For most user-submitted tasks, MaxCompute first analyzes them and then generates a task execution plan. The execution plan is composed of multiple execution stages that are dependent on each other.
Currently, the execution plan is displayed as a directed acyclic graph. The vertex in the graph designates the execution phase, while edges of the graph indicate the dependence of each execution phase. MaxCompute follows the dependency of execution plan to run each phase.
In an execution stage, multiple processes, known as Workers, complete the computing work. Different Workers handle different data, but the execution logic is the same. Computational tasks are executed directly in MaxCompute instances, for example, Status Instance and Kill Instance.
For MaxCompute tasks that are not computational tasks, such as DDL statement in SQL, these tasks can only read and modify the metadata information in MaxCompute. This means that no execution plan can be analyzed and generated from the task.