[jordanchan]

You raise a good point. I don't expect graph queries that go more than a few links away. To the extent that there will be such queries, I think appropriate indexing will help a great deal.

Implementation wise, it is not that different from a query with many joins anyway - which is a standard problem, and is addressed by materialized views as well as indexing. So as long as the complex queries don't need to be ad hoc and real time, I can prepare for them using materialized views.

However you are quite right in that complex, ad hoc, and realtime graph queries Will be a pain point. That is a risk I am prepared to tolerate in the short term. The thing is, I also need transactional integrity, so I am stuck with an rdbms as the primary solution. In the long term, I anticipate the need for both graph queries and transactions - thus, I think it is better to start with postgres and incorporate a graphdb as the need arises than the other way round.

Hopefully this line of reasoning make sense? I am really not experienced at this, so am trying to not bump into too many walls.