Running a cluster on various versions of Hammer and Jewel I haven't had any problems. I haven't upgraded to Luminous quite yet, but I'd be surprised if there is that severe of a regression especially since they did so many improvements to Erasure Coding.
On Thu, Oct 19, 2017 at 4:59 PM Jorge Pinilla López <jorp...@unizar.es> wrote: > Well I was trying it some days ago and it didn't work for me. > > maybe because of this: > > http://tracker.ceph.com/issues/18749 > > https://github.com/ceph/ceph/pull/17619 > > I don't know if now it's actually working > > El 19/10/2017 a las 22:55, David Turner escribió: > > In a 3 node cluster with EC k=2 m=1, you can turn off one of the nodes and > the cluster will still operate normally. If you lose a disk during this > state or another server goes offline, then you lose access to your data. > But assuming that you bring up the third node and let it finish > backfilling/recovering before restarting any other nodes, then you're fine. > > On Thu, Oct 19, 2017 at 4:49 PM Jorge Pinilla López <jorp...@unizar.es> > wrote: > >> Imagine we have a 3 OSDs cluster and I make an erasure pool with k=2 m=1. >> >> If I have an OSD fail, we can rebuild the data but (I think) the hole >> cluster won't be able to perform IOS. >> >> Wouldn't be possible to make the cluster work in a degraded mode? >> I think it would be a good idea to make the cluster work on degraded mode >> and promise to re balance/re build whenever a third OSD comes alive. >> On reads, it could serve the data using the live data chunks and >> rebuilding (if necessary) the missing ones(using cpu to calculate the data >> before serving// with 0 RTA) or trying to rebuild the missing parts so it >> actually has the 2 data chunks on the 2 live OSDs (with some RTA and space >> usage) or even doing both things at the same time (with high network and >> cpu and storage cost). >> On writes, it could write the 2 data parts into the live OSDs and >> whenever the third OSD comes up, the cluster could re balance rebuilding >> the parity chunk and re positioning the parts so all OSDs have the same >> amount of data/work. >> >> would this be possible? >> >> ------------------------------ >> *Jorge Pinilla López* >> jorp...@unizar.es >> Estudiante de ingenieria informática >> Becario del area de sistemas (SICUZ) >> Universidad de Zaragoza >> PGP-KeyID: A34331932EBC715A >> <http://pgp.rediris.es:11371/pks/lookup?op=get&search=0xA34331932EBC715A> >> ------------------------------ >> _______________________________________________ >> ceph-users mailing list >> ceph-users@lists.ceph.com >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >> > > -- > ------------------------------ > *Jorge Pinilla López* > jorp...@unizar.es > Estudiante de ingenieria informática > Becario del area de sistemas (SICUZ) > Universidad de Zaragoza > PGP-KeyID: A34331932EBC715A > <http://pgp.rediris.es:11371/pks/lookup?op=get&search=0xA34331932EBC715A> > ------------------------------ >
_______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
