On Tue, 18 Aug 2009, Jorn Argelo wrote:
For the sake of the KISS concept I decided to scrap the multi-master replication and made it a master-slave setup only, where all 3 boxes point to one master. Then the slave is present as a backup server only. I was expecting to need 2 masters because of performance reasons, but the single master works like a charm in production. I'm sticking to that for now and works surprisingly well with InnoDB and the my-4GB-huge-InnoDB.cnf (or whatever it's called).
I assume you're using autolearning. Master+multislave can support distributed scoring if you're manually learning, as sa-learn can be configured to talk to the master while the SA daemons cal query the slave(s).
Suggestion: open a feature request bug to allow bayes autolearn to use a different database connection string than bayes scoring. That way you could configure all the daemons' autolearns to write to the master, but distribute their scoring queries across X number of replicated slaves...
-- John Hardin KA7OHZ http://www.impsec.org/~jhardin/ jhar...@impsec.org FALaholic #11174 pgpk -a jhar...@impsec.org key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79 ----------------------------------------------------------------------- The real opiate of the masses isn't religion; it's the belief that somewhere there is a benefit that can be delivered without a corresponding cost. -- Tom of "Radio Free NJ" ----------------------------------------------------------------------- 7 days until the 1930th anniversary of the destruction of Pompeii
