Re: [bailey-developers] lattice master

[Yonik S. <yo...@ap...>]

On Fri, Feb 29, 2008 at 11:23 AM, Ning Li <nin...@gm...> wrote:
> On Fri, Feb 29, 2008 at 9:08 AM, Yonik Seeley <yo...@ap...> wrote:
>  >  We'll need to make sure that it handles this case:
>  >  Say C syncs from B, then B syncs from A and get some new docs.  Now A
>  >  goes away for a while... it seems like C will need to get the new docs
>  >  from B next time it syncs.  If the docs that B got from A are treated
>  >  like any other new additions, this should work right?
>
>  I think this is how it works with replication=3:
>  Node A serves range A-D, node B serves range B-E, node C serves range C-F...
>  Node C syncs with nodes A & B on range C-D, syncs with nodes B and D
>  on range D-E, and syncs with nodes D & E on range E-F.
>
>  Now node D goes away. Node C continues to serve range C-F until it gets
>  range F-G from node E or node F and starts to serve range C-G. Then,
>  node C syncs with nodes A & B on range C-D (same as before), syncs with
>  nodes B and E on range D-F, and syncs with nodes E & F on range F-G.
>
>  Right?

The issue I'm concerned about is losing documents that are only
partially replicated and there are changes to the configuration.  It's
not clear to me if you were answering that or not.  I guess that's why
we need a good simulator!

-Yonik