Feature of PGCluster-1.3

PGCluster-1.3 is the synchronous replication system of the multi-master system for PostgreSQL (8.0).

License BSD License, same as PostgreSQL.
Function Load Balancing Since PGCluster is the synchronous replication of a multi-master, it can acquire the same value even if it refers to different Cluster DB. It is very suitable for load balancing for reference system.
When the load of an updating process is high, you may use partial replication feature to decentralize the load.
Problem Handling Cluster server can be set up three or more machines. When a problem occurs in Cluster DB, loader balancer and a replication server will be separated from that machine, and continues service using the other Cluster DBs.
Several replication server can be set up. When an problem occurs at the replication server, Cluster DB automatically changes connection to the standing-by replication server.
When all replication servers stop, Cluster DB operates in stand-alone mode. There are two modes for stand-alone mode: read only mode which only permits references, and read write mode, which permits updating.
Several loader balancer can be set up, however, they will not have the problem detection and switching over functions that the loader balancer itself has.
Recovery When the Cluster DBs were started with a recovery option, it will come back to a replication system after taking the synchronization to master DB's data. If there are three or more Cluster DBs, a query can be processed without stopping service during recovery.
Replication Query-based The replication is done in query-based method.
COPY Since data read from the COPY (FROM) command are also replicated, the backup files which were created by pg_dump can also be replicated.
NOW( ) When inserting the present time data with a NOW function (ex. the NOW function is contained in the query, or the default value of a table is being set to 'NOW'), that values will also be replicated. The time will be the time of a replication server.
The sequence operation function (nextval, setval) will be replicated.
Serial type Since the order of INSERT processing is controlled to become the same in each Cluster DB, serial type data is synchronized as well.
Stored procedure The stored procedure accompanied by data update is replicated.
Large Object The data of large object is replicated.
OID The OID is replicated only the large object.


(1) When replicating a large object, it needs to be placed on the directory which the file of a large object can read from all the clusters DB.