On Tue, Apr 3, 2018 at 11:36 PM, YuShengzuo <yu.shengzuo@xxxxxxxxxxx> wrote: > Hi Jason, > > > > Some questions about rbd-mirror: > > > > About the ‘split brain’ state of RBD: > > One use case is primary cluster all down, then promote secondary cluster > rbd. > > And Now if we recover primary cluster and demote rbd to sync from one in > secondary. > > The state of this rbd will be ‘split-brain’, and we have to do ‘image > resync’. > > > > In this case, > > why the ‘split-brain’ state is necessary? The split-brain state is necessary because site A and site B versions of the image can now be different. > Any soft way to do ‘resync’ , just sync different data from secondary > rbd-journal? Not right now but it would be possible provided that the force-promoted image was 100% up-to-date w/ the original primary image. There is also a goal (of mine at least) to get OSD support for deep-deletion of objects (where all snapshot clones are deleted, not just the HEAD revision) so that could roll-back changes from the original primary image by copying the affected objects from the force-promoted image. > （So violence about resync， delete and wait 30s to sync again） > > About network of rbd-mirror process: > > The bandwidth will be used much when syncing. Any way to set the work of > rbd-mirror to use > > an independence network card ? > If your site A and site B Ceph clusters are in different subnets, standard network routing rules would move your mirroring traffic to the specified NIC. > > > > That is all, Hope your return. Thanks > > > > Best regards > > Yu Shengzuo > > -- Jason -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
