Lossless Conditional Schema Evolution

Conditional schema changes change the schema of the tuples that satisfy the change condition. When the schema of a relation changes some tuples may no longer fit the current schema. Handling the mismatch between the intended schema of tuples and the recorded schema of tuples is at the core of a DBMS that supports schema evolution. We propose to keep track of schema mismatches at the level of individual tuples, and prove that evolving schemas with conditional schema changes, in contrast to database systems relying on data migration, are lossless when the schema evolves. The lossless property is a precondition for a flexible semantics that allows to correctly answer general queries over evolving schemas. The key challenge is to handle attribute mismatches between the intended and recorded schema in a consistent way. We provide a parametric approach to resolve mismatches according to the needs of the application. We introduce the mismatch extended completed schema (MECS) which records attributes along with their mismatches, and we prove that relations with MECS are lossless.