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsement...@virtuozzo.com> >> >> side question: should not we change discard_passthrough[QCOW2_DISCARD_OTHER] >> to >> true or at least flags&BDRV_O_UNMAP by default? What is the reason of not >> discarding >> things in qcow2-cluster? > As far as I remember the reason is that whenever you clean up something > its cluster is probably going to be reused rather soon. So cleaning up > takes longer, repopulating that cluster takes longer, and you save only > rather little space. > > This is also why I don't know whether this patch makes much sense. > > Max > This depands upon the amount of actually free clusters in the image. If there a lot of them at the moment, this one could be refilled any time later on.
Den
