On 08/19/10 20:19, Benjamin Smith wrote: > On Wednesday, August 18, 2010 04:58:08 pm Joshua D. Drake wrote: >> Well if you are just using it for updates to the schema etc... you >> should only need to launch a single connection to each database to make >> those changes. > > And that's exactly the problem. On each server, we have at least dozens of > active databases - one for each client. Opening a connection for each database > starts to become problematic. Hmm, let's talk about this problem. If I understand you correctly, you have an arrangement where you have one server which will push updates, and other, database servers, with "dozens" of databases per server? Let's try to convert this to numbers and assume that "dozens of databases per server" means "50" and that you have "50 more servers". This means that the server which pushes the updates needs to connect to 2500 databases. This is way to small a number of connections (sockets) from a single client to create problems on the client side. On the other hand, if you have, say, 50 databases per server, this means that the update clients connects 50 times to the same server, which is only a problem if the server has a small total number of connections configured (max_connections) - smaller than 2x the number of databases. Unless you are very careful not to actually exceed this number of connections during normal database use (and depending on what you do this may or may not be possible), it would actually be more benefitial to raise max_connections to a sensible value - e.g. 500 (which would probably need a bump in SEMMNI). -- Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general