本期內容:繼承
1,動態Batch Size深刻it
2,RateController解析date
Spark Streaming中採用的模型以下,基本思想是按照100ms的批次發送數據給Controller,剛開始Controller直接轉給JobGenerator,JobGenerator在生成Job,交給JobProcessor來處理,JobProcessor將job統計信息發送給Controller,Controller接收到統計信息,會動態改變Batch Size來給Job發送數據。sso
RateController的子類有ReceiverRateController、DirectKafkaRateController。方法
ReceiverRateController會向receiverTracker發送UpdateReceiverRateLimit消息。im
ReceiverTracker接收到UpdateReceiverRateLimit消息後,向ReceiverSupervisorImpl轉發UpdateRateLimit消息。統計
ReceiverSupervisorImpl接收到UpdateRateLimit消息後,調用BlockGenerator的updateRate方法,來控制數據的接收速度。數據
BlockGenerator繼承自RateLimiter。img