Here are my notes from a conversation with Andy Salnikov, Simon Krughoff, and Chris Morrison:
Andy has implemented an SQLAlchemy-based prototype system for testing the L1 database system: https://github.com/lsst-dm/l1dbproto He's been testing it on MySQL and Postgres at IN2P3. The ap_proto script (https://github.com/lsst-dm/l1dbproto/blob/master/bin/ap_proto) simulates the L1 workflow.
We'll need to have a database set up on whatever system we want the MVS pipelines to run on. Andy recommends Postres on SSDs for best performance, but that's not possible yet at NCSA. We will investigate just using lsst-db at NCSA for now, and possibly also an AWS system for continuous integration.
We discussed the question of whether we expect the science pipelines to contain SQL directly or if those functions might be dealt with by the Butler in the future. Andy thinks that for the L1 database performance will likely be impacted enough that we will want to work with direct database access. For L1 pipeline minimum viable system we (UW) will write a small abstraction layer so we don't bake our assumptions in too deeply. It's possible the l1dbproto package has much of this functionality already.
Andy Salnikov will send us the l1db.cfg file and point us to the most appropriate ticket branch of l1dbproto.