Schema evolution and foreign keys: a study on usage, heartbeat of change and relationship of foreign keys to table activity

Abstract

In this paper, we study the evolution of foreign keys in the context of schema evolution for relational databases. Specifically, we study the schema histories of a six free, open-source databases that contain foreign keys. Our findings verify previous results that schemata grow in the long run in terms of tables. To our surprise, we discovered that foreign keys appear to be fairly scarce in the projects that we have studied and they do not necessarily grow in sync with table growth. In fact, we have observed different “cultures” for the handling of foreign keys, ranging from treating foreign keys as an indispensable part of the schema, in full sync with the growth of tables, to the unexpected extreme of treating foreign keys as an optional add-on that twice resulted in their full removal from the schema of the database. Apart from the behavior of entire schemata, we have also studied the behavior of individual tables. We model the schema of any version of the history as a graph, with tables being nodes and foreign keys being edges. The union of these graphs is called the diachronic graph of the schema and contains all the tables and foreign keys that ever appeared in the schema history. The study of the total degree of tables at the diachronic graph, reveals several patterns. The population of tables with total degree in the range of [0–2] includes almost all the tables that were eventually removed from the schema, as well as the vast majority of survivor tables. These low-degree tables (especially the dead ones) tend to be mostly with zero or very few internal updates in their entire history. At the same time, the few tables with degree higher than 2 are typically born very early in the life of the schema, overwhelmingly survivors, and, unusually active, typically undergoing medium or high update activity.