On Thu, May 28, 2020 at 8:44 AM Hans van den Bogert <hansbogert@gmail.com> wrote:
Hi list,
When reading the documentation for the new way of mirroring [1], some questions arose, especially with the following sentence:
Since this mode is not point-in-time consistent, the full snapshot delta will need to be synced prior to use during a failover scenario.
1) I'm not sure I follow. Why are snapshots not point-in-time, whereas using the journal-based mirroring is?
I agree that it's probably a poor choice of verbiage. Both are technically crash / point-in-time consistent. The difference is that journal-based mirroring provides a replayable stream of updates that are synced with a very fine granularity (down to single IO updates) whereas snapshot-based mirroring has a very coarse granularity. 2) Also, I am not sure what the implications are for the second part of
the referenced sentence, what significance does the word "full" bear in "full snapshot delta" ?
Given mirroring snapshot A and B, the full delta of changes between snapshots A and B will need to be successfully synced / applied to the non-primary image. If you force a failover before a the delta between snapshot A and B can be applied, the system will rollback the non-primary image to snapshot A to get the non-primary image to a consistent state.
Further, Isn't it obvious that in order to do a failover, the other side should have the latest snapshot? - I am probably missing what the documentation is trying to tell me.
Not all failovers are planned. All in all, I think some background concepts are missing in order to
read this chapter of the documentation. It might well be that this is explained somewhere else in the docs, but maybe we should link to that.
If someone can explain the above 2 points, I'm willing to take a shot at making a patch if I can make this clearer in the docs.
Please feel free to submit any clarifying updates.
Thanks,
Hans
[1] https://docs.ceph.com/docs/octopus/rbd/rbdcourse-mirroring/ <https://docs.ceph.com/docs/octopus/rbd/rbd-mirroring/> _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Jason