Query Stages

Nov 2, 2018

Overview

Apache Drill is a system for interactive analysis of large-scale datasets. It was designed to allow users to query across multiple large big data systems using traditional query technologies such as SQL. It is built as a flexible framework to support a wide variety of data operations, query languages and storage engines.

Query Parsing

A Drillbit is capable of parsing a provided query into a logical plan. In theory, Drill is capable of parsing a large range of query languages. At launch, this will likely be restricted to an enhanced SQL2003 language.

Physical Planning

Once a query is parsed into a logical plan, a Drillbit will then translate the plan into a physical plan. The physical plan will then be optimized for performance. Since plan optimization can be computationally intensive, a distributed in-memory cache will provide LRU retrieval of previously generated optimized plans to speed query execution.

Execution Planning

Once a physical plan is generated, the physical plan is then rendered into a set of detailed executional plan fragments (EPFs). This rendering is based on available resources, cluster load, query priority and detailed information about data distribution. In the case of large clusters, a subset of nodes will be responsible for rendering the EPFs. Shared state will be managed through the use of a distributed in-memory cache.

Execution Operation

Query execution starts with each Drillbit being provided with one or more EPFs associated with query execution. A portion of these EPFs may be identified as initial EPFs and thus they are executed immediately. Other EPFs are executed as data flows into them.