The role of applicationmaster in spark on Yan's cluster mode

Why? Spark On Yarn Through ApplicationMaster Whereabouts ResourceManager Apply to start Executors, Not through Driver To do these things ？

• Because in standalone In cluster mode ,Driver towards Master apply Worker Resources for starting Executor, And the same , stay Yarn In cluster mode ,Spark The application needs to ResourceManager apply NodeManager Resources for starting Executor, This responsibility can make ApplicationMaster To do , Rather than let Driver To do , It can make Driver More relaxed , Achieve decoupling .