On Mon, Oct 17, 2016 at 4:58 AM, Manuel Lausch <manuel.lau...@1und1.de> wrote:
> Hi Gregory,
> each datacenter has its own IP subnet which is routed. We created
> simultaneously iptables rules on each host wich drops all packages in and
> outgoing to the other datacenter. After this our application wrote to DC A,
> there are 3 of 5 Monitor Nodes.
> Now we modified in B the monmap (removed all mon nodes from DC A, so there
> are now 2 of 2 mon active). The monmap in A is untouched. The cluster in B
> was now active as well and the applications in B could now write to it. So
> we wrote definitely data in both clusterparts.
> After this we shut down the mon nodes in A. The part in A was now
> unavailable.
> Some hours later we removed the iptables rules and tried to rejoin the tow
> parts.
> we rejoined he three mon nodes from A as new nodes. the old mon data from
> this nodes was destroyed.
> Do you need further information?

Oh, so you actually forced both data centers to go active on purpose.
Yeah, there's no realistic recovery from that besides throwing out one
side and then adding it back to the cluster in the other DC. Sorry.
ceph-users mailing list

Reply via email to