Hello,
I had observed that the quorum does not holding values which was set on the fly previously if every mon got restarted but remaining set of monitors was enough for a quorum at every moment, seems to be a bit non-logical. For example, down-out interval will be reset to the config values if such operation is performed, so it breaks idea of the continuous operation - I`m seeing that the cluster has a functioning quorum over time, but if someone restarted mons one by one, values are set previously in runtime is dropped to the config defaults and I have a wrong picture of the parameters set at the end, without knowledge of every mon and this could be happening natively on long production runs, where mons will be turned off and on in time by hardware means, for example. Though it is quite hard and ugly to say "since quorum has started and until it breaks, discard config values at the mon` join and accept only runtime config changes", there is no obvious idea of how to make it better.
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
