On Sat, Jul 23, 2011 at 3:39 AM, Andrew Dunstan
<andrew@dunslane.net> wrote:
1. I think the proposed use is of very marginal value at best, and certainly not worth importing an external library for.
Now that I've seen two people who seem to think that this is not an important feature I'll wade in and respond to this idea.
I think it's very easy to doubt the value of a definitively recognizable string that represents a postgres database when you don't have a heterogenous environment with more than a hundred thousand applications of all types in it. To make matters worse, as language support on that platform continues to widen beyond its humble beginnings, there isn't a standard across those languages for what constitutes a postgres URL. This is the current situation at Heroku, where we currently run ~150,000 individual databases on our infrastructure as well as a variety of other databases such as MySQL, Redis, Mongo, Couch, Riak, Cassandra, &c.
To head off the most obvious criticism, we aren't using connection strings in our system because there isn't any reasonable way to recognize them. A PG conninfo string is just a set of key value pairs with no dependably present signifier. This is why almost every database library from Ruby to Python to Java takes some form of a URL with a protocol called "postgres" in it in order to help select which driver to use.
Further, support (and syntax!) for the more esoteric connection parameters varies from library to library as well as between languages. A good spec by the project would go a long way in resolving this, and I can at least be confident that we could get it adopted very quickly by all three of the Ruby-community Postgres libraries.
In conclusion, this is a serious operational concern for me and my team and I will be personally dealing with fires caused by this for years to come regardless of the outcome of this thread.
Best,
-pvh
--
Peter van Hardenberg
Department of Data
Heroku
"Everything was beautiful, and nothing hurt." -- Kurt Vonnegut
