On Wed, Dec 01, 2004 at 06:39:00PM +0000, Tim Bunce wrote: > Lots of people do, it seems, but I'm not getting much background about why. > > FWIW, the reason I'm digging here is because I agree there may be > some value in the DBI supporting something along these lines, but > I need a better understanding of the underlying issues. More real- > world examples would help. > > It'll always come down to the issue of "why not store complete DSNs?" > and so far that's not been well covered by the feedback I've got.
Here's one from a place where I used to work, where we had a very similar module to the one proposed: A given server would be set up to provide a number of "backend" services, including an RDBMS, a Web Server, and more. We had a number of similar servers for similar purposes, including (but not limited to) failover. To configure which server(s) to use, we had a simple config of the form server = victoria rdbms = postgres http_port = 8080 [etc] and so failing over to another server is a matter of simply changing the [server] portion., likewise (as we did for performance testing, or differing SLAs), swithcing the RDBMS used. If we had used the DSN directly, we would be duplicating information such as the hostname (and the RDBMS type, which is used elsewhere for construction & admin tasks); redundancy in this form leads to frustrating silly errors consuming time (such as when one value gets changed but the another is forgotten). Mx.
