Stage and Source Scheduler and Grouped Execution

Background

When Presto executes a query, it does so by breaking up the execution into a hierarchy of stages. Stage models a particular part of a distributed query plan. Each stage reads from data source and writes to an output buffer, and there are two types of sources:

Table scan source (sometimes referred to as partitioned source in code)
Remote source (sometimes referred to as unpartitioned source in code)

The following figure shows part of a query plan with two stages:

Two Stages

See more details about stage at https://prestodb.github.io/docs/current/overview/concepts.html#stage

Stage Scheduler

StageScheduler is responsible for the following jobs:

Create tasks for this stage
Schedule table splits to tasks

This section will discuss about stage scheduler before grouped execution is introduced. ScaledWriterScheduler is not discussed here.

The StageScheduler interface is quite concise (https://github.com/prestodb/presto/blob/b430a562679aab0a04df37b7a1e77cfd5d941c81/presto-main/src/main/java/com/facebook/presto/execution/scheduler/StageScheduler.java):

public interface StageScheduler
        extends Closeable
{
    /**
     * Schedules as much work as possible without blocking.
     * The schedule results is a hint to the query scheduler if and
     * when the stage scheduler should be invoked again.  It is
     * important to note that this is only a hint and the query
     * scheduler may call the schedule method at any time.
     */
    ScheduleResult schedule();

    @Override
    default void close() {}
}