On Sun, May 27, 2012 at 8:38 AM, Joseph Glanville <
jos...@or...> wrote:
> Hi List,
>
> I have bene taking a look at postgres-xc (installing a test cluster
> atm) but I was wondering if anyone could clarify what setups are
> available to facilitate HA/failover/replication etc?
> We primarily need read scalability and would like to syncrnously
> replicate all data to all or atleast a subset of datanodes with a hope
> that postgres-xc etc would handle failover and recovery seamlessly.
> I saw some presentation indicating this wasn't the case and that you
> require streaming replication in adition to postgres-xc clustering but
> there isn't much in the way of firm information that I could find.
>
You only need to use syncrep on a node when this node contains data that is
unique in the cluster. This is for example necessary if you distribute a
table data by hash, modulo or round robin among multiple nodes. Those nodes
need to be covered by any syncrep mechanism at node level. This is
particularly useful in case of disaster recovery.
If you plan to replicate all your tables at XC level, I mean that you use
the extension DISTRIBUTE BY REPLICATION in CREATE TABLE on a subset of
nodes (at least 2), in this case you do not need any syncrep as all your
data will be already replicated among multiple nodes. However, such a
configuration is expensive in writes, but perhaps in your application your
replicated tables are not updated that often.
For more details about recovery and HA stuff, the document indicated by
Koichi is really useful.
--
Michael Paquier
http://michael.otacoo.com
|
