Hi Sylvain, Thanks for pointing this out. Is this clearer? * The default CRUSH rules and layouts are now using the 'bobtail' tunables and defaults. Upgaded clusters using the old values will now present with a health WARN state. This can be disabled by adding 'mon warn on legacy crush tunables = false' to ceph.conf and restarting the monitors. Alternatively, you can switch to the new tunables with 'ceph osd crush tunables firefly,' but keep in mind that this will involve moving a significant portion of the data already stored in the cluster and in a large cluster may take several days to complete. * We now default to the 'bobtail' CRUSH tunable values that are first supported by Ceph clients in bobtail (v0.56) and Linux kernel version v3.9. If you plan to access a newly created Ceph cluster with an older kernel client, you should use 'ceph osd crush tunables legacy' to switch back to the legacy behavior. Note that making that change will likely result in some data movement in the system, so adjust the setting before populating the new cluster with data. sage On Mon, 7 Jul 2014, Sylvain Munaut wrote: > Hi, > > > In the release notes for firefly, more precisely in the "upgrading > from emperor" section, you can find theses two notes : > > > * The default CRUSH rules and layouts are now using the latest and > greatest tunables and defaults. Clusters using the old values will now > present with a health WARN state. This can be disabled by adding ?mon > warn on legacy crush tunables = false? to ceph.conf. > > * We now default to the ?bobtail? CRUSH tunable values that are first > supported by Ceph clients in bobtail (v0.56) and Linux kernel version > v3.9. If you plan to access a newly created Ceph cluster with an older > kernel client, you should use ?ceph osd crush tunables legacy? to > switch back to the legacy behavior. Note that making that change will > likely result in some data movement in the system, so adjust the > setting before populating the new cluster with data. > > > And they don't agree with each other. The "latest and greatest > tunables" are _not_ the one from bobtail as the second item says. The > 'optimal' tunables (the one you get from "ceph osd crush tunables > optimal") include "tunable chooseleaf_vary_r 1" and that's rather a > big deal since it will break RBD for anything other kernel than >= > v3.15 . > > > Cheers, > > Sylvain > _______________________________________________ > ceph-users mailing list > ceph-users at lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >
 |