Skip to content

PluginLifecycle.md

Latest commit

 

History

History
21 lines (16 loc) · 1.18 KB

PluginLifecycle.md

File metadata and controls

21 lines (16 loc) · 1.18 KB

Spark Cyclone Plugin Lifecycle

The general flow is as such:

  1. The Driver and Executor plug-ins are launched
  2. Each Executor plug-in spawns its own VE process
  3. SQL queries pass through VeRewriteStrategy to capture any plans that coule be processed by the plug-in; expression evaluation happens here.
  4. CombinedCompilationColumnarRule compiles all the generated C++ sources using NCC into a .so, and rewrites all the plans to reference the new .so that is located on the driver.
  5. Spark eventually calls a VectorEngineToSparkPlan to produce an RDD of data, which calls SupportsVeColBatch#executeVeColumnar of its child plan, and its child plan does the same. During this calling, VeProcess of the executor is summoned. Data is freed at the earlist possible opportunity or at Task completion.

Compilation lifecycle

The Spark Cyclone plugin will translate your Spark SQL queries into a C++ kernel to execute them on the Vector Engine. Compilation can take anywhere from a few seconds to a couple minutes. While insignificant if your queries take hours you can optimize the compilation time by specifying a directory to cache kernels using the following config.