Merge happens either due to movement of objects due to CRUSH recalculation (when cluster grows or shrinks due to various reasons) or deletion of objects.
Split happens when portions of objects/volumes get populated that were previously sparse. Each RADOS object by default is 4MB chunk and volumes comprise of these objects There is no RADOS object created when there is no write on that region. When write spans sparse portions of the volume, new RADOS objects are created under directory that maps the PG to which the object belongs. Regards, Anand On Mon, Jul 11, 2016 at 5:38 PM, Nick Fisk <[email protected]> wrote: > I believe splitting will happen on writes, merging I think only happens on > deletions. > > > > *From:* Paul Renner [mailto:[email protected]] > *Sent:* 10 July 2016 19:40 > *To:* [email protected] > *Cc:* [email protected] > *Subject:* Re: [ceph-users] Filestore merge and split > > > > Thanks... > > Do you know when splitting or merging will happen? Is it enough that a > directory is read, eg. through scrub? If possible I would like to initiate > the process > > Regards > > Paul > > > > On Sun, Jul 10, 2016 at 10:47 AM, Nick Fisk <[email protected]> wrote: > > You need to set the option in the ceph.conf and restart the OSD I think. > But it will only take effect when splitting or merging in the future, it > won't adjust the current folder layout. > > > > -----Original Message----- > > From: ceph-users [mailto:[email protected]] On Behalf > Of Paul Renner > > Sent: 09 July 2016 22:18 > > To: [email protected] > > Subject: [ceph-users] Filestore merge and split > > > > Hello cephers > > we have many (millions, small objects in our RadosGW system and are > getting not very good write performance, 100-200 PUTs /sec. > > > > I have read on the mailinglist that one possible tuning option would be > to increase the max. number of files per directory on OSDs with > > eg. > > > > filestore merge threshold = 40 > > filestore split multiple = 8 > > Now my question is, do we need to rebuild the OSDs to make this > effective? Or is it a runtime setting? > > I'm asking because when setting this with injectargs I get the message > "unchangeable" back. > > Thanks for any insight. > > > > > _______________________________________________ > ceph-users mailing list > [email protected] > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > > -- ---------------------------------------------------------------------------- Never say never.
_______________________________________________ ceph-users mailing list [email protected] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
