Eric Ridge <ebr@tcdi.com> writes:
> Xapian has it's own storage subsystem, and that's what I'm using to
> store the index... not using anything internal to postgres (although
> this could change).
I would say you have absolutely zero chance of making it work that way.
You will not be able to get it to interoperate reliably with
transactions, checkpointing, or WAL replay; to say nothing of features
we might add in the future, such as tablespaces and point-in-time recovery.
You need to migrate all the data into the Postgres storage mechanism.
It might be worth pointing out here than an index AM is not bound to use
exactly the typical Postgres page layout. I think you probably do have
to use the standard page header, but the page contents don't have to
look like tuples if you don't want 'em to. For precedent see the hash
index AM, which stores ordinary index tuples on some index pages but
uses other pages for its own purposes.
regards, tom lane