The application of scripts to deadlock avoidance

Abstract

We describe the prototype of an expert system software advisor for the lock manager of a database system. The software advisor, called EAGLE (Expert Advisor for Granting Locks Effectively), is intended to become an embedded expert system within a database management system. EAGLE maintains a record of lock request and lock status within a database management system as an application processes transactions. Eag uses this dynamic lock data to avoid the granting of locks which could lead to a future deadlock. The sequence of lock requests and lock grantings is held as a script(s). EAGLE uses its collected record of lock request sequence to match against stereotypical lock event sequence (script base) and to learn to avoid such sequences in future. As EAGLE gains experience of lock event sequences leading to deadlock it recognises patterns which have led to deadlock, an avoids granting locks which would repeat a previous deadlock-inducing sequence of locks, thereby reducing the occurrence of deadlock. EAGLE treats the deadlock problem as a plan recognition issue rather than a problem resolution issue. We describe the general design of EAGLE, present some results from the EAGLE prototype implementation and discuss planned enhancements to EAGLE.