On 3/12/06, huang mingyou <therods@stripped> wrote:
> thank you reply !
> and I want ask another question about the ndb node's heartbeat check.
> if the ndb nodes in 2 wan,and between them have about 80 ms delay.the
> cluster can work correct? and in the 2 wan ,have 2 group ndb nodes,if
> the 2 wan have lose connect.then have data update in each node group
> ,when the network is connect,the cluster can auto coordinate the data
> charge?
> my English is ugly,I'm so sorry, hope you can
> understand my meaning. :->
nothing prevents you from splitting a cluster across data centers. it
will work, but there will be a performance penalty since the system
will often have to traverses the slow(er) link. i have not tried this
myself (yet,) but i have spoken to someone who has. in this case your
using cluster for availability more than performance. i was also told
it would be a good idea to have the management node in a third data
center to help prevent split-brain scenarios.
unless i'm missing something you will have to have numReplicas > the
number of ndb nodes in each data center to ensure that the data exists
in both data centers. otherwise all n replicas could sit in a single
data center and be lost if it goes away. this is just going by what
i've read. there may be an undocumented behavior to address this. it
would be nice to be able to specify node groups (data centers, racks,
etc.) so that ndb could place replicas optimally.
--
-rm
