twitter storm源码走读之2 -- tuple消息发送场景分析

欢迎转载，转载请注明出处源自徽沪一郎。本文尝试分析tuple发送时的具体细节，本博的另一篇文章《bolt消息传递路径之源码解读》主要从消息接收方面来阐述问题，两篇文章互为补充。

worker进程内消息接收与处理全景图

先上幅图简要勾勒出worker进程接收到tuple消息之后的处理全过程

IConnection的建立与使用

话说在mk-threads :bolt函数的实现中有这么一段代码,其主要功能是实现tuple的emit功能

bolt-emit (fn [stream anchors values task]
          (let [out-tasks (if task
                  (tasks-fn task stream values)
                (tasks-fn stream values))]
        (fast-list-iter [t out-tasks]
                (let [anchors-to-ids (HashMap.)]
                  (fast-list-iter [^TupleImpl a anchors]
                          (let [root-ids (-> a .getMessageId .getAnchorsToIds .keySet)]
                            (when (pos? (count root-ids))
                              (let [edge-id (MessageId/generateId rand)]
                            (.updateAckVal a edge-id)
                            (fast-list-iter [root-id root-ids]
                                    (put-xor! anchors-to-ids root-id edge-id))
                            ))))
                  (transfer-fn t
                           (TupleImpl. worker-context
                               values
                               task-id
                               stream
                               (MessageId/makeId anchors-to-ids)))))
        (or out-tasks [])))

 加亮为蓝色的部分实现的功能是另外发送tuple,那么transfer-fn函数的定义在哪呢？见mk-threads的let部分，能见到下述一行代码

:transfer-fn (mk-executor-transfer-fn batch-transfer->worker)

 

在继续往下看每个函数实现之前，先确定一下这节代码阅读的目的。storm在线程之间使用disruptor进行通讯，在进程之间进行消息通讯使用的是zeromq或netty, 所以需要从transfer-fn追踪到使用zeromq或netty api的位置。

再看mk-executor-transfer-fn函数实现

(defn mk-executor-transfer-fn [batch-transfer->worker]
  (fn this
      ([task tuple block? ^List overflow-buffer]
       (if (and overflow-buffer (not (.isEmpty overflow-buffer)))
       (.add overflow-buffer [task tuple])
     (try-cause
      (disruptor/publish batch-transfer->worker [task tuple] block?)
      (catch InsufficientCapacityException e
         (if overflow-buffer
             (.add overflow-buffer [task tuple])
           (throw e))
         ))))
      ([task tuple overflow-buffer]
       (this task tuple (nil? overflow-buffer) overflow-buffer))
      ([task tuple]
       (this task tuple nil)
       )))

 

disruptor/publish表示将消息从本线程发送出去，至于谁是该消息的接收者，请继续往下看。

worker进程中，有一个receiver-thread是用来专门接收来自外部进程的消息，那么与之相对的是有一个transfer-thread用来将本进程的消息发送给外部进程。所以刚才的disruptor/publish发送出来的消息应该被transfer-thread接收到。

在transfer-thread中，能找到这行下述一行代码

transfer-thread (disruptor/consume-loop* (:transfer-queue worker) transfer-tuples)

 对于接收到来自本进程中其它线程发送过来的消息利用transfer-tuples进行处理，transfer-tuples使用mk-transfer-tuples-handler来创建，所以需要看看mk-transfer-tuples-handler能否与zeromq或netty联系上呢？

(defn mk-transfer-tuples-handler [worker]
  (let [^DisruptorQueue transfer-queue (:transfer-queue worker)
            drainer (ArrayList.)
            node+port->socket (:cached-node+port->socket worker)
            task->node+port (:cached-task->node+port worker)
            endpoint-socket-lock (:endpoint-socket-lock worker)
            ]
    (disruptor/clojure-handler
     (fn [packets _ batch-end?]
     (.addAll drainer packets)
     (when batch-end?
       (read-locked endpoint-socket-lock
            (let [node+port->socket @node+port->socket
                        task->node+port @task->node+port]
              ;; consider doing some automatic batching here (would need to not be serialized at this point to remo
              ;; try using multipart messages ... first sort the tuples by the target node (without changing the lo
              17
              (fast-list-iter [[task ser-tuple] drainer]
                      ;; TODO: consider write a batch of tuples here to every target worker
                      ;; group by node+port, do multipart send
                      (let [node-port (get task->node+port task)]
                        (when node-port
                          (.send ^IConnection (get node+port->socket node-port) task ser-tuple))
                        ))))
       (.clear drainer))))))

 

 

上述代码中出现了与zeromq可能有联系的部分了即加亮为红色的一行。

那凭什么说加亮的IConnection一行与zeromq有关系的，这话得慢慢说起，需要从配置文件开始。

在storm.yaml中有这么一行配置项，即

storm.messaging.transport: "backtype.storm.messaging.zmq"

这个配置项与worker中的mqcontext相对应,所以在worker中以mqcontext为线索,就能够一步步找到IConnection的实现。connections在函数mk-refresh-connections中建立

refresh-connections (mk-refresh-connections worker)

mk-refresh-connection函数中与mq-context相关联的一部分代码如下所示

(swap! (:cached-node+port->socket worker)
       #(HashMap. (merge (into {} %1) %2))
       (into {}
         (dofor [endpoint-str new-connections
                  :let [[node port] (string->endpoint endpoint-str)]]
             [endpoint-str
               (.connect
            ^IContext (:mq-context worker)
            storm-id
            ((:node->host assignment) node)
            port)
               ]
            )))

 

注意加亮部分，利用mq-conext中connect函数来创建IConnection. 当打开zmq.clj时候，就能验证我们的猜测。

(^IConnection connect [this ^String storm-id ^String host ^int port]
          (require 'backtype.storm.messaging.zmq)
          (-> context
          (mq/socket mq/push)
          (mq/set-hwm hwm)
          (mq/set-linger linger-ms)
          (mq/connect (get-connect-zmq-url local? host port))
          mk-connection))

 代码走到这里，IConnection什么时候建立起来的谜底就揭开了，消息是如何从bolt或spout线程传递到transfer-thread，再由zeromq将tuple发送给下跳的路径打通了。

tuple的分发策略 grouping

从一个bolt中产生的tuple可以有多个bolt接收,到底发送给哪一个bolt呢?这牵扯到分发策略问题,其实在twitter storm中有两个层面的分发策略问题,一个是对于task level的,在讲topology submit的时候已经涉及到。另一个就是现在要讨论的针对tuple level的分发。

再次将视线拉回到bolt-emit中，这次将目光集中在变量t的前前后后。

(let [out-tasks (if task
		    (tasks-fn task stream values)
		  (tasks-fn stream values))]
  (fast-list-iter [t out-tasks]
		  (let [anchors-to-ids (HashMap.)]
		    (fast-list-iter [^TupleImpl a anchors]
				    (let [root-ids (-> a .getMessageId .getAnchorsToIds .keySet)]
				      (when (pos? (count root-ids))
					(let [edge-id (MessageId/generateId rand)]
					  (.updateAckVal a edge-id)
					  (fast-list-iter [root-id root-ids]
							  (put-xor! anchors-to-ids root-id edge-id))
					  ))))
 (transfer-fn t
				 (TupleImpl. worker-context
					     values
					     task-id
					     stream
					     (MessageId/makeId anchors-to-ids)))))

 上述代码显示t从out-tasks来，而out-tasks是tasks-fn的返回值

tasks-fn (:tasks-fn task-data)

 

一谈tasks-fn，原来从未涉及的文件task.clj这次被挂上了，task-data与由task/mk-task创建。将中间环节跳过，调用关系如下所列。

mk-task
mk-task-data
mk-tasks-fn

 

tasks-fn中会使用到grouping,处理代码如下

fn ([^Integer out-task-id ^String stream ^List values]
          (when debug?
            (log-message "Emitting direct: " out-task-id "; " component-id " " stream " " values))
          (let [target-component (.getComponentId worker-context out-task-id)
                component->grouping (get stream->component->grouper stream)
                grouping (get component->grouping target-component)
                out-task-id (if grouping out-task-id)]
            (when (and (not-nil? grouping) (not= :direct grouping))
              (throw (IllegalArgumentException. "Cannot emitDirect to a task expecting a regular grouping")))                          
            (apply-hooks user-context .emit (EmitInfo. values stream task-id [out-task-id]))
            (when (emit-sampler)
              (builtin-metrics/emitted-tuple! (:builtin-metrics task-data) executor-stats stream)
              (stats/emitted-tuple! executor-stats stream)
              (if out-task-id
                (stats/transferred-tuples! executor-stats stream 1)
                (builtin-metrics/transferred-tuple! (:builtin-metrics task-data) executor-stats stream 1)))
            (if out-task-id [out-task-id])
            ))

 而每个topology中的grouping策略又是如何被executor知道的呢，这从另一端executor-data说起。

在mk-executor-data中有下面一行代码 

:stream->component->grouper (outbound-components worker-context component-id)

outbound-components的定义如下

(defn outbound-components
  "Returns map of stream id to component id to grouper"
  [^WorkerTopologyContext worker-context component-id]
  (->> (.getTargets worker-context component-id)
       clojurify-structure
       (map (fn [[stream-id component->grouping]]
        [stream-id
         (outbound-groupings
          worker-context
          component-id
          stream-id
          (.getComponentOutputFields worker-context component-id stream-id)
          component->grouping)]))
       (into {})
       (HashMap.)))