Transactions and indexes

Noob here.

I'm trying to implement a new type of index access method. I don't think I
can use the built-in storage manager and buffer manager efficiently because
files with 8k blocks aren't going to work. I really need to implement a
log-structured file.

I'm confused on how to handle transactions and visibility. I don't see
anything in the index action method functions (am*()) that tell me when to
commit or rollback new index entries, or which transaction we're currently
in so I can know whether recently-added index entries should be visible to
the current scan. I'm guessing that all that magically happens in the
storage and buffer managers.

So... how do I handle this? Is there some way for me to implement my own
storage manager that manages visibility?

I'd be grateful for any guidance.

--
Chris Cleveland
312-339-2677 mobile