Hi James,
Cannot help you too much on the specifics, but can give you some
general opinions. We were very much in the same situation, and there
just wasn't the money for Oracle or Sybase at that point. I wasn't going
to use MySQL, as a system without transactions is not a database IMHO.
So I started using postgres at version 6.1 or 6.2. Since then I've seen
postgres develop in leaps and bounds and I think it has got an excellent
developer community. If you work on interfaces to postgres, put them in
the public domain and you are liable to get quite a bit of help and lots
of advice. And it is invaluable to have the source-code: I run on
Alphas, and in some of the earlier versions things didn't always run as
they should on Alphas. Having the source code I managed to track down
quite a few problems and fix them in hours -- try that with Oracle
support. Also, some of the enhancements promised in 7.0 and 7.1 are
going to move postgres ahead by quite a bit again.
On the minus side: tools are lacking in a big way. Currently you cannot
change constraints on columns or drop columns. Until 7.1 you are stuck
with an 8 K tuple length (although Jan Wieck is coming up with a very
nice solution to this). Referential integrity is coming in 7.0.
Currently you can still have trouble with views, as the query plan is
too big to fit into the system tables (I believe this is supposed to be
largely solved in 7.0 by compressing the query plans). You cannot have
unions in views. Outer joins may be in 7.0. And a query-tree redesign is
in the offing to enable the implementation of some more complex queries.
So quite a few things that are kind-of necessary for a production system
are on the verge of happening, but not there yet. Of course you may be
able to spare somebody to give a hand with some of these issues if you
save so much money from not buying Oracle ;-)
Even so, I have been running a production system on postgres for quite a
while now. And yes, referential integrity is implemented with kludgy
triggers, etc. But it works just fine. So i would suggest: have a look
at the current to-do list, decide whether you can live with it or
otherwise help to do something about it, and put together a small sample
system and put it under load.
BTW, I seem to recall an announcement from ColdFusion that they were
bringing out a Linux version with Postgres. It may be worth checking
that out.
Good luck,
Adriaan
