Reducing query optimizer plan regressions with machine learning classification

A non-transitory computer-readable tangible medium, on which is recorded a computer program, the computer program comprising executable instructions, that, when executed, perform a method comprising:

• a database system receiving a query;
• the database system retrieving an old query execution plan (QEP), OldPlan, for the query;
• the database system submitting the query to an optimizer;
• the optimizer returning a new QEP, NewPlan, for the query;
• the database system submitting the OldPlan and the NewPlan to a machine learning classifier (ML classifier);
• the ML classifier computing a feature vector describing a difference between:
• an OldPlan graph of the OldPlan, wherein the OldPlan graph includes:
• a plurality of OldPlan nodes, each associated with a physical step in the OldPlan, each OldPlan node including regression-relevant properties and costs previously determined to be relevant to performance regression, and
• one or more OldPlan edges, each associated with a connection between OldPlan nodes, and
• a NewPlan graph of the NewPlan, wherein the NewPlan graph includes:
• a plurality of NewPlan nodes, each associated with a physical step in the NewPlan, each NewPlan node including the regression-relevant properties and costs, and
• one or more NewPlan edges, each associated with a connection between NewPlan nodes;
• the ML classifier predicting, based on the feature vector, that executing the NewPlan will result in a performance regression as compared to executing the OldPlan;
• and
• the database system executing the OldPlan instead of the NewPlan.