Flipping Control Flow of DBMS Execution Engine

With hardware advancements, ever-changing customer demands and evolving computing environment, DBMS query engines remain to be a relevant research topic. OLAP workloads are growing in popularity, complexity and volumes of analyzed data. Computing environment of execution engines is shifting towards managed multi-tenant DBMS installations offered by the cloud provider. Related work suggests multiple optimizations that improve execution engine performance for the mentioned scenario. Their application to an existing executor, however, can be difficult. Traditional query engines use a pull-model to execute queries, while the mentioned optimizations are only applicable to a push-based engine. Rewriting the executor from scratch is time-consuming and can cause regressions, stall of other ongoing projects and loss of developer expertise. This work explores an alternative approach, which is to modify the existing executor to use a push-model while retaining most of its original code-base intact. The procedure of "Push-based conversion" and results of its application to an existing executor are described. It is concluded that writing a new executor is preferred over the mentioned procedure, unless the prior is considered impossible.