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
;; 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
( 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:
( 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:
( 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:
;; 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:
( 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:
( 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:
( 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:
( 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:
"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:
;; 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:
;; 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:
( 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:
( 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:
( 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:
;; 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:
( 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:
;; 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:
( 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:
;; 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:
;; 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:
( 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:
( 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:
( 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:
( 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:
;; 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。