While investigating a report of very slow fact-paths responses we discovered that analyzing the value_types table provided a roughly 4x improvement for 10.5m facts and 629k fact_values. Further examination demonstrated that the table had never been analyzed. We believe this is because the autovacuum analyzer won't look at a table until it passes certain thresholds in both size and update/delete counts.

To fix that, we're planning to add code to run a vacuum analyze (or possibly just analyze) at startup whenever at least one migration fires. In addition we'll add a new dummy migration to make sure all existing tables are analyzed.