Monitoring the progress of a long-running analysis query is a cool problem -- it's been discussed on -hackers a few times in the past (e.g. by Greg Stark). In that thread, I pointed to some Wisconsin research on this topic (2004, 2006). That work was prototyped with Postgres. I just noticed that there's apparently a new project at the DB group at U of T that is tackling similar problems: ConEx. Apparently they are also using Postgres to build their prototype, which is always cool to see.