| In the past few years there has been considerable interest in expert database systems, i.e. in integrating conventional data base managers and rule based production systems. POSTGRES is a tightly coupled expert database system, where the rule system is integrated with the data manager. The rules follow the familiar event-qualification-action paradigm, and provide both backward and forward chaining semantics. The POSTGRES rule system can be used to provide data base services including view and version processing, alerters, integrity control and protection.;This dissertation explores the semantics and the implementation of the POSTGRES rule system. The possible options concerning the time of rule activation are discussed. Rule execution can be either immediate (at the time the event happens) or deferred (at the end of the command or transaction). A priority scheme for rule conflict resolution is also presented. Exclusive priorities can be used to decide which rules to activate when more than one apply, and sequential priorities determine the order of rule execution.;There are two alternate implementations for rules in POSTGRES. The POSTGRES tuple level system implementation is described in detail. This system processes rules on a tuple by tuple basis. The rule activation mechanism is based on a physical marking (locking) scheme. When a rule is defined special markers, called rule locks, are placed in the tuples that satisfy the rule's qualification. Rule stubs are used to appropriately place rule locks to tuples inserted after the rule definition. The alternate implementation, based on a query rewriting scheme is also discussed. User queries are combined with rules to generate a set of new queries that when executed create the desired effect. A performance analysis that compares the tuple level and query rewrite system implementations is presented.;Finally, three other more complex, higher level rule systems are presented. These systems can be built on top of the basic POSTGRES rule system. The first provides deferred execution semantics. The second can be used to enforce integrity constraints, and the third one allows access to the sets of tuples that have been inserted, deleted or updated during a transaction. |
