supervisor启动worker源码分析-worker.clj

supervisor通过调用sync-processes函数来启动worker,关于sync-processes函数的详细分析请参见"storm启动supervisor源码分析-supervisor.clj"。sync-processes函数代码片段如下:

sync-processes函数代码片段

;; sync-processes函数用于管理workers, 比如处理不正常的worker或dead worker, 并创建新的workers
;; supervisor标识supervisor的元数据
(defn sync-processes [supervisor]
                .
            .
            .
              ;; 忽略了部分代码
              .
              .
              .
       (wait-for-workers-launch
            conf
            (dofor [[port assignment] reassign-executors]
              (let [id (new-worker-ids port)]
                (log-message "Launching worker with assignment "
                             (pr-str assignment)
                             " for this supervisor "
                             (:supervisor-id supervisor)
                             " on port "
                             port
                             " with id "
                             id
                             )
                ;; launch-worker函数负责启动worker              
                (launch-worker supervisor
                               (:storm-id assignment)
                               port
                               id)
                id)))
  ))

sync-processes函数调用launch-worker函数启动worker,launch-worker函数是一个"多重函数",定义如下:
宏defmulti和defmethod经常被用在一起来定义multimethod-"多重函数"。宏defmulti的参数包括一个方法名以及一个dispatch函数,这个dispatch函数的返回值会被用来选择到底调用哪个重载的函数。宏defmethod的参数则包括方法名,dispatch的值,参数列表以及方法体。一个特殊的dispatch值:default 是用来表示默认情况的—即如果其它的dispatch值都不匹配的话,那么就调用这个方法。defmethod定义名字相同的方法,它们的参数个数必须一样。传给multimethod的参数会传给dipatch函数。实现类似java的重载

launch-worker函数

(defmulti launch-worker (fn [supervisor & _] (cluster-mode (:conf supervisor))))

 

;; 如果dispatch函数的返回值为关键字:distributed,即storm集群运行在分布式模式下,则执行该方法
(defmethod launch-worker
   ;; supervisor标识supervisor的元数据,storm-id标识该worker所属的topology,port标识该worker占用的端口号,worker-id是一个32位的uuid,用于标识worker
   :distributed [supervisor storm-id port worker-id]
   ;; conf绑定集群配置信息
   (let [conf (:conf supervisor)
         ;; storm-home绑定storm本地安装路径
         storm-home (System/getProperty "storm.home")
         ;; storm-log-dir绑定日志路径
         storm-log-dir (or (System/getProperty "storm.log.dir") (str storm-home "/logs"))
         ;; stormroot绑定supervisor本地路径"{storm.local.dir}/supervisor/stormdist/{storm-id}"
         stormroot (supervisor-stormdist-root conf storm-id)
         ;; jlp绑定运行时所依赖的本地库的路径,jlp函数生成本地库路径,参见jlp函数定义部分
         jlp (jlp stormroot conf)
         ;; stormjar绑定stormjar.jar文件的路径"{storm.local.dir}/supervisor/stormdist/{storm-id}/stormjar.jar"
         stormjar (supervisor-stormjar-path stormroot)
         ;; storm-conf绑定集群配置信息和storm-id配置信息的并集
         storm-conf (read-supervisor-storm-conf conf storm-id)
         ;; topo-classpath绑定storm-id的classpath集合
         topo-classpath (if-let [cp (storm-conf TOPOLOGY-CLASSPATH)]
                          [cp]
                          [])
         ;; 将stormjar和topo-classpath所标识的路径添加到Java的classpath中                
         classpath (-> (current-classpath)
                       (add-to-classpath [stormjar])
                       (add-to-classpath topo-classpath))
         ;; 从集群配置信息中获取默认情况下supervisor启动worker的jvm参数              
         worker-childopts (when-let [s (conf WORKER-CHILDOPTS)]
                            (substitute-childopts s worker-id storm-id port))
         ;; 从topology的配置信息中获取为该topology的worker指定的jvm参数                
         topo-worker-childopts (when-let [s (storm-conf TOPOLOGY-WORKER-CHILDOPTS)]
                                 (substitute-childopts s worker-id storm-id port))
         ;; 将该topology特有的依赖库路径合并到jlp中,这样topology-worker-environment绑定的map中就包含了启动该topology的worker所需的所有的依赖库                                  
         topology-worker-environment (if-let [env (storm-conf TOPOLOGY-ENVIRONMENT)]
                                       (merge env {"LD_LIBRARY_PATH" jlp})
                                       {"LD_LIBRARY_PATH" jlp})
         ;; 生成该worker的日志文件worker-{port}.log                              
         logfilename (str "worker-" port ".log")
         ;; command绑定一个Java -server xxxxxx -cp classpath classname arg_0 arg_1 ... arg_n命令,xxxxxx表示传递给java命令的jvm参数
         command (concat
                   [(java-cmd) "-server"]
                   worker-childopts
                   topo-worker-childopts
                   [(str "-Djava.library.path=" jlp)
                    (str "-Dlogfile.name=" logfilename)
                    (str "-Dstorm.home=" storm-home)
                    (str "-Dstorm.log.dir=" storm-log-dir)
                    (str "-Dlogback.configurationFile=" storm-home "/logback/cluster.xml")
                    (str "-Dstorm.id=" storm-id)
                    (str "-Dworker.id=" worker-id)
                    (str "-Dworker.port=" port)
                    "-cp" classpath
                    "backtype.storm.daemon.worker"
                    storm-id
                    (:assignment-id supervisor)
                    port
                    worker-id])
         ;; 去掉command命令数组中的空值
         command (->> command (map str) (filter (complement empty?)))
         ;; 获取command命令数组的字符串形式
         shell-cmd (->> command
                        (map #(str \' (clojure.string/escape % {\' "\\'"}) \'))
                        (clojure.string/join " "))]
     (log-message "Launching worker with command: " shell-cmd)
     ;; 通过ProcessBuilder类来执行command命令,即执行java命令运行backtype.storm.daemon.worker类的main方法创建一个新的进程,传递给main方法的参数为storm-id,supervisor-id,port和worker-id
     ;; 关于backtype.storm.daemon.worker类的main方法请参见其定义部分
     (launch-process command :environment topology-worker-environment)
     ))

;; 如果dispatch函数的返回值为关键字:local,即storm集群运行在本地模式下,则执行该方法      
(defmethod launch-worker
   :local [supervisor storm-id port worker-id]
   (let [conf (:conf supervisor)
         pid (uuid)
         worker (worker/mk-worker conf
                                  (:shared-context supervisor)
                                  storm-id
                                  (:assignment-id supervisor)
                                  port
                                  worker-id)]
     (psim/register-process pid worker)
     (swap! (:worker-thread-pids-atom supervisor) assoc worker-id pid)
     ))

jlp函数定义如下:

;; stormroot绑定supervisor本地路径"{storm.local.dir}/supervisor/stormdist/{storm-id}",conf绑定集群配置
(defn jlp [stormroot conf]
 ;; resource-root绑定supervisor本地路径"{storm.local.dir}/supervisor/stormdist/{storm-id}/resources"
 (let [resource-root (str stormroot File/separator RESOURCES-SUBDIR)
         ;; os绑定supervisor服务器的操作系统名
       os (clojure.string/replace (System/getProperty "os.name") #"\s+" "_")
       ;; arch绑定操作系统的架构,如"x86"和"i386"
       arch (System/getProperty "os.arch")
       ;; arch-resource-root绑定路径"{storm.local.dir}/supervisor/stormdist/{storm-id}/resources/{os}-{arch}"
       arch-resource-root (str resource-root File/separator os "-" arch)]
   ;; 返回"{storm.local.dir}/supervisor/stormdist/{storm-id}/resources/{os}-{arch}:{storm.local.dir}/supervisor/stormdist/{storm-id}/resources:{java.library.path}"
   (str arch-resource-root File/pathSeparator resource-root File/pathSeparator (conf JAVA-LIBRARY-PATH))))

read-supervisor-storm-conf函数定义如下:

;; 从supervisor本地路径"{storm.local.dir}/supervisor/stormdist/stormconf.ser"读取topology运行配置信息
(defn read-supervisor-storm-conf
 [conf storm-id]
 ;; stormroot绑定目录路径"{storm.local.dir}/supervisor/stormdist"
 (let [stormroot (supervisor-stormdist-root conf storm-id)
       ;; conf-path绑定文件路径"{storm.local.dir}/supervisor/stormdist/stormconf.ser"
       conf-path (supervisor-stormconf-path stormroot)
       ;; topology-path绑定文件路径"{storm.local.dir}/supervisor/stormdist/stormcode.ser"
       topology-path (supervisor-stormcode-path stormroot)]
   ;; 返回集群配置信息和topology配置信息合并后的配置信息map
   (merge conf (Utils/deserialize (FileUtils/readFileToByteArray (File. conf-path))))
   ))

backtype.storm.daemon.worker类定义在worker.clj文件中,通过:gen-class生成一个lava类,其main方法如下:

(defn -main [storm-id assignment-id port-str worker-id]  
 ;; 读取storm集群配置信息
 (let [conf (read-storm-config)]
   ;; 验证配置信息
   (validate-distributed-mode! conf)
   ;; 调用mk-worker函数,mk-worker函数请参见其定义部分
   (mk-worker conf nil storm-id assignment-id (Integer/parseInt port-str) worker-id)))

mk-worker函数:

;; conf绑定集群配置信息,shared-mq-context绑定共享mq,storm-id标识topology-id,assignment-id标识supervisor-id
(defserverfn mk-worker [conf shared-mq-context storm-id assignment-id port worker-id]
 (log-message "Launching worker for " storm-id " on " assignment-id ":" port " with id " worker-id
              " and conf " conf)
 ;; 如果storm不是"本地模式"运行(即"分布式模式"运行),则将标准输入输出流重定向到slf4j
 (if-not (local-mode? conf)
   (redirect-stdio-to-slf4j!))
 ;; because in local mode, its not a separate
 ;; process. supervisor will register it in this case
 ;; 如果storm是"分布式模式"运行,则在supervisor服务器本地创建文件"{storm.local.dir}/workers/{worker-id}/pids/{process-pid}",process-pid函数主要功能就是获取jvm进程的id
 ;; 需要特别注意的是worker-id是我们人为分配给该进程的一个标识,创建进程时,我们无法指定一个jvm进程的id,进程id是由操作系统分配的,所以我们需要获取该进程的实际id,并将我们指定的worker-id与进程id进行关联
 (when (= :distributed (cluster-mode conf))
   (touch (worker-pid-path conf worker-id (process-pid))))
 ;; worker绑定该进程的"元数据",worker-data函数的主要功能就是生成进程的"元数据",worker-data函数请参见其定义部分
 (let [worker (worker-data conf shared-mq-context storm-id assignment-id port worker-id)
             ;; heartbeat-fn绑定一个匿名函数,该匿名函数的功能就是生成worker"本地心跳信息",这里相当定义了heartbeat-fn函数,do-heartbeat函数请参见其定义部分
       heartbeat-fn #(do-heartbeat worker)

       ;; do this here so that the worker process dies if this fails
       ;; it's important that worker heartbeat to supervisor ASAP when launching so that the supervisor knows it's running (and can move on)
       ;; 调用heartbeat-fn函数将worker进程心跳信息保存到本地LocalState对象中
       _ (heartbeat-fn)
        ;; 定义一个原子类型的引用executors
       executors (atom nil)
       ;; launch heartbeat threads immediately so that slow-loading tasks don't cause the worker to timeout
       ;; to the supervisor
       ;; 将heartbeat-fn函数添加到定时器heartbeat-timer中,延迟执行时间为0s,每隔WORKER-HEARTBEAT-FREQUENCY-SECS执行一次
       _ (schedule-recurring (:heartbeat-timer worker) 0 (conf WORKER-HEARTBEAT-FREQUENCY-SECS) heartbeat-fn)
       ;; 将#(do-executor-heartbeats worker :executors @executors)函数添加到定时器executor-heartbeat-timer中,延迟执行时间为0s,每隔TASK-HEARTBEAT-FREQUENCY-SECS执行一次
       ;; 这样就可以将worker进程心跳信息同步到zookeeper中, 以便nimbus可以立刻知道worker进程已经启动,do-executor-heartbeats函数请参见其定义部分
       _ (schedule-recurring (:executor-heartbeat-timer worker) 0 (conf TASK-HEARTBEAT-FREQUENCY-SECS) #(do-executor-heartbeats worker :executors @executors))

       ;; 更新发送connections,mk-refresh-connections函数请参见其定义部分
       refresh-connections (mk-refresh-connections worker)
         ;; 主动调用refresh-connections函数refresh该worker进程所拥有的connections,并且不向zookeeper注册回调函数
       _ (refresh-connections nil)
         ;; 调用refresh-storm-active函数refresh该worker进程缓存的所属topology的活跃状态,refresh-storm-active函数请其参见定义部分
       _ (refresh-storm-active worker nil)
         ;; 调用mk-executor函数生成executor对象,保存到executors集合中。关于executor对象的创建将会在以后文章中具体分析
       _ (reset! executors (dofor [e (:executors worker)] (executor/mk-executor worker e)))
       ;; 启动worker进程专有的接收线程,将数据从worker进程的侦听端口,不停的放到task对应的接收队列,receive-thread-shutdown绑定该接收线程的关闭函数。launch-receive-thread函数请参见其定义部分
       receive-thread-shutdown (launch-receive-thread worker)
       
       ;; 定义event handler来处理transfer queue里面的数据。关于消息处理的流程会在以后文章中具体分析
       transfer-tuples (mk-transfer-tuples-handler worker)
       
       ;; 创建transfer-thread。关于消息处理的流程会在以后文章中具体分析
       transfer-thread (disruptor/consume-loop* (:transfer-queue worker) transfer-tuples)
       ;; 定义worker进程关闭回调函数,当关闭worker进程时调用该函数释放worker进程所占有的资源
       shutdown* (fn []
                   (log-message "Shutting down worker " storm-id " " assignment-id " " port)
                   ;; 关闭该worker进程到其他worker进程的连接
                   (doseq [[_ socket] @(:cached-node+port->socket worker)]
                     ;; this will do best effort flushing since the linger period
                     ;; was set on creation
                     (.close socket))
                   (log-message "Shutting down receive thread")
                   ;; 调用receive-thread-shutdown函数关闭该worker进程的接收线程
                   (receive-thread-shutdown)
                   (log-message "Shut down receive thread")
                   (log-message "Terminating messaging context")
                   (log-message "Shutting down executors")
                   ;; 关闭该worker进程所拥有的executor
                   (doseq [executor @executors] (.shutdown executor))
                   (log-message "Shut down executors")
                                       
                   ;;this is fine because the only time this is shared is when it's a local context,
                   ;;in which case it's a noop
                   ;; 关闭该worker进程所拥有的backtype.storm.messaging.netty.Context实例
                   (.term ^IContext (:mq-context worker))
                   (log-message "Shutting down transfer thread")
                   ;; 关闭transfer-queue
                   (disruptor/halt-with-interrupt! (:transfer-queue worker))
                                       ;; 中断transfer-thread
                   (.interrupt transfer-thread)
                   ;; 等待transfer-thread结束
                   (.join transfer-thread)
                   (log-message "Shut down transfer thread")
                   ;; 调用cancel-timer函数中断heartbeat-timer定时器线程
                   (cancel-timer (:heartbeat-timer worker))
                   ;; 调用cancel-timer函数中断refresh-connections-timer定时器线程
                   (cancel-timer (:refresh-connections-timer worker))
                   ;; 调用cancel-timer函数中断refresh-active-timer定时器线程
                   (cancel-timer (:refresh-active-timer worker))
                   ;; 调用cancel-timer函数中断executor-heartbeat-timer定时器线程
                   (cancel-timer (:executor-heartbeat-timer worker))
                   ;; 调用cancel-timer函数中断user-timer定时器线程
                   (cancel-timer (:user-timer worker))
                   
                   ;; 关闭该worker进程所拥有的线程池
                   (close-resources worker)
                   
                   ;; TODO: here need to invoke the "shutdown" method of WorkerHook
                   
                   ;; 调用StormClusterState实例的remove-worker-heartbeat!函数从zookeeper上删除worker心跳信息
                   (.remove-worker-heartbeat! (:storm-cluster-state worker) storm-id assignment-id port)
                   (log-message "Disconnecting from storm cluster state context")
                   ;; 关闭zookeeper连接
                   (.disconnect (:storm-cluster-state worker))
                   (.close (:cluster-state worker))
                   (log-message "Shut down worker " storm-id " " assignment-id " " port))
       ;; ret实现了Shutdownable和DaemonCommon协议
       ret (reify
            Shutdownable
            (shutdown
             [this]
             (shutdown*))
            DaemonCommon
            (waiting? [this]
              (and
                (timer-waiting? (:heartbeat-timer worker))
                (timer-waiting? (:refresh-connections-timer worker))
                (timer-waiting? (:refresh-active-timer worker))
                (timer-waiting? (:executor-heartbeat-timer worker))
                (timer-waiting? (:user-timer worker))
                ))
            )]
   
   ;; 将refresh-connections函数添加到定时器refresh-connections-timer中,每隔TASK-REFRESH-POLL-SECS执行一次。refresh-connections函数的无参版本提供一个默认回调函数调用其有参版本来更新所属           worker进程所拥有的collections,默认回调函数就是再次将refresh-connections函数无参版本添加到定时器refresh-connections-timer中
   ;; 这样只要zookeeper上分配信息发生变化,refresh-connections函数的有参版本就会执行,这里之所以周期执行refresh-connections函数是以防zookeeper的"watcher机制"失效
   (schedule-recurring (:refresh-connections-timer worker) 0 (conf TASK-REFRESH-POLL-SECS) refresh-connections)
   ;; 将函数(partial refresh-storm-active worker)添加到定时器refresh-active-timer中,每隔TASK-REFRESH-POLL-SECS执行一次。refresh-storm-active函数的执行逻辑与refresh-connections函数完全相      同
   (schedule-recurring (:refresh-active-timer worker) 0 (conf TASK-REFRESH-POLL-SECS) (partial refresh-storm-active worker))

   (log-message "Worker has topology config " (:storm-conf worker))
   (log-message "Worker " worker-id " for storm " storm-id " on " assignment-id ":" port " has finished loading")
   ;; 返回实现了Shutdownable协议和DaemonCommon协议的实例ret,通过ret我们可以关闭worker进程
   ret
   ))

worker-data函数:

;; worker-data函数生成进程的"元数据"
(defn worker-data [conf mq-context storm-id assignment-id port worker-id]
 ;; 为该进程生成ClusterState实例
 (let [cluster-state (cluster/mk-distributed-cluster-state conf)
       ;; 为该进程生成StormClusterState实例,这样进程就可以通过StormClusterState与zookeeper进行交互了
       storm-cluster-state (cluster/mk-storm-cluster-state cluster-state)
       ;; 调用read-supervisor-storm-conf函数读取storm-id的配置信息,read-supervisor-storm-conf函数请参见其定义部分
       storm-conf (read-supervisor-storm-conf conf storm-id)
       ;; executors绑定分配给该进程的executor的id集合,包含system executor的id
       executors (set (read-worker-executors storm-conf storm-cluster-state storm-id assignment-id port))
       ;; 进程内executor间通信是通过disruptor实现的,所以这里为该worker创建了一个名为"worker-transfer-queue"的disruptor queue,关于disruptor的内容会在以后详细介绍
       ;; 注意transfer-queue是worker相关的,与executor无关
       transfer-queue (disruptor/disruptor-queue "worker-transfer-queue" (storm-conf TOPOLOGY-TRANSFER-BUFFER-SIZE)
                                                 :wait-strategy (storm-conf TOPOLOGY-DISRUPTOR-WAIT-STRATEGY))
       ;; mk-receive-queue-map函数为每个executor创建一个名为"receive-queue{executor-id}"的disruptor queue,executor-receive-queue-map绑定executor-id->"disruptor接收queue"的map  
       ;; 注意executor-receive-queue-map是executor相关,与worker无关                                    
       executor-receive-queue-map (mk-receive-queue-map storm-conf executors)
       ;; executor可能有多个tasks,相同executor的tasks共用一个"disruptor接收queue",将executor-id->"disruptor接收queue"的map转化为task-id->"disruptor接收queue"的map,
       ;; 如executor-receive-queue-map={[1 2] receive-queue[1 2], [3 4] receive-queue[3 4]},那么receive-queue-map={1 receive-queue[1 2], 2 receive-queue[1 2], 3 receive-queue[3             4], 4 receive-queue[3 4]}
       receive-queue-map (->> executor-receive-queue-map
                              (mapcat (fn [[e queue]] (for [t (executor-id->tasks e)] [t queue])))
                              (into {}))
               ;; 调用read-supervisor-topology函数从supervisor本地路径"{storm.local.dir}/supervisor/stormdist/stormcode.ser"读取topology对象的序列化文件
       topology (read-supervisor-topology conf storm-id)]
   ;; recursive-map宏会将下面value都执行一遍,用返回值和key生成新的map作为worker的"元数据",recursive-map宏见其定义部分
   (recursive-map
     ;; 保存集群配置信息
     :conf conf
     ;; 保存一个传输层实例用于worker进程间消息传递,storm传输层被定义成了"可插拔式"插件,通过实现backtype.storm.messaging.IContext接口就可以定义自己的消息传输层。storm 0.8.x默认传输层实例是             backtype.storm.messaging.zmq,但是由于
     ;; 1.ZeroMQ是一个本地化的消息库,它过度依赖操作系统环境,而且ZeroMQ使用的是"堆外内存",无法使用jvm相关的内存监控工具进行监控管理,存在"堆外内存"泄漏风险
     ;; 2.安装起来比较麻烦
     ;; 3.ZeroMQ的稳定性在不同版本之间差异巨大,并且目前只有2.1.7版本的ZeroMQ能与Storm协调的工作。
     ;; 所以storm 0.9之后默认传出层实例为backtype.storm.messaging.netty.Context,Netty有如下优点:
     ;; 1.平台隔离,Netty是一个纯Java实现的消息队列,可以帮助Storm实现更好的跨平台特性,同时基于JVM的实现可以让我们对消息有更好的控制,因为Netty使用jvm的堆内存,而不是堆外内存
     ;; 2.高性能,Netty的性能要比ZeroMQ快两倍左右
     ;; 3. 安全性认证,使得我们将来要做的worker进程之间的认证授权机制成为可能。
     :mq-context (if mq-context
                     mq-context
                     (TransportFactory/makeContext storm-conf))
     ;; 记录所属storm-id
     :storm-id storm-id
     ;; 记录所属supervisor-id
     :assignment-id assignment-id
     ;; 记录端口
     :port port
     ;; 记录我们分配给该进程的worker-id
     :worker-id worker-id
     ;; 记录ClusterState实例
     :cluster-state cluster-state
     ;; 记录StormClusterState实例,以便worker进程与zookeeper进行交互
     :storm-cluster-state storm-cluster-state
     ;; 记录topology的当前活跃状态为false
     :storm-active-atom (atom false)
     ;; 记录分布在该worker进程上的executors的id
     :executors executors
     ;; 记录排序后的分布在该worker进程上的tasks的id
     :task-ids (->> receive-queue-map keys (map int) sort)
     ;; 记录该topology的配置信息
     :storm-conf storm-conf
     ;; 记录topology实例
     :topology topology
     ;; 记录添加了acker,system bolt,metric bolt后的topology实例
     :system-topology (system-topology! storm-conf topology)
     ;; 记录一个名为"heartbeat-timer"的定时器
     :heartbeat-timer (mk-halting-timer "heartbeat-timer")
     ;; 记录一个名为"refresh-connections-timer"的定时器
     :refresh-connections-timer (mk-halting-timer "refresh-connections-timer")
     ;; 记录一个名为"refresh-active-timer"的定时器
     :refresh-active-timer (mk-halting-timer "refresh-active-timer")
     ;; 记录一个名为"executor-heartbeat-timer"的定时器
     :executor-heartbeat-timer (mk-halting-timer "executor-heartbeat-timer")
     ;; 记录一个名为"user-timer"的定时器
     :user-timer (mk-halting-timer "user-timer")
     ;; 记录任务id->组件名称键值对的map,形如:{1 "boltA", 2 "boltA", 3 "boltA", 4 "boltA", 5 "boltB", 6 "boltB"},storm-task-info函数请参见其定义部分
     :task->component (HashMap. (storm-task-info topology storm-conf)) ; for optimized access when used in tasks later on
     ;; 记录"组件名称"->"stream_id->输出域Fields对象的map"的map,component->stream->fields函数请参见其定义部分
     :component->stream->fields (component->stream->fields (:system-topology <>))
     ;; 记录"组件名称"->排序后task-id集合的map,形如:{"boltA" [1 2 3 4], "boltB" [5 6]}
     :component->sorted-tasks (->> (:task->component <>) reverse-map (map-val sort))
     ;; 记录一个ReentrantReadWriteLock对象
     :endpoint-socket-lock (mk-rw-lock)
     ;; 记录一个node+port->socket的原子类型的map
     :cached-node+port->socket (atom {})
     ;; 记录一个task->node+port的原子类型的map
     :cached-task->node+port (atom {})
     ;; 记录该worker进程的传输队列transfer-queue
     :transfer-queue transfer-queue
     ;; 记录executor接收队列executor-receive-queue-map
     :executor-receive-queue-map executor-receive-queue-map
     ;; 记录executor中"开始任务id"->executor接收queue的map,如executor-receive-queue-map={[1 2] receive-queue[1 2], [3 4] receive-queue[3 4]},那么short-executor-receive-queue-map={1 receive-queue[1 2], 3 receive-queue[3 4]}
     :short-executor-receive-queue-map (map-key first executor-receive-queue-map)
     ;; 记录task_id->executor中"开始任务id"的map,如executors=#{[1 2] [3 4] [5 6]},task->short-executor={1 1, 2 1, 3 3, 4 3, 5 5, 6 5}
     :task->short-executor (->> executors
                                (mapcat (fn [e] (for [t (executor-id->tasks e)] [t (first e)])))
                                (into {})
                                (HashMap.))
     ;; 记录一个可以终止该worker进程的"自杀函数"
     :suicide-fn (mk-suicide-fn conf)
     ;; 记录一个可以计算该worker进程启动了多长时间的函数
     :uptime (uptime-computer)
     ;; 为该worker进程生成一个线程池
     :default-shared-resources (mk-default-resources <>)
     ;; mk-user-resources函数目前版本为空实现
     :user-shared-resources (mk-user-resources <>)
     ;; 记录一个函数,该函数的主要功能就是接收messages并将message发送到task对应的接收队列,mk-transfer-local-fn函数请参见其定义部分
     :transfer-local-fn (mk-transfer-local-fn <>)
     ;; 记录每个worker进程特有的接收线程的个数
     :receiver-thread-count (get storm-conf WORKER-RECEIVER-THREAD-COUNT)
     ;; 将executor处理过的message放到worker进程发送队列transfer-queue中,mk-transfer-fn函数请参见其定义部分
     :transfer-fn ( <>)
     )))

read-worker-executors函数:

;; read-worker-executors函数用于读取分布在该进程上的executor信息
(defn read-worker-executors [storm-conf storm-cluster-state storm-id assignment-id port]
 ;; assignment绑定executor->node+port的map,调用StormClusterState实例的assignment-info函数从zookeeper上读取storm-id的分配信息AssignmentInfo实例
 ;; AssignmentInfo定义如下:(defrecord Assignment [master-code-dir node->host executor->node+port executor->start-time-secs])
 (let [assignment (:executor->node+port (.assignment-info storm-cluster-state storm-id nil))]
   ;; 返回分配给该进程的executor的id集合,包含system executor的id
   (doall
    ;; 将system executor的id和topology executor的id合并
    (concat
     ;; system executor的id,[-1 -1]    
     [Constants/SYSTEM_EXECUTOR_ID]
     ;; 从分配信息assignment中获取分配给该进程的executor
     (mapcat (fn [[executor loc]]
               (if (= loc [assignment-id port])
                 [executor]
                 ))
             assignment)))))

mk-receive-queue-map函数:

;; mk-receive-queue-map函数为每个executor创建一个名为"receive-queue{executor-id}"的disruptor queue,如"receive-queue[1 3]",并返回executor-id->receive-queue的map
(defn- mk-receive-queue-map [storm-conf executors]
 ;; executors标识了executor-id集合
 (->> executors
      ;; TODO: this depends on the type of executor
      ;; 通过调用map函数为每个executor-id创建一个"disruptor接收queue"
      (map (fn [e] [e (disruptor/disruptor-queue (str "receive-queue" e)
                                                 (storm-conf TOPOLOGY-EXECUTOR-RECEIVE-BUFFER-SIZE)
                                                 :wait-strategy (storm-conf TOPOLOGY-DISRUPTOR-WAIT-STRATEGY))]))
      ;; 返回executor-id->receive-queue的map                                          
      (into {})
      ))

storm-task-info函数:

(defn storm-task-info
 "Returns map from task -> component id"
 [^StormTopology user-topology storm-conf]
 (->> (system-topology! storm-conf user-topology)
      ;; 获取组件名称->组件对象键值对的map
      all-components
      ;; 返回组件名称->组件任务数键值对的map,如{"boltA" 4, "boltB" 2}
      (map-val (comp #(get % TOPOLOGY-TASKS) component-conf))
      ;; 按照组件名称对map进行排序返回结果序列,如(["boltA" 4] ["boltB" 2])
      (sort-by first)
      ;; mapcat函数等价于对(map (fn...))的返回结果执行concat函数,返回("boltA" "boltA" "boltA" "boltA" "boltB" "boltB")
      (mapcat (fn [[c num-tasks]] (repeat num-tasks c)))
      ;; {1 "boltA", 2 "boltA",3 "boltA", 4 "boltA", 5 "boltB", 6 "boltB"}
      (map (fn [id comp] [id comp]) (iterate (comp int inc) (int 1)))
      (into {})
      ))

component->stream->fields函数:

(defn component->stream->fields [^StormTopology topology]
 ;; 调用ThriftTopologyUtils/getComponentIds方法获取topology所有组件名称集合,如#{"boltA", "boltB", "boltC"}
 (->> (ThriftTopologyUtils/getComponentIds topology)
          ;; 获取每个组件的stream_id->StreamInfo对象的map,stream->fields函数请参见其定义部分
      (map (fn [c] [c (stream->fields topology c)]))
      ;; 生成"组件名称"->"stream_id->输出域Fields对象的map"的map
      (into {})
      ;; 将其转化成Java的HashMap
      (HashMap.)))

stream->fields函数:

(defn- stream->fields [^StormTopology topology component]
 ;; 获取指定组件名的ComponentCommon对象
 (->> (ThriftTopologyUtils/getComponentCommon topology component)
        ;; 调用ComponentCommon对象的get_streams方法获取stream_id->StreamInfo对象的map,一个组件可以有多个输出流
      .get_streams
      ;; s绑定stream_id,info绑定StremInfo对象,调用StreamInfo对象的get_output_fields获取输出域List<String>对象,再用输出域List<String>对象生成Fields对象
      (map (fn [[s info]] [s (Fields. (.get_output_fields info))]))
      ;; 生成stream_id->Fields对象的map
      (into {})
      ;; 将clojure结构的map转换成java中的HashMap
      (HashMap.)))

mk-transfer-local-fn函数:

;; mk-transfer-local-fn函数返回一个匿名函数,该匿名函数的主要功能就是接收messages并将message发送到task对应的接收队列    
(defn mk-transfer-local-fn [worker]
 ;; short-executor-receive-queue-map绑定"开始任务id"->executor接收queue的map,如:{1 receive-queue[1 2], 3 receive-queue[3 4]}
 (let [short-executor-receive-queue-map (:short-executor-receive-queue-map worker)
       ;; task->short-executor绑定task_id->executor中"开始任务id"的map,如:{1 1, 2 1, 3 3, 4 3}
       task->short-executor (:task->short-executor worker)
       ;; task-getter绑定一个由comp生成的组合函数
       task-getter (comp #(get task->short-executor %) fast-first)]
   ;; 返回一个匿名函数,tuple-batch是一个ArrayList对象,ArrayList的每个元素都是一个长度为2的数组[task_id, message],task_id表示该消息由哪个task处理,message表示消息
   (fn [tuple-batch]
     ;; 调用fast-group-by函数获取"executor简写id"->需要该executor处理的消息List的map
     (let [grouped (fast-group-by task-getter tuple-batch)]
       ;; fast-map-iters宏主要用于遍历map,short-executor标识"executor简写id",pairs标识消息[task_id, message]
       (fast-map-iter [[short-executor pairs] grouped]
         ;; 获取该executor的接收queue
         (let [q (short-executor-receive-queue-map short-executor)]
           ;; 如果q不为空,则调用disruptor的publish方法将消息放入disruptor中
           (if q
             (disruptor/publish q pairs)
             (log-warn "Received invalid messages for unknown tasks. Dropping... ")
             )))))))

fast-group-by函数:

;; fast-group-by函数的主要功能就是生成"executor简写id"->需要该executor处理的消息List的map            
(defn fast-group-by
 ;; afn绑定mk-transfer-local-fn函数中定义的task-getter函数,alist绑定一个ArrayList对象,ArrayList的每个元素都是一个长度为2的数组[task_id, message],task_id表示该消息由哪个task处理,message表示消息
 [afn alist]
 ;; 创建一个HashMap对象ret
 (let [ret (HashMap.)]
   ;; fast-list-iter是一个宏,主要功能就是遍历list
   (fast-list-iter
     ;; e绑定每个[task_id, message]数组对象
     [e alist]
     ;; 调用afn绑定的task-getter函数获取该task_id所属的"executor的简写id",所以key绑定"executor简写id"
     (let [key (afn e)
              ;; 从ret中获取key所对应的ArrayList对象,即需要该executor处理的消息列表
           ^List curr (get-with-default ret key (ArrayList.))]
       ;; [task_id, message]数组对象添加到list中
       (.add curr e)))
   ;; 返回ret
   ret))

mk-transfer-fn函数:

;; mk-transfer-fn函数主要功能就是将executor处理过的message放到worker进程发送队列transfer-queue中
(defn mk-transfer-fn [worker]
 ;; local-tasks绑定分布在该worker进程上的task的id集合
 (let [local-tasks (-> worker :task-ids set)
       ;; local-transfer标识mk-transfer-local-fn返回的匿名函数
       local-transfer (:transfer-local-fn worker)
       ;; transfer-queue绑定该worker进程的传输队列transfer-queue
       ^DisruptorQueue transfer-queue (:transfer-queue worker)
       ;; task->node+port绑定task_id->node+port的map
       task->node+port (:cached-task->node+port worker)]
   ;; 返回一个匿名函数,serializer标识一个Kryo序列化器,tuple-batch是一个ArrayList对象,ArrayList的每个元素都是一个长度为2的数组[task_id, message],task_id表示该消息由哪个task处理,即message的目标task,message表示消息
   (fn [^KryoTupleSerializer serializer tuple-batch]
     ;; local为ArrayList
     (let [local (ArrayList.)
           ;; remoteMap为HashMap
           remoteMap (HashMap.)]
       ;; 遍历tuple-batch
       (fast-list-iter [[task tuple :as pair] tuple-batch]
         ;; 如果接收该消息的task为本地task,即该task也分布在该worker进程上,那么将该消息添加到local中
         (if (local-tasks task)
           (.add local pair)
           
           ;;Using java objects directly to avoid performance issues in java code
           ;; 否则说明接收该消息的task不是本地task,即该task分布在其他worker进程上;node+port标识了运行该task的worker进程所在的节点和端口
           (let [node+port (get @task->node+port task)]
             ;; 如果remoteMap不包含node+port,则添加
             (when (not (.get remoteMap node+port))
               (.put remoteMap node+port (ArrayList.)))
             (let [remote (.get remoteMap node+port)]
               ;; 首先用task_id和序列化后的tuple生成TaskMessage对象,然后将TaskMessage对象添加到ArrayList中
               (.add remote (TaskMessage. task (.serialize serializer tuple)))
                ))))
       ;; 调用local-transfer函数发送需要本地task处理的消息
       (local-transfer local)
       ;; 调用disruptor的publish方法将remoteMap放入worker进程的传输队列transfer-queue中,remoteMap的key为node+port,value为ArrayList,ArrayList中每个元素都是需要node+port所对应的worker进行处理
       (disruptor/publish transfer-queue remoteMap)
         ))))

do-heartbeat函数:

(defn do-heartbeat [worker]
 ;; 获取集群配置信息
 (let [conf (:conf worker)
       ;; 创建WorkerHeartbeat对象
       hb (WorkerHeartbeat.
            ;; 本次心跳时间
            (current-time-secs)
            ;; 该worker进程所属的topology-id
            (:storm-id worker)
            ;; 分布在该worker进程上的executor-id集合
            (:executors worker)
            ;; 该worker进程所占用的端口
            (:port worker))
       ;; 创建一个基于目录"{storm.local.dir}/workers/{worker-id}/heartbeats"的LocalState对象,用于存放worker进程的"本地心跳信息",通过LocalState对象我们可以访问一个序列化到磁盘的map对象
       state (worker-state conf (:worker-id worker))]
   (log-debug "Doing heartbeat " (pr-str hb))
   ;; do the local-file-system heartbeat.
   ;; 将worker进程心跳信息通过LocalState对象存入磁盘,map对象的key为"worker-heartbeat"字符串,value为worker心跳信息
   (.put state
       LS-WORKER-HEARTBEAT
       hb
       false
       )
   ;; 调用LocalState对象的clearup方法,只保留最近60次心跳信息
   (.cleanup state 60) ; this is just in case supervisor is down so that disk doesn't fill up.
                        ; it shouldn't take supervisor 120 seconds between listing dir and reading it

   ))

do-executor-heartbeats函数:

;; do-executor-heartbeats函数主要功能就是通过worker-heartbeat!函数将worker进程心跳信息写入zookeeper的workerbeats节点中
(defnk do-executor-heartbeats [worker :executors nil]
 ;; stats is how we know what executors are assigned to this worker
 ;; stats绑定executor对象->executor统计信息的map。当第一次调用do-executor-heartbeats函数时,即第一次心跳时,executors为nil,map形如:{executor_1 nil, executor_2 nil, ... }
 ;; 当再次心跳时,将会调用executor对象的get-executor-id函数和render-stats函数,获取executor_id->executor统计信息的map,所以stats绑定的map在第一次心跳时和再次心跳时是不同的,有关executor统计信  息的计算会在以后文章中具体分析。
 (let [stats (if-not executors
                 (into {} (map (fn [e] {e nil}) (:executors worker)))
                 (->> executors
                   (map (fn [e] {(executor/get-executor-id e) (executor/render-stats e)}))
                   (apply merge)))
       ;; 构建worker进程的心跳信息
       zk-hb {:storm-id (:storm-id worker)
              ;; 记录executor统计信息
              :executor-stats stats
              ;; 记录worker进程运行了多次时间
              :uptime ((:uptime worker))
              ;; 记录worker进程心跳时间
              :time-secs (current-time-secs)
              }]
   ;; do the zookeeper heartbeat
   ;; 调用StormClusterState对象的worker-heartbeat!函数将worker进程心跳信息zk-hb同步到zookeeper的"/workerbeats/{topology-id}/{supervisorId-port}/"节点中
   (.worker-heartbeat! (:storm-cluster-state worker) (:storm-id worker) (:assignment-id worker) (:port worker) zk-hb)    
   ))

mk-refresh-connections函数:

;; mk-refresh-connections函数返回一个名为this的函数,在"storm启动supervisor源码分析-supervisor.clj"中,我们在mk-synchronize-supervisor函数也见过这种定义函数的方式,是因为这个函数本身要在函数体内被使用。
;; 并且refresh-connections是需要反复被执行的,即当每次assignment-info发生变化的时候,就需要refresh一次,这里是通过zookeeper的"watcher机制"实现的
(defn mk-refresh-connections [worker]
 ;; outbound-tasks绑定用于接收该worker进程输出消息的所有任务,worker-outbound-tasks函数请参见其定义部分
 (let [outbound-tasks (worker-outbound-tasks worker)
       ;; conf绑定worker配置信息
       conf (:conf worker)
       ;; storm-cluster-state绑定StormClusterState实例
       storm-cluster-state (:storm-cluster-state worker)
       ;; storm-id标识该worker进程所属的topology的id
       storm-id (:storm-id worker)]
   ;; 返回名称为this的函数,每次assignment-info发生变化时,就执行一次来refresh该worker进程的connections
   (fn this
     ;; 无参版本,提供一个"默认回调函数"调用有参版本,"默认回调函数"就是将this函数无参版本本身添加到worker进程的refresh-connections-timer定时器中,这样当assignment-info发生变化时,zookeeper的"watcher机制"
     ;; 就会执行回调函数,refresh-connections-timer定时器线程将会执行this函数。这样就可以保证,每次assignment发生变化,定时器都会在后台做refresh-connections的操作
     ([]
       (this (fn [& ignored] (schedule (:refresh-connections-timer worker) 0 this))))
     ;; 有参版本
     ([callback]
        ;; 调用StormClusterState实例的assignment-version函数获取storm-id的当前分配信息版本,并将callback函数注册到zookeeper
        (let [version (.assignment-version storm-cluster-state storm-id callback)
              ;; 如果worker本地缓存的分配版本和zookeeper上获取的分配版本相等,那么说明storm-id的分配信息未发生变化,直接从worker本地获取分配信息
              assignment (if (= version (:version (get @(:assignment-versions worker) storm-id)))
                           (:data (get @(:assignment-versions worker) storm-id))
                           ;; 否则调用assignment-info-with-version函数从zookeeper的"/assignments/{storm-id}"节点重新获取带有版本号的分配信息,并注册回调函数,这样worker就能感知某个已存在的assignment是否被重新分配
                           (let [new-assignment (.assignment-info-with-version storm-cluster-state storm-id callback)]
                             ;; 将最近分配信息保存到worker本地缓存
                             (swap! (:assignment-versions worker) assoc storm-id new-assignment)
                             (:data new-assignment)))
             ;; my-assignment标识"接收该worker进程输出消息的任务"->[node port]的map
             my-assignment (-> assignment
                                                 ;; 获取executor_id->[node port]的map,如:{[1 1] [node1 port1], [4 4] [node1 port1], [2 2] [node2 port1], [5 5] [node2 port1], [3 3] [node3 port1], [6 6] [node3 port1]}
                               :executor->node+port
                               ;; 获取task_id->[node port]的map,如:{[1 [node1 port1], 4 [node1 port1], 2 [node2 port1], 5 [node2 port1], 3 [node3 port1], 6 [node3 port1]}
                               to-task->node+port
                               ;; 选择"键"包含在outbound-tasks集合的键值对,假设outbound-tasks=#{4 5 6},过滤后为{4 [node1 port1], 5 [node2 port1], 6 [node3 port1]}
                               (select-keys outbound-tasks)
                               ;; {4 "node1/port1", 5 "node2/port1", 6 "node3/port1"}
                               (#(map-val endpoint->string %)))
             ;; we dont need a connection for the local tasks anymore
             ;; 过滤掉分布在该worker进程上的task,因为分布在通一个进程上不需要建立socket连接。假设该worker进程位于node1的port1上,则needed-assignment={5 "node2/port1", 6 "node3/port1"}
             needed-assignment (->> my-assignment
                                     (filter-key (complement (-> worker :task-ids set))))
             ;; needed-connections绑定"需要的连接"的集合,needed-connections=#{"node2/port1", "node3/port1"}
             needed-connections (-> needed-assignment vals set)
             ;; needed-tasks绑定需要建立连接的任务集合,needed-tasks=#{5, 6}
             needed-tasks (-> needed-assignment keys)
             
             ;; current-connections绑定当前该worker进程"已建立的连接"的集合
             current-connections (set (keys @(:cached-node+port->socket worker)))
             ;; needed-connections和current-connections的差集表示需要"新建的连接"的集合,假设current-connections=#{},则new-connections=#{"node2/port1", "node3/port1"}
             new-connections (set/difference needed-connections current-connections)
             ;; current-connections和needed-connections的差集表示需要"删除的连接"的集合
             remove-connections (set/difference current-connections needed-connections)]
             ;; 将新建的连接合并到cached-node+port->socket中
             (swap! (:cached-node+port->socket worker)
                    #(HashMap. (merge (into {} %1) %2))
                    ;; 创建endpoint-str->connection对象的map,即建立新的连接。如:{"node2/port1" connect1, "node3/port1" connect2}
                    (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)
                         ]
                        )))
             ;; 将my-assignment保存到worker进程本地缓存cached-task->node+port中
             (write-locked (:endpoint-socket-lock worker)
               (reset! (:cached-task->node+port worker)
                       (HashMap. my-assignment)))
             ;; close需要"删除的连接"
             (doseq [endpoint remove-connections]
               (.close (get @(:cached-node+port->socket worker) endpoint)))
             ;; 将需要"删除的连接"从worker进程本地缓存cached-node+port->socket中删除,通过worker进程本地缓存cached-task->node+port和cached-node+port->socket,我们就可以或得task和socket的对应关系
             (apply swap!
                    (:cached-node+port->socket worker)
                    #(HashMap. (apply dissoc (into {} %1) %&))
                    remove-connections)
             ;; 查找出未建立连接的task
             (let [missing-tasks (->> needed-tasks
                                      (filter (complement my-assignment)))]
               ;; 如果存在未建立连接的task,则记录日志文件
               (when-not (empty? missing-tasks)
                 (log-warn "Missing assignment for following tasks: " (pr-str missing-tasks))
                 )))))))

worker-outbound-tasks函数:

;; worker-outbound-tasks函数主要功能就是获取接收来自该worker消息的组件的task-id集合                
(defn worker-outbound-tasks
 "Returns seq of task-ids that receive messages from this worker"
 [worker]
 ;; context绑定backtype.storm.task.WorkerTopologyContext对象,worker-context函数请参见其定义部分
 (let [context (worker-context worker)
       ;; 对分布在该worker进程上的每个任务的task_id调用匿名函数(fn [task-id] ... ),并对返回结果进行concat操作,components绑定了接收组件id的集合
       components (mapcat
                    (fn [task-id]
                      ;; 调用context的getComponentId方法获取该task-id所属的组件(spout/bolt)的名称
                      (->> (.getComponentId context (int task-id))
                           ;; 调用context的getTargets方法,获取哪些组件接收了componentId输出的消息
                           (.getTargets context)
                           vals
                           ;; 获取接收组件id的集合
                           (map keys)
                           (apply concat)))
                    ;; 获取分布在该worker进程上的task_id集合
                    (:task-ids worker))]
   (-> worker
       ;; 获取任务id->组件名称键值对的map,形如:{1 "boltA", 2 "boltA", 3 "boltA", 4 "boltA", 5 "boltB", 6 "boltB"}
       :task->component
       ;; 结果形如:{"boltA" [1 2 3 4], "boltB" [5 6]}
       reverse-map
       ;; 过滤出"键"包含在components集合中的键值对
       (select-keys components)
       vals
       flatten
       ;; 获取接收组件所有任务的id的集合
       set )))

worker-context函数:

(defn worker-context [worker]
 ;; 返回backtype.storm.task.WorkerTopologyContext对象
 (WorkerTopologyContext. (:system-topology worker)
                         (:storm-conf worker)
                         (:task->component worker)
                         (:component->sorted-tasks worker)
                         (:component->stream->fields worker)
                         (:storm-id worker)
                         (supervisor-storm-resources-path
                           (supervisor-stormdist-root (:conf worker) (:storm-id worker)))
                         (worker-pids-root (:conf worker) (:worker-id worker))
                         (:port worker)
                         (:task-ids worker)
                         (:default-shared-resources worker)
                         (:user-shared-resources worker)
                         ))

getTargets方法:

;; WorkerTopologyContext类继承GeneralTopologyContext类,getTargets方法是GeneralTopologyContext类实例方法,主要功能就是获取哪些组件接收了componentId输出的消息
;; 返回值为一个stream_id->{receive_component_id->Grouping}的map,receive_component_id就是接收组件的id              
public Map<String, Map<String, Grouping>> getTargets(String componentId) {
       ;; 创建返回结果map,ret
       Map<String, Map<String, Grouping>> ret = new HashMap<String, Map<String, Grouping>>();
       ;; 获取该topology的所有组件ids,并遍历
       for(String otherComponentId: getComponentIds()) {
           ;; 通过组件id获取组件的ComponentCommon对象,然后再获取其输入信息inputs
           Map<GlobalStreamId, Grouping> inputs = getComponentCommon(otherComponentId).get_inputs();
           ;; 遍历输入信息,GlobalStreamId对象有两个成员属性,一个是流id,一个是发送该流的组件id
           for(GlobalStreamId id: inputs.keySet()) {
               ;; 如果输入流的组件id和componentId相等,那么说明该组件接收来自componentId的输出,则将其添加到ret中
               if(id.get_componentId().equals(componentId)) {
                   Map<String, Grouping> curr = ret.get(id.get_streamId());
                   if(curr==null) curr = new HashMap<String, Grouping>();
                   curr.put(otherComponentId, inputs.get(id));
                   ret.put(id.get_streamId(), curr);
               }
           }
       }
       return ret;
   }

refresh-storm-active函数:

;; refresh-storm-active函数主要功能就是refresh指定worker进程缓存的所属topology的活跃状态                
(defn refresh-storm-active
 ;; "无回调函数"版本,使用默认回调函数调用"有回调函数"版本,默认回调函数将refresh-storm-active函数本身添加到refresh-active-timer定时器
 ([worker]
   (refresh-storm-active worker (fn [& ignored] (schedule (:refresh-active-timer worker) 0 (partial refresh-storm-active worker)))))
 ;; "有回调函数"版本
 ([worker callback]
   ;; 调用StormClusterState实例的storm-base函数,从zookeeper的"/storms/{storm-id}"节点获取该topology的StormBase数据,并将回调函数callback注册到zookeeper的"/storms/{storm-id}"节点
   ;; 这样当该节点数据发生变化时,callback函数将被执行,即将refresh-storm-active函数添加到refresh-active-timer定时器,refresh-active-timer定时器线程将会执行refresh-storm-active函数
   (let [base (.storm-base (:storm-cluster-state worker) (:storm-id worker) callback)]
    ;; 更新worker进程缓存的topology的活跃状态
    (reset!
     (:storm-active-atom worker)
     (= :active (-> base :status :type))
     ))
    ))

launch-receive-thread函数:

;; 为worker进程启动专有接收线程  
(defn launch-receive-thread [worker]
 (log-message "Launching receive-thread for " (:assignment-id worker) ":" (:port worker))
 ;; launch-receive-thread!函数请参见其定义部分
 (msg-loader/launch-receive-thread!
   ;; 连接实例,0.9版本开始默认使用netty,backtype.storm.messaging.netty.Context实例
   (:mq-context worker)
   (:storm-id worker)
   ;; 接收线程数
   (:receiver-thread-count worker)
   (:port worker)
   ;; 获取本地消息传输函数transfer-local-fn,transfer-local-fn函数将消息发送给分布在该worker进程上的task相应队列
   (:transfer-local-fn worker)
   ;; 获取worker进程输入队列大小
   (-> worker :storm-conf (get TOPOLOGY-RECEIVER-BUFFER-SIZE))
   :kill-fn (fn [t] (exit-process! 11))))

launch-receive-thread!函数:

;; launch-receive-thread!函数定义在loader.clj文件中,用于启动指定worker进程的接收线程  
(defnk launch-receive-thread!
 [context storm-id receiver-thread-count port transfer-local-fn max-buffer-size
  :daemon true
  :kill-fn (fn [t] (System/exit 1))
  :priority Thread/NORM_PRIORITY]
 ;; max-buffer-size绑定worker进程最大输入队列大小
 (let [max-buffer-size (int max-buffer-size)
     ;; 调用backtype.storm.messaging.netty.Context的bind方法建立一个服务器端的连接,socket绑定backtype.storm.messaging.netty.Server实例
       socket (.bind ^IContext context storm-id port)
       ;; thread-count绑定接收线程数,默认值为1
       thread-count (if receiver-thread-count receiver-thread-count 1)
       ;; 调用mk-receive-threads函数创建接收线程,vthreads绑定接收线程所对应的SmartThread实例,通过该实例我们可以start、join、interrupt接收线程,mk-receive-threads函数请参见其定义部分
       vthreads (mk-receive-threads context storm-id port transfer-local-fn daemon kill-fn priority socket max-buffer-size thread-count)]
   ;; 返回一个匿名函数,该匿名函数的主要功能就是通过向task_id=-1的任务发送一个空消息来关闭接收线程
   (fn []
     ;; 向本地端口port创建连接
     (let [kill-socket (.connect ^IContext context storm-id "localhost" port)]
       (log-message "Shutting down receiving-thread: [" storm-id ", " port "]")
       ;; 向task_id=-1的任务发送一个空消息,接收线程在接收消息时,首先检查是否是发送给task_id=-1消息,如果是则关闭接收线程
       (.send ^IConnection kill-socket
                 -1 (byte-array []))
       ;; 关闭连接
       (.close ^IConnection kill-socket)
       
       (log-message "Waiting for receiving-thread:[" storm-id ", " port "] to die")
       ;; 等待所有接收线程结束
       (for [thread-id (range thread-count)]
            (.join (vthreads thread-id)))
       
       (log-message "Shutdown receiving-thread: [" storm-id ", " port "]")
       ))))

mk-receive-threads函数:

;; mk-receive-threads函数循环调用mk-receive-thread函数创建接收线程,mk-receive-thread请参见其定义部分
(defn- mk-receive-threads [context storm-id port transfer-local-fn  daemon kill-fn priority socket max-buffer-size thread-count]
 (into [] (for [thread-id (range thread-count)]
            (mk-receive-thread context storm-id port transfer-local-fn  daemon kill-fn priority socket max-buffer-size thread-id))))

mk-receive-thread函数:

(defn- mk-receive-thread [context storm-id port transfer-local-fn  daemon kill-fn priority socket max-buffer-size thread-id]
   ;; async-loop函数接收一个"函数"或"函数工厂"作为参数生成一个java thread,这个java thread不断循环执行这个"函数"或"函数工厂"生产的函数。async-loop函数返回实现SmartThread协议的实例,通过该实例我们可以start、join、interrupt接收线程
   (async-loop
      ;; 这个参数就是一个"函数工厂","函数工厂"就是一个返回函数的函数
      (fn []
        (log-message "Starting receive-thread: [stormId: " storm-id ", port: " port ", thread-id: " thread-id  " ]")
        ;; 生成的java thread的run方法不断循环执行该函数
        (fn []
          ;; batched是一个ArrayList对象
          (let [batched (ArrayList.)
                ;; backtype.storm.messaging.netty.Server的recv方法返回ArrayList<TaskMessage>的Iterator<TaskMessage>。关于消息的处理流程会在以后文章中具体分析
                ^Iterator iter (.recv ^IConnection socket 0 thread-id)
                closed (atom false)]
            ;; 当iter不为nil,遍历iter
            (when iter
              (while (and (not @closed) (.hasNext iter))
                 ;; packet绑定一个TaskMessage对象,TaskMessage有两个成员属性task和message,task表示处理该消息的任务id,message表示消息的byte数组
                 (let [packet (.next iter)
                     ;; task绑定接收该消息的任务id
                       task (if packet (.task ^TaskMessage packet))
                       ;; message绑定消息的byte数组
                       message (if packet (.message ^TaskMessage packet))]
                     ;; 如果task=-1,则关闭接收线程
                     (if (= task -1)
                        (do (log-message "Receiving-thread:[" storm-id ", " port "] received shutdown notice")
                          (.close socket)
                          (reset! closed  true))
                        ;; 否则将数组[task message]添加到batched
                        (when packet (.add batched [task message]))))))
            ;; 如果接收线程关闭标识closed值为false,则调用transfer-local-fn函数将接收到的一批消息发送给task对应的接收队列
            (when (not @closed)
              (do
                (if (> (.size batched) 0)
                  (transfer-local-fn batched))
                ;; 0表示函数执行完一次不需要sleep,直接进行下一次执行
                0)))))
        ;; 表示参数是一个"函数工厂"
        :factory? true
        ;; daemon的值为true,所以接收线程是一个守护线程
        :daemon daemon
        ;; 指定kill函数
        :kill-fn kill-fn
        ;; 指定java thread的优先级
        :priority priority
        ;; 指定接收线程的名称为"worker-receiver-thread-"+thread-id
        :thread-name (str "worker-receiver-thread-" thread-id)))

 

以上就是supervisor启动worker的源码分析,启动worker的过程中涉及了executor的相关内容,这里没有详细分析,会在以后进行分析。同时也涉及了跟消息队列相关的内容
也会在以后进行详细分析。

posted @ 2015-04-03 10:35  不懂0604  阅读(1717)  评论(1编辑  收藏  举报