[Openais] [whitetank / corosync trunk] Fix checkpoint sync in certain scenarios

Andrew Beekhof beekhof at gmail.com
Fri Nov 7 05:48:49 PST 2008


On Fri, Nov 7, 2008 at 10:24, Steven Dake <sdake at redhat.com> wrote:
> In a certain rare scenario, the checkpoint service throws away the
> current checkpoint database.
>
> An example of when this occurs is when there are 3 nodes A, B, C, node A
> and C are killed

Does it have to be killed, or could shutdown trigger this too?

> then node B syncs.  After this completes, Node C is
> started and node B again begins resyncing, but during this sync process
> node A starts up.
>
> This results in node b no longer believing it is required to sync its
> current database contents.  The abort called on node b throws away all
> checkpoints in the system but since node b is no longer the lowest node
> id in the system it believes it doesn't have to sync.
>
> The design change is that once a node has been declared as a responsible
> for synchronization, any aborts or configuration changes will never
> change the fact that node is still responsible for synchronization.
>
> Regards
> -steve
>
> _______________________________________________
> Openais mailing list
> Openais at lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/openais
>


More information about the Openais mailing list