Storm-源码分析-acker (backtype.storm.daemon.acker)

backtype.storm.daemon.acker
设计的巧妙在于, 不用分别记录和track, stream过程中所有的tuple, 而只需要track root tuple, 而所有中间过程都通过异或更新track entry

acker-init, 在spout发送一个tuple时触发, 初始化这个root tuple的track entry 
acker-ack, 在blot ack一个tuple的时候触发, 会对该tuple的anchors-to-ids中记录的每个(root, edge)进行ack, 并出于优化还会附带登记新的edge(对acker透明, 在发送前已经完成)
acker-fail, 任一个过程中的tuple fail, 都会导致这个root tuple失败

 

(defn mk-acker-bolt []
  (let [output-collector (MutableObject.)
        pending (MutableObject.)]
    (reify IBolt
      (^void prepare [this ^Map storm-conf ^TopologyContext context ^OutputCollector collector]
               (.setObject output-collector collector)
               (.setObject pending (RotatingMap. 2)) ;;用RotatingMap来缓存每个tuple的track信息
               )
      (^void execute [this ^Tuple tuple]
             (let [^RotatingMap pending (.getObject pending)
                   stream-id (.getSourceStreamId tuple)]  ;;从ack tuple中取出streamid
               (if (= stream-id Constants/SYSTEM_TICK_STREAM_ID) ;;收到system_tick_stream, rotate pending, spout的pending和acker的pending超期时间是一样的, 都取决于system-tick
                 (.rotate pending)
                 (let [id (.getValue tuple 0) ;;else,其他的stream,取出tuple id
                       ^OutputCollector output-collector (.getObject output-collector)
                       curr (.get pending id) ;;取出相应tuple的track entry
                       curr (condp = stream-id
                                ACKER-INIT-STREAM-ID (-> curr  ;;初始化tuple的track entry
                                                         (update-ack (.getValue tuple 1)) ;;更新entry中的track value
                                                         (assoc :spout-task (.getValue tuple 2))) ;;记录该tuple和spout-task的关系, 这样在ack或fail的时候才知道通知谁
                                ACKER-ACK-STREAM-ID (update-ack curr (.getValue tuple 1));;ack, 用val和原来的entry value做异或
                                ACKER-FAIL-STREAM-ID (assoc curr :failed true))] ;;fail, 直接把entry的:failed设true
                   (.put pending id curr)
                   (when (and curr (:spout-task curr))
                     (cond (= 0 (:val curr)) ;;val为0, 表示该tuple的所有edge都被成功ack
                           (do
                             (.remove pending id) ;;从pending中删除track entry, 并向相应的spout-task发送ack消息
                             (acker-emit-direct output-collector
                                                (:spout-task curr)
                                                ACKER-ACK-STREAM-ID
                                                [id]
                                                ))
                           (:failed curr) ;;:failed为true, 表示该tuple失败
                           (do
                             (.remove pending id) ;;从pending中删除track entry, 并向相应的spout-task发送fail消息 
                             (acker-emit-direct output-collector
                                                (:spout-task curr)
                                                ACKER-FAIL-STREAM-ID
                                                [id]
                                                ))
                           ))
                   (.ack output-collector tuple) ;;acker bolt也是bolt, 所以最后完成对该ack tuple的ack
                   ))))
      (^void cleanup [this]
        )
      )))

 

(defn- update-ack [curr-entry val]
  (let [old (get curr-entry :val 0)] ;;取出entry中的value值,默认设为0
    (assoc curr-entry :val (bit-xor old val)) ;;将old和新val异或, 赋给entry的value
    ))

posted on 2013-08-06 10:34  fxjwind  阅读(1435)  评论(5编辑  收藏  举报