Constraint-preserving snapshot isolation

Autor: Stephen J. Hegner
Rok vydání: 2015
Předmět:
Zdroj: Annals of Mathematics and Artificial Intelligence. 76:281-326
ISSN: 1573-7470
1012-2443
DOI: 10.1007/s10472-015-9482-6
Popis: A method for detecting potential violations of integrity constraints of concurrent transactions running under snapshot isolation (SI) is presented. Although SI provides a high level of isolation, it does not, by itself, ensure that all integrity constraints are satisfied. In particular, while current implementations of SI enforce all internal integrity constraints, in particular key constraints, they fail to enforce constraints implemented via triggers. One remedy is to turn to serializable SI (SSI), in which full serializability is guaranteed. However, SSI comes at the price of either a substantial number of false positives, or else a high cost of constructing the full direct serialization graph. In this work, a compromise approach, called constraint-preserving snapshot isolation (CPSI), is developed, which while not guaranteeing full serializability, does guarantee that all constraints, including those enforced via triggers, are satisfied. In contrast to full SSI, CPSI requires testing concurrent transactions for conflict only pairwise, and thus involves substantially less overhead while providing a foundation for resolving conflicts via negotiation rather than via abort and restart. As is the case with SSI, CPSI can result in false positives. To address this, a hybrid approach is also developed which combines CPSI with a special version of SSI called CSSI, resulting in substantially fewer false positives than would occur using either approach alone.
Databáze: OpenAIRE
Externí odkaz: