A Research And Implementation Of Provenance For Relational Databases

Data provenance is the information besides data to explains where the data comes from(provenance data)and how the data transfroms from(provenance transformation).Using data provenance,it's possible to compute quality of data,trace error,debug data,understand references between data,etc.However,many researchs of provenance in RDBMS have some limitations in supporting full-feature SQL or procedure languages,and most researchs pay little attention on provenance transformation.To solve the limitations,we propose a provenance model of RDBMS,which can help understanding relationship/references between data in different levels.Finally we implement the model in PostgreSQL.Specificly,our works contains several aspects:(1)Using the Where-Provenance which is a copy-contribute semantic to define provenance data,and proposing a definition named Pet(means Provenance explains transformation)to explain the copy-transformation of data in queries.We also provide a comprehensive explanation on how the Pet fit full-feature SQL and procedure languages.(2)Base on Where-Provenance and Pet,we propose a directed graph named SPGraph to descript provenance relationship in relational schema level,at the mean time use sub-graph of SPGraph to descript provenance transformation for every tuple,which named TupPRoute.Derivatively we propose FPGraph and VPRoute for field level of schema and value level.(3)Extending PostgreSQL to implement features including extracting,querying and visualizing the data provenance.Specificly,we modify the plan-executor and PL/PgSQL module to extract SPGraph of query and TupPRoute of each tuple exports.To query SPGraph and TupPRoute,we provide a gui-client which also including the visualization work,responding to user,client can extract the FPGraph/VPRoute from SPGraph/TupPRoute.It's worth mentioning that our implemention takes no apparent burden for query engine.