As a PostgreSQL newbie, I've got a question about load-balancing/high-availability which isn't covered by Chapter 25 ("High Availability, Load Balancing and Replication") of the documentation. But first, some background.
Background
----------
I'm planning a distributed database deployment. The deployment will run across a small number of geographically remote sites. I'll use Slony-I for master-slave replication between all machines in the deployment. (The write rate is extremely low and the write latency for remote sites is acceptable.)
There will be 2 (or perhaps more if necessary) database nodes in each site, accessed by many clients. For read-only operations, clients will only access nodes in their local site only. (For write operations, they obviously have to go to the master.)
I want read-only operations to be tolerant of the loss of a single database node in the client's site.
Question
--------
Will a standard TCP load-balancer suffice, or do I *need* to use PgPool in raw mode (or some other equivalent)?
Many thanks,
Andrew