(1)ProvSQL: Provenance and Probability Management in PostgreSQL Pierre Senellart, Louis Jachiet, Silviu Maniu, Yann Ramusat RESEARCH

Texte intégral

(1)ProvSQL: Provenance and Probability Management in PostgreSQL Pierre Senellart, Louis Jachiet, Silviu Maniu, Yann Ramusat RESEARCH. UNIVERSITY PARIS. ÉCOLE NORMALE S U P É R I E U R E. ProvSQL in Brief. Supported Queries and Provenance Annotations. Light-weight extension for PostgreSQL ě 9.5. Large subset of (nested) SQL, without aggregates or UDFs:. Transparent computation of provenance of SQL queries as a term algebra circuit. – SELECT … FROM … WHERE, JOIN, DISTINCT, GROUP BY – UNION, UNION ALL, EXCEPT. Allows probabilistic query evaluation with various techniques:. Variety of provenance formalisms:. – Enumeration of possible worlds – Knowledge compilation to d-DNNF via external tools (c2d, Dsharp, d4) – Monte-Carlo sampling. – arbitrary provenance semirings – arbitrary provenance semirings with monus (m-semirings) – where-provenance. Open-source and available at https://github.com/PierreSenellart/provsql. How ProvSQL Computes Various Provenance Forms and Probabilities Personnel id name. position. city. classification.       . Director Janitor Analyst Field agent Double agent HR Analyst. New York New York Paris Berlin Paris Paris Berlin. unclassified restricted confidential secret top_secret restricted secret. John Paul Dave Ellen Magdalen Nancy Susan. database with provenance annotations. t1 t2 t3 t4 t5 t6 t7. g2 circuit in the universal m-semiring. a. g1. g3. a. ‘. a. city. query evaluation. New York g1 Paris g2 Berlin g3. SELECT DISTINCT city FROM Personnel EXCEPT SELECT DISTINCT P1.city FROM Personnel P1, Personnel P2 WHERE P1.city = P2.city AND P1.id < P2.id. ‘. b. ‘. t1. t2. b. b. b. ‘. b. t3. t5. t6. t4. t7. counting m-semiring (bag semantics). Boolean function m-semiring (Boolean provenance). g2 ´. g2 g1. ^. g1 ␣. ␣. _. t1. _. ^. t2. ␣. ^. t3. ^. t5. _. t6. t4. +. +. ˆ. ˆ. ˆ. ˆ. +. ˆ. 3. 5. 6. 4. 7. ^. 1. t7. 2. ( = 1). ( = 0). g2. d-DNNF. +. _. (deterministic decomposable negation normal form). ˆ. ^ ␣. _. ^. probability computation (independent tuples). t6. ^. +. 1´¨. ˆ. ˆ. ˆ. Pr(t6). 1´¨ ␣. t3. ( = 0). a ´ b = max(a ´ b, 0). knowledge compilation. ^. ´. ^. _. ^. +. ´. g3. ^. g3. 1´¨. ␣. t5. Pr(t3). Pr(t5). References [] Pierre Senellart. Provenance and probabilities in relational databases: From theory to practice. SIGMOD Record, ():–, . [] Pierre Senellart, Louis Jachiet, Silviu Maniu, and Yann Ramusat. ProvSQL: Provenance and probability management in PostgreSQL. Proceedings of the VLDB Endowment (PVLDB), ():–, . Demonstration..

(2)