|
From: Michael P. <mic...@gm...> - 2012-07-16 05:43:19
|
On Sat, Jul 14, 2012 at 2:06 PM, Aris Setyawan <ari...@gm...> wrote: > Hi, > > I was able to configure postgresql 9.1 for total memory usage about > less than(in lowend VPS) 8MB. > That's kind of low... Do you know that shared_buffers is usually set at 25% of your machine shared memory in average for PostgreSQL? > > How about postgres-xc data-node behavior? Is it still the same way with > ordinary postgresql to configure? > Yes, it should be the same. The only differences are: 1) XC uses 2PC internally when a transaction involves more than 2 nodes in a write operation. So be sure to set up max_prepared_xacts to a good value. 2) Set max_connections on Datanodes in accordance with your coordinators. Each Datanode may receive a connection from each Coordinator backend. For example, if you have a cluster with 20 Coordinators having the same max_connections value set. The absolute maximum of connections that 1 Datanode of your cluster can get is 20 * max_connections. You should never reach this absolute value in practice, but just be aware of that depending on your application. > How about GTM memory usage? Can I limit their memory? > There are no GUC parameters that allow in gtm.conf that allow to control the memory used at GTM level. You can still change the maximum number of backends allowed in GTM. This could bring some kind of control on the max memory a GTM uses. -- Michael Paquier http://michael.otacoo.com |