标签：des style http color io os 使用 java ar
supervisor是storm集群重要组成部分，supervisor主要负责管理各个"工作节点"。supervisor与zookeeper进行通信，通过zookeeper的"watch机制"可以感知到是否有新的任务需要认领或哪些任务被重新分配。我们可以通用执行bin/storm supervisor >/dev/null 2>&1 &来启动supervisor。bin/storm是一个python脚本，在这个脚本中定义了一个supervisor函数：
supervisor函数
def supervisor(klass="backtype.storm.daemon.supervisor"):
"""Syntax: [storm supervisor]
Launches the supervisor daemon. This command should be run
under supervision with a tool like daemontools or monit.
See Setting up a Storm cluster for more information.
(https://github.com/nathanmarz/storm/wiki/Setting-up-a-Storm-cluster)
"""
cppaths = [STORM_DIR + "/log4j", STORM_DIR + "/conf"]
jvmopts = parse_args(confvalue("supervisor.childopts", cppaths)) + [
"-Dlogfile.name=supervisor.log",
"-Dlog4j.configuration=storm.log.properties",
]
exec_storm_class(
klass,
jvmtype="-server",
extrajars=cppaths,
jvmopts=jvmopts)
klass参数的默认值为backtype.storm.daemon.supervisor，backtype.storm.daemon.supervisor标识一个java类。STORM_DIR标识storm的安装目录，cppaths集合存放了log4j配置文件路径和storm配置文件storm.yaml路径，jvmopts存放传递给jvm的参数，包括log4j配文件路径、storm.yaml路径、log4j日志名称和log4j配置文件名称。exec_storm_class函数的逻辑比较简单，具体实现如下：
exec_storm_class函数
def exec_storm_class(klass, jvmtype="-server", jvmopts=[], extrajars=[], args=[], fork=False):
global CONFFILE
all_args = [
"java", jvmtype, get_config_opts(),
"-Dstorm.home=" + STORM_DIR,
"-Djava.library.path=" + confvalue("java.library.path", extrajars),
"-Dstorm.conf.file=" + CONFFILE,
"-cp", get_classpath(extrajars),
] + jvmopts + [klass] + list(args)
print "Running: " + " ".join(all_args)
if fork:
os.spawnvp(os.P_WAIT, "java", all_args)
else:
os.execvp("java", all_args) # replaces the current process and never returns
get_config_opts()获取jvm的默认配置信息，confvalue("java.library.path", extrajars)获取storm使用的本地库JZMQ加载路径，get_classpath(extrajars)获取所有依赖jar包的完整路径，然后拼接一个java -cp命令运行klass的main方法。klass默认值为backtype.storm.daemon.supervisor，所以exec_storm_class函数最终调用backtype.storm.daemon.supervisor类的main方法。
backtype.storm.daemon.supervisor类定义在supervisor.clj文件中，定义如下：
backtype.storm.daemon.supervisor类
(ns backtype.storm.daemon.supervisor
(:import [backtype.storm.scheduler ISupervisor])
(:use [backtype.storm bootstrap])
(:use [backtype.storm.daemon common])
(:require [backtype.storm.daemon [worker :as worker]])
(:gen-class
:methods [^{:static true} [launch [backtype.storm.scheduler.ISupervisor] void]]))
(bootstrap)
;; ... ...
;; 其他方法
;; ... ...
(defn -main []
(-launch (standalone-supervisor)))
:gen-class指示Clojure生成Java类backtype.storm.daemon.supervisor，并且声明一个静态方法launch，launch方法接收一个实现backtype.storm.scheduler.ISupervisor接口的实例作为参数。launch函数的参数是由standalone-supervisor函数生成的。standalone-supervisor函数定义如下：返回一个实现ISupervisor接口的实例。
standalone-supervisor函数
;; 该函数主要返回一个实现了ISupervisor接口的实例。
(defn standalone-supervisor []
(let [conf-atom (atom nil)
id-atom (atom nil)]
(reify ISupervisor
;; prepare方法主要功能是创建一个基于磁盘的存放K/V对的database——LocalState对象，LocalState类参见其定义部分
(prepare [this conf local-dir]
;; conf-atom原子类型，绑定storm集群配置信息
(reset! conf-atom conf)
;; state绑定LocalState对象，local-dir标识database在磁盘上的根目录，database实际就是一个HashMap对象序列化后存放到磁盘local-dir目录下
(let [state (LocalState. local-dir)
;; LS-ID值为字符串"supervisor-id"，定义在common.clj文件中。如果state中存放了该supervisor的id，那么curr-id绑定该id，否则curr-id绑定32为uuid
curr-id (if-let [id (.get state LS-ID)]
id
;; 调用uuid函数生成一个32的id
(generate-supervisor-id))]
;; 调用state的put函数，更新该supervisor的id
(.put state LS-ID curr-id)
;; id-atom原子类型，绑定该supervisor的id
(reset! id-atom curr-id))
)
;; 返回true
(confirmAssigned [this port]
true)
;; 从storm配置信息中获取supervisor的所有端口，因为clojure中的map函数返回的是"懒惰序列"，所以需要调用doall函数对"懒惰序列"进行完全实例化
(getMetadata [this]
(doall (map int (get @conf-atom SUPERVISOR-SLOTS-PORTS))))
;; 获取supervisor的id
(getSupervisorId [this]
@id-atom)
;; 获取supervisor的分配id即其id
(getAssignmentId [this]
@id-atom)
;; killedWorker空实现
(killedWorker [this port]
)
;; assigned空实现
(assigned [this ports]
))))
LocalState类是个java类，定义见LocalState.java，这个类有一个VersionedStore类型对象，VersionedStore类见VersionedStore.java，由于这两个类是java实现，而且也比较简单，这样就在详细分析。
mk-supervisor函数定义如下：
mk-supervisor函数
;; conf绑定storm集群配置信息，isupervisor绑定standalone-supervisor函数返回的实现ISupervisor接口的实例
(defserverfn mk-supervisor [conf shared-context ^ISupervisor isupervisor]、
;; 打印日志信息
(log-message "Starting Supervisor with conf " conf)
;; supervisor-isupervisor-dir函数调用了supervisor-local-dir函数，supervisor-local-dir函数从storm配置中获取storm的安装路径，然后在supervisor上创建目录{storm.local.dir}/supervisor，并返回目录
;; supervisor-isupervisor-dir函数返回字符串"{storm.local.dir}/supervisor/isupervisor"作为一个LocalState对象的根目录，该LocalState对象只用来存放supervisor的id
;; 调用isupervisor的prepare方法，创建一个LocalState对象，并生成该supervisor的id，将其存入LocalState对象
(.prepare isupervisor conf (supervisor-isupervisor-dir conf))
;; supervisor-tmp-dir函数在supervisor上创建{storm.local.dir}/supervisor/tmp目录；FileUtils类的cleanDirectory方法清空该目录
(FileUtils/cleanDirectory (File. (supervisor-tmp-dir conf)))
;; supervisor绑定supervisor元数据信息，supervisor-data参见其定义部分
(let [supervisor (supervisor-data conf shared-context isupervisor)
;; event-manager和processes-event-manager分别绑定一个EventManager实例，managers绑定包含两个EventManager实例的集合。event-manager函数请参见文章"storm事件管理器EventManager源码分析-event.clj"
[event-manager processes-event-manager :as managers] [(event/event-manager false) (event/event-manager false)]
;; partial用于定义"偏函数"，所谓偏函数就是给一个指定函数的某些参数预赋值，这样就得到了一个新函数。sync-processes就绑定这个新函数，sync-processes参见其定义部分
sync-processes (partial sync-processes supervisor)
;; synchronize-supervisor绑定一个函数，该函数主要功能就是当assignment发生变化时, 从nimbus同步topology的代码到本地，当assignment发生变化时, check workers状态, 保证被分配的work的状态都是valid
synchronize-supervisor (mk-synchronize-supervisor supervisor sync-processes event-manager processes-event-manager)
;; heartbeat-fn绑定一个匿名函数，该匿名函数主要功能是调用StormClusterState实例的supervisor-heartbeat!函数将该supervisor的心跳信息SupervisorInfo实例写入zookeeper的"/supervisors/supervisor-id"节点中
heartbeat-fn (fn [] (.supervisor-heartbeat!
;; StormClusterState实例
(:storm-cluster-state supervisor)
;; supervisor-id
(:supervisor-id supervisor)
;; 创建SupervisorInfo实例，即该supervisor的心跳信息
(SupervisorInfo. (current-time-secs)
;; 主机名
(:my-hostname supervisor)
;; assignment-id即supervisor-id
(:assignment-id supervisor)
;; 当前集群已使用的所有port
(keys @(:curr-assignment supervisor))
;; used ports
;; 该supervisor上所有可用port，即在storm配置文件中配置的port
(.getMetadata isupervisor)
(conf SUPERVISOR-SCHEDULER-META)
;; supervisor启动时间
((:uptime supervisor)))))]
;; 调用heartbeat-fn绑定的匿名函数，将supervisor心跳心跳写入zookeeper
(heartbeat-fn)
;; should synchronize supervisor so it doesn‘t launch anything after being down (optimization)
;; 调用timer.clj中的schedule-recurring函数向该supervisor的定时器中添加一个周期执行的定时任务heartbeat-fn--"向zookeeper汇报superior的心跳信息"，关于storm定时器的详细信息请参看"storm定时器timer源码分析-timer.clj"
(schedule-recurring (:timer supervisor)
0
;; 指定每隔多长时间汇报一次心跳信息
(conf SUPERVISOR-HEARTBEAT-FREQUENCY-SECS)
heartbeat-fn)
;; 如果supervisor.enable值为true时(默认值就是true，而且不会改变，所以一定会执行)，那么将synchronize-supervisor绑定的函数（mk-synchronize-supervisor函数返回的函数）每隔10s加入event-manager事件管理器中，
;; 这样即使zookeeper的"watcher机制"异常时，supervisor也可以主动的获取分配信息的变化。同时将sync-processes绑定的函数（sync-processes函数）每隔SUPERVISOR-MONITOR-FREQUENCY-SECS秒加入processes-event-manager事件管理器中，
;; 这样即使zookeeper的"watcher机制"异常时，supervisor也可以正常管理worker
(when (conf SUPERVISOR-ENABLE)
;; This isn‘t strictly necessary, but it doesn‘t hurt and ensures that the machine stays up
;; to date even if callbacks don‘t all work exactly right
(schedule-recurring (:timer supervisor) 0 10 (fn [] (.add event-manager synchronize-supervisor)))
(schedule-recurring (:timer supervisor)
0
(conf SUPERVISOR-MONITOR-FREQUENCY-SECS)
(fn [] (.add processes-event-manager sync-processes))))
(log-message "Starting supervisor with id " (:supervisor-id supervisor) " at host " (:my-hostname supervisor))
;; 返回实现了Shutdownable接口、SupervisorDaemon协议和DaemonCommon协议的实例
(reify
Shutdownable
;; 关闭supervisor，就是关闭该supervisor所拥护的资源
(shutdown [this]
(log-message "Shutting down supervisor " (:supervisor-id supervisor))
(reset! (:active supervisor) false)
(cancel-timer (:timer supervisor))
(.shutdown event-manager)
(.shutdown processes-event-manager)
(.disconnect (:storm-cluster-state supervisor)))
SupervisorDaemon
;; 返回集群配置信息
(get-conf [this]
conf)
;; 返回supervisor-id
(get-id [this]
(:supervisor-id supervisor))
;; 见名知意，关闭所有worker
(shutdown-all-workers [this]
(let [ids (my-worker-ids conf)]
(doseq [id ids]
(shutdown-worker supervisor id)
)))
DaemonCommon
(waiting? [this]
(or (not @(:active supervisor))
(and
;; 定时器线程是否处于sleep状态
(timer-waiting? (:timer supervisor))
;; 调用事件管理器的waiting?函数检查event-manager和processes-event-manager内事件执行线程是否处于sleep状态，memfn宏可以自动生成代码以使得java方法可以当成clojure里面的函数
(every? (memfn waiting?) managers)))
))))
supervisor-data函数定义如下：
supervisor-data函数返回一个包含了supervisor元数据的map对象。
supervisor-data函数
(defn supervisor-data [conf shared-context ^ISupervisor isupervisor]
;; 保存集群配置信息
{:conf conf
;; 启动supervisor时，shared-context为nil
:shared-context shared-context
;; 保存supervisor实例
:isupervisor isupervisor
;; 保存supervisor是否是活跃的（默认是活跃的）
:active (atom true)
;; 保存supervisor启动时间
:uptime (uptime-computer)
;; 保存工作线程id
:worker-thread-pids-atom (atom {})
;; 保存StormClusterState对象
:storm-cluster-state (cluster/mk-storm-cluster-state conf)
;; 保存supervisor的LocalState对象，该LocalState对象的根目录是"{storm.local.dir}/supervisor/localstate"
:local-state (supervisor-state conf)
;; 保存supervisor的id
:supervisor-id (.getSupervisorId isupervisor)
;; 保存supervisor的分配id，分配id与supervisor_id相同
:assignment-id (.getAssignmentId isupervisor)
;; 保存supervisor的主机名，如果配置conf（map对象）中包含"storm.local.hostname"，那么就使用配置的主机名，否则通过调用InetAddress.getLocalHost().getCanonicalHostName()获取主机名
:my-hostname (if (contains? conf STORM-LOCAL-HOSTNAME)
(conf STORM-LOCAL-HOSTNAME)
(local-hostname))
;; 心跳时汇报当前集群的所有分配信息
:curr-assignment (atom nil) ;; used for reporting used ports when heartbeating
;; 保存一个storm定时器timer，kill-fn函数会在timer-thread发生exception的时候被调用
:timer (mk-timer :kill-fn (fn [t]
(log-error t "Error when processing event")
(exit-process! 20 "Error when processing an event")
))
;; 创建一个用于存放带有版本号的分配信息的map
:assignment-versions (atom {})
})
sync-processes函数定义如下：
sync-processes函数
;; 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]
;; conf绑定storm的配置信息map
(let [conf (:conf supervisor)
;; local-state绑定supervisor的LocalState实例
^LocalState local-state (:local-state supervisor)
;; 从supervisor的LocalState实例中获取本地分配信息端口port->LocalAssignment实例的map，LocalAssignment实例封装了storm-id和分配给该storm-id的executors
assigned-executors (defaulted (.get local-state LS-LOCAL-ASSIGNMENTS) {})
;; now绑定当前时间
now (current-time-secs)
;; allocated绑定worker-id->worker状态和心跳的map，read-allocated-workers函数请参见其定义部分
allocated (read-allocated-workers supervisor assigned-executors now)
;; 过滤掉allocated中state不等于:valid的元素，并将过滤后的结果绑定到keepers
keepers (filter-val
(fn [[state _]] (= state :valid))
allocated)
;; keep-ports绑定keepers中心跳信息所包含的端口
keep-ports (set (for [[id [_ hb]] keepers] (:port hb)))
;; reassign-executors绑定assigned-executors中端口不在集合keep-ports的键值对构成的map，也就是说已分配的线程所对应的进程挂掉了，需要重新进行分配
reassign-executors (select-keys-pred (complement keep-ports) assigned-executors)
;; new-worker-ids绑定port->worker-id的map，new-worker-ids保存了需要重新启动进程的worker-id
new-worker-ids (into
{}
(for [port (keys reassign-executors)]
[port (uuid)]))
]
;; 1. to kill are those in allocated that are dead or disallowed
;; 2. kill the ones that should be dead
;; - read pids, kill -9 and individually remove file
;; - rmr heartbeat dir, rmdir pid dir, rmdir id dir (catch exception and log)
;; 3. of the rest, figure out what assignments aren‘t yet satisfied
;; 4. generate new worker ids, write new "approved workers" to LS
;; 5. create local dir for worker id
;; 5. launch new workers (give worker-id, port, and supervisor-id)
;; 6. wait for workers launch
(log-debug "Syncing processes")
(log-debug "Assigned executors: " assigned-executors)
(log-debug "Allocated: " allocated)
;; allocated绑定worker-id->worker状态和心跳的map，id绑定worker-id，state绑定worker状态，heartbeat绑定worker心跳时间
(doseq [[id [state heartbeat]] allocated]
;; 如果worker的状态不是:valid，那么就关闭worker
(when (not= :valid state)
(log-message
"Shutting down and clearing state for id " id
". Current supervisor time: " now
". State: " state
", Heartbeat: " (pr-str heartbeat))
;; shutdown-worker函数关闭进程，shutdown-worker函数请参见其定义部分
(shutdown-worker supervisor id)
))
;; new-worker-ids保存了需要重新启动进程的worker-id，遍历new-worker-ids，为每个worker-id创建本地目录"{storm.local.dir}/workers/{worker_id}"
(doseq [id (vals new-worker-ids)]
(local-mkdirs (worker-pids-root conf id)))
;; 将合并后的map重新保存到local-state的LS-APPROVED-WORKERS中
(.put local-state LS-APPROVED-WORKERS
;; 将new-worker-ids的键值交换由原来的port->worker-id转换成worker-id->port，并与local-state的LS-APPROVED-WORKERS合并
(merge
;; select-keys函数从local-state的LS-APPROVED-WORKERS中获取key包含在keepers中的键值对，返回结果是一个map
(select-keys (.get local-state LS-APPROVED-WORKERS)
(keys keepers))
;; zipmap函数返回new-worker-ids的worker-id->port的map
(zipmap (vals new-worker-ids) (keys new-worker-ids))
))
;; wait-for-workers-launch函数等待所有worker启动完成，请参见wait-for-workers-launch函数定义部分
(wait-for-workers-launch
conf
;; assignment绑定在该port运行的executor信息
(dofor [[port assignment] reassign-executors]
;; id为port所对应的worker-id
(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，关于worker启动的相关分析会在以后的文章中详细介绍，在此不再介绍
(launch-worker supervisor
(:storm-id assignment)
port
id)
id)))
))
read-allocated-workers函数定义如下：
read-allocated-workers函数
;; 返回worker-id->worker状态和心跳的map，如果worker心跳为nil，那么worker是"dead"
(defn read-allocated-workers
"Returns map from worker id to worker heartbeat. if the heartbeat is nil, then the worker is dead (timed out or never wrote heartbeat)"
;; supervisor绑定supervisor元数据，assigned-executors绑定supervisor分配信息端口port->LocalAssignment实例的map，now绑定当前时间
[supervisor assigned-executors now]
;; 获取集群配置信息
(let [conf (:conf supervisor)
;; 获取supervisor的LocalState实例
^LocalState local-state (:local-state supervisor)
;; id->heartbeat绑定supervisor上运行进程的worker-id->心跳信息的map
id->heartbeat (read-worker-heartbeats conf)
;; approved-ids绑定supervisor的LocalState实例中保存的worker-id的集合
approved-ids (set (keys (.get local-state LS-APPROVED-WORKERS)))]
;; 生成worker-id->[state hb]的map
(into
{}
(dofor [[id hb] id->heartbeat]
;; cond相当于if...else嵌套
(let [state (cond
;; 如果心跳信息为nil，那么state值为:not-started关键字
(not hb)
:not-started
;; 如果approved-ids不包含id或者matches-an-assignment?返回false，那么state值为:disallowed关键字
(or (not (contains? approved-ids id))
;; matches-an-assignment?函数通过比较心跳信息和分配信息中的storm-id和线程id集合是否相同，来判定该worker是否已分配
(not (matches-an-assignment? hb assigned-executors)))
:disallowed
;; 如果当前时间-上次心跳时间>心跳超时时间，state值为:timed-out关键字
(> (- now (:time-secs hb))
(conf SUPERVISOR-WORKER-TIMEOUT-SECS))
:timed-out
;; 以上条件均不满足时，state值为:valid关键字
true
:valid)]
(log-debug "Worker " id " is " state ": " (pr-str hb) " at supervisor time-secs " now)
[id [state hb]]
))
)))
read-worker-heartbeats函数定义如下：
read-worker-heartbeats函数
;; 获取supervisor上运行进程的worker-id->心跳信息的map
(defn read-worker-heartbeats
"Returns map from worker id to heartbeat"
[conf]
;; ids绑定supervisor上进程的worker-id集合
(let [ids (my-worker-ids conf)]
;; 生成worker-id->心跳信息的map
(into {}
(dofor [id ids]
;; read-worker-heartbeat函数获取指定worker-id的心跳信息，从supervisor上"{storm.local.dir}/workers/{worker-id}/heartbeats"中获取心跳信息
[id (read-worker-heartbeat conf id)]))
))
my-worker-ids函数定义如下：
my-worker-ids函数
;; 获取supervisor上运行的进程的worker-id
(defn my-worker-ids [conf]
;; worker-root函数返回supervisor本地目录"{storm.local.dir}/workers"，read-dir-contents函数获取目录"{storm.local.dir}/workers"下所有文件名的集合（即该supervisor上正在运行的所有进程的worker-id）
(read-dir-contents (worker-root conf)))
matches-an-assignment?函数定义如下：
matches-an-assignment?函数
;; worker-heartbeat标识心跳信息，assigned-executors标识supervisor分配信息port->LocalAssignment实例的map
(defn matches-an-assignment? [worker-heartbeat assigned-executors]
;; 从worker-heartbeat中获取进程占用的端口，进而从assigned-executors中获取LocalAssignment实例
(let [local-assignment (assigned-executors (:port worker-heartbeat))]
;; 如果local-assignment不为nil，且心跳信息中的storm-id和分配信息中的storm-id相等，且心跳信息中的线程id集合和分配信息中的线程id集合相等，那么返回true；否则返回false
(and local-assignment
(= (:storm-id worker-heartbeat) (:storm-id local-assignment))
;; Constants/SYSTEM_EXECUTOR_ID标识"系统bolt"的线程id，我定义的topology除了我们指定的spout和bolt外，还包含一些"系统bolt"
(= (disj (set (:executors worker-heartbeat)) Constants/SYSTEM_EXECUTOR_ID)
(set (:executors local-assignment))))))
shutdown-worker函数定义如下：
shutdown-worker函数
;; 关闭进程，supervisor标识supervisor元数据，id标识worker_id
(defn shutdown-worker [supervisor id]
(log-message "Shutting down " (:supervisor-id supervisor) ":" id)
;; conf绑定集群配置信
(let [conf (:conf supervisor)
;; 注意当storm集群"分布式模式"运行时，supervisor的"{storm.local.dir}/workers/{worker_id}/pids"路径中存放了worker实际对应的jvm进程id
;; 从supervisor的"{storm.local.dir}/workers/{worker_id}/pids"路径获取进程id，worker_id标识我们指定的进程id，pids目录存放了该worker实际对应的jvm进程的id
pids (read-dir-contents (worker-pids-root conf id))
;; 注意当storm集群"本地模式"运行时，supervisor元数据中关键字:worker-thread-pids-atom所对应的map用于存放worker_id->线程id集合的键值对
;; 先从supervisor元数据中获取worker-id(我们人为分配给worker的id)->jvm进程id的map，thread-pid实际上绑定的是worker的jvm进程id
thread-pid (@(:worker-thread-pids-atom supervisor) id)]
;; 当thread-pid不为空时，kill掉该进程
(when thread-pid
;; 调用backtype.storm.process-simulator中的kill-process函数kill掉进程
(psim/kill-process thread-pid))
;; 遍历pids集合，kill掉每个进程
(doseq [pid pids]
;; 调用backtype.storm.util中的kill-process-with-sig-term函数，kill-process-with-sig-term函数又调用了send-signal-to-process函数，send-signal-to-process函数实现比较简单就是执行系统命令"kill -15 pid"，kill掉进程
;; 注意在创建worker进程时为worker进程指定了关闭回调函数，当调用"kill -15 pid"关闭worker进程时会触发回调函数执行，回调函数是在worker.clj的mk-worker函数中添加的
(kill-process-with-sig-term pid))
;; 如果pids不为空，sleep 1秒，等着"清理函数"--关闭回调函数执行完毕
(if-not (empty? pids) (sleep-secs 1)) ;; allow 1 second for execution of cleanup threads on worker.
;; 通过调用"kill -15 pid"命令未能关闭的进程，将通过调用force-kill-process函数关闭，force-kill-process函数只是调用了"kill -9 pid"命令
(doseq [pid pids]
(force-kill-process pid)
(try
;; 删除"{storm.local.dir}/workers/{worker_id}/pids"
(rmpath (worker-pid-path conf id pid))
(catch Exception e))) ;; on windows, the supervisor may still holds the lock on the worker directory
;; try-cleanup-worker函数清理本地目录，try-cleanup-worker函数参见其定义部分
(try-cleanup-worker conf id))
(log-message "Shut down " (:supervisor-id supervisor) ":" id))
try-cleanup-worker函数定义如下：
try-cleanup-worker函数
;; 清理本地目录
(defn try-cleanup-worker [conf id]
(try
;; 删除"{storm.local.dir}/workers/{worker_id}/heartbeats"目录
(rmr (worker-heartbeats-root conf id))
;; this avoids a race condition with worker or subprocess writing pid around same time
;; 删除"{storm.local.dir}/workers/{worker_id}/pids"目录
(rmpath (worker-pids-root conf id))
;; 删除"{storm.local.dir}/workers/{worker_id}"目录
(rmpath (worker-root conf id))
(catch RuntimeException e
(log-warn-error e "Failed to cleanup worker " id ". Will retry later")
)
(catch java.io.FileNotFoundException e (log-message (.getMessage e)))
(catch java.io.IOException e (log-message (.getMessage e)))
))
wait-for-workers-launch函数定义如下：
wait-for-workers-launch函数
;; wait-for-workers-launch函数等待所有worker启动完成
(defn- wait-for-workers-launch [conf ids]
(let [start-time (current-time-secs)]
(doseq [id ids]
;; 调用wait-for-worker-launch函数
(wait-for-worker-launch conf id start-time))
))
wait-for-worker-launch函数定义如下：
wait-for-worker-launch函数
;; wait-for-worker-launch函数等待worker启动完成，worker启动完成的条件是：如果worker在规定的心跳超时时间内有一次心跳那么就说明worker成功启动
(defn- wait-for-worker-launch [conf id start-time]
(let [state (worker-state conf id)]
(loop []
(let [hb (.get state LS-WORKER-HEARTBEAT)]
(when (and
(not hb)
(<
(- (current-time-secs) start-time)
(conf SUPERVISOR-WORKER-START-TIMEOUT-SECS)
))
(log-message id " still hasn‘t started")
(Time/sleep 500)
(recur)
)))
(when-not (.get state LS-WORKER-HEARTBEAT)
(log-message "Worker " id " failed to start")
)))
mk-synchronize-supervisor函数定义如下：
mk-synchronize-supervisor函数
;; mk-synchronize-supervisor函数返回一个名字为"this"的函数，
(defn mk-synchronize-supervisor [supervisor sync-processes event-manager processes-event-manager]
(fn this []
;; conf绑定集群配置信息
(let [conf (:conf supervisor)
;; storm-cluster-state绑定StormClusterState对象
storm-cluster-state (:storm-cluster-state supervisor)
;; isupervisor绑定实现了ISupervisor接口的实例
^ISupervisor isupervisor (:isupervisor supervisor)
;; local-state绑定LocalState实例
^LocalState local-state (:local-state supervisor)
;; sync-callback绑定一个匿名函数，这个匿名函数的主要功能就是将上面定义的"this"函数添加到event-manager中，这样"this"函数将会在一个新的线程内执行
;; 每次执行，都需要再一次把sync-callback注册到zookeeper中作为回调函数，以保证下次可以被继续触发，当zookeeper的子节点"/assignments"发生变化时执行回调函数sync-callback
sync-callback (fn [& ignored] (.add event-manager this))
;; assignment-versions绑定带有版本号的分配信息，topology-id->分配信息的map
assignment-versions @(:assignment-versions supervisor)
;; assignments-snapshot绑定topoloy-id->分配信息AssignmentInfo对象的map，versions绑定带有版本号的分配信息，assignments-snapshot函数从zookeeper的子节点"/assignments"获取分配信息（当前集群分配信息快照），并将回调函数添加到子节点"/assignments"上，assignments-snapshot函数参见其定义部分
{assignments-snapshot :assignments versions :versions} (assignments-snapshot
storm-cluster-state sync-callback
assignment-versions)
;; 调用read-storm-code-locations函数获取topology-id->nimbus上该topology代码目录的map
storm-code-map (read-storm-code-locations assignments-snapshot)
;; read-downloaded-storm-ids函数从supervisor本地的"{storm.local.dir}/stormdist"目录读取已经下载了代码jar包的topology-id
downloaded-storm-ids (set (read-downloaded-storm-ids conf))
;; all-assignment绑定该supervisor上的所有分配信息，即port->LocalAssignment对象的map
all-assignment (read-assignments
assignments-snapshot
(:assignment-id supervisor))
;; 调用isupervisor对象的confirmAssigned函数验证all-assignment的key即port的有效性，将通过验证的保存到new-assignment中。isupervisor对象是在standalone-supervisor函数中创建的，查看standalone-supervisor函数，我们可以发现isupervisor对象的confirmAssigned函数只是返回true，所以new-assignment=all-assignment
new-assignment (->> all-assignment
(filter-key #(.confirmAssigned isupervisor %)))
;; assigned-storm-ids绑定分配给该supervisor的topology-id的集合
assigned-storm-ids (assigned-storm-ids-from-port-assignments new-assignment)
;; existing-assignment绑定该supervisor上已经存在的分配信息
existing-assignment (.get local-state LS-LOCAL-ASSIGNMENTS)]
(log-debug "Synchronizing supervisor")
(log-debug "Storm code map: " storm-code-map)
(log-debug "Downloaded storm ids: " downloaded-storm-ids)
(log-debug "All assignment: " all-assignment)
(log-debug "New assignment: " new-assignment)
;; download code first
;; This might take awhile
;; - should this be done separately from usual monitoring?
;; should we only download when topology is assigned to this supervisor?
;; storm-code-map绑定当前集群上已分配的所有topology-id->nimbus上代码jar包目录的键值对的map
(doseq [[storm-id master-code-dir] storm-code-map]
;; 如果downloaded-storm-ids集合不包含该storm-id，且assigned-storm-ids集合包含该storm-id（表明该storm-id需要在该superior上运行，但是该storm-id的代码jar包还没有从nimbus服务器下载到本地），则调用download-storm-code函数下载代码jar包
(when (and (not (downloaded-storm-ids storm-id))
(assigned-storm-ids storm-id))
(log-message "Downloading code for storm id "
storm-id
" from "
master-code-dir)
;; 从nimbus服务器上下载该storm-id相关的代码jar包，序列化后的topology对象，运行时所需的配置信息，并将其保存到"{storm.local.dir}/nimbus/stormdist/{storm-id}/"目录
(download-storm-code conf storm-id master-code-dir)
(log-message "Finished downloading code for storm id "
storm-id
" from "
master-code-dir)
))
(log-debug "Writing new assignment "
(pr-str new-assignment))
;; existing-assignment与new-assignment的差集表示不需要在该supervisor上运行的分配的集合，所以要把这些分配对应的worker关闭
(doseq [p (set/difference (set (keys existing-assignment))
(set (keys new-assignment)))]
;; 当前storm版本0.9.2中，killedWorker为空实现，所以什么都没做
(.killedWorker isupervisor (int p)))
;; assigned函数为空实现，什么也没有做
(.assigned isupervisor (keys new-assignment))
;; 将最新分配信息new-assignment保存到local-state数据库中
(.put local-state
LS-LOCAL-ASSIGNMENTS
new-assignment)
;; 将带有版本号的分配信息versions存入supervisor缓存:assignment-versions中
(swap! (:assignment-versions supervisor) versions)
;; 重新设置supervisor缓存的:curr-assignment值为new-assignment，即保存当前storm集群上最新分配信息
(reset! (:curr-assignment supervisor) new-assignment)
;; remove any downloaded code that‘s no longer assigned or active
;; important that this happens after setting the local assignment so that
;; synchronize-supervisor doesn‘t try to launch workers for which the
;; resources don‘t exist
;; 如果当前supervisor服务器的操作系统是"Windows_NT"系统，那么执行shutdown-disallowed-workers函数，关闭状态为:disallowed的worker
(if on-windows? (shutdown-disallowed-workers supervisor))
;; 遍历downloaded-storm-ids集合，该集合内存放了已经下载了jar包等信息的topology的id
(doseq [storm-id downloaded-storm-ids]
;; 如果storm-id不在assigned-storm-ids集合内，则递归删除"{storm.local.dir}/supervisor/stormdist/{storm-id}"目录。assigned-storm-ids表示当前需要在该supervisor上运行的topology的id
(when-not (assigned-storm-ids storm-id)
(log-message "Removing code for storm id "
storm-id)
(try
(rmr (supervisor-stormdist-root conf storm-id))
(catch Exception e (log-message (.getMessage e))))
))
;; 将sync-processes函数添加到processes-event-manager事件管理器中，这样就可以在一个单独线程内执行sync-processes函数。因为sync-processes函数比较耗时，所以需要在一个新的线程内执行
(.add processes-event-manager sync-processes)
)))
assignments-snapshot函数定义如下：
assignments-snapshot函数
;; assignments-snapshot函数从zookeeper的子节点"/assignments"获取分配信息，并将回调函数添加到子节点"/assignments"上，assignment-versions绑定该supervisor本地缓存的带有版本号的分配信息
(defn- assignments-snapshot [storm-cluster-state callback assignment-versions]
;; storm-ids绑定已分配的topology-id的集合，获取/assignments的子节点列表，如果callback不为空，将其赋值给assignments-callback，并对/assignments添加"节点观察"，这样supervisor就能感知集群是否有新的assignment或者有assignment被删除
(let [storm-ids (.assignments storm-cluster-state callback)]
;; new-assignments绑定最新分配信息
(let [new-assignments
(->>
;; sid绑定topology-id
(dofor [sid storm-ids]
;; recorded-version绑定该supervisor上缓存的该sid的分配信息版本号
(let [recorded-version (:version (get assignment-versions sid))]
;; assignment-version绑定zookeeper上"/assignments/{sid}"节点数据及其版本号，并注册回调函数
(if-let [assignment-version (.assignment-version storm-cluster-state sid callback)]
;; 如果缓存的分配版本号和zookeeper上获取的分配版本号相等，则返回sid->缓存的分配信息的map，否则从zookeeper的"/assignments/{sid}"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supervisor就能感知某个已存在的assignment是否被重新分配
(if (= assignment-version recorded-version)
{sid (get assignment-versions sid)}
{sid (.assignment-info-with-version storm-cluster-state sid callback)})
;; 如果从zookeeper上获取分配信息失败，值为{sid nil}
{sid nil})))
;; 将dofor结果进行合并，形如：{sid_1 {:data data_1 :version version_1}, sid_2 {:data data_2 :version version_2},......sid_n {:data data_n :version version_n} }
(apply merge)
;; 保留值不空的键值对
(filter-val not-nil?))]
;; 返回的map形如：{:assignments {sid_1 data_1, sid_2 data_2, ...... , sid_n data_n}, :versions {sid_1 {:data data_1 :version version_1}, sid_2 {:data data_2 :version version_2},......sid_n {:data data_n :version version_n} } }
;; data_x是一个AssignmentInfo对象，AssignmentInfo对象包含对应的nimbus上的代码目录,所有task的启动时间,每个task与机器、端口的映射
{:assignments (into {} (for [[k v] new-assignments] [k (:data v)]))
:versions new-assignments})))
read-assignments函数定义如下：
read-assignments函数
(defn- read-assignments
"Returns map from port to struct containing :storm-id and :executors"
;; assignments-snapshot绑定topology-id->分配信息AssignmentInfo对象的map，assignment-id绑定supervisor-id
[assignments-snapshot assignment-id]
;; 遍历read-my-executors函数返回结果，检查是否存在多个topology分配到同一个端口，如果存在则抛出异常。检查的方式特别巧妙，通过对返回结果调用merge-with函数，如果返回结果中存在相同的port，那么就会调用
;; 匿名函数(fn [& ignored] ......)，这样就会抛出异常
(->> (dofor [sid (keys assignments-snapshot)] (read-my-executors assignments-snapshot sid assignment-id))
(apply merge-with (fn [& ignored] (throw-runtime "Should not have multiple topologies assigned to one port")))))
read-my-executor函数定义如下：
read-my-executor函数
;; assignments-snapshot绑定topology-id->分配信息AssignmentInfo对象的map，assignment-id绑定supervisor-id，storm-id为topoloy-id
(defn- read-my-executors [assignments-snapshot storm-id assignment-id]
(let [assignment (get assignments-snapshot storm-id)
;; my-executors绑定分配给该supervisor的executor信息，即executor->node+port的map
my-executors (filter (fn [[_ [node _]]] (= node assignment-id))
(:executor->node+port assignment))
;; port-executors绑定port->executor-id集合的map，merge-with函数的作用就是对key相同的value调用concat函数
port-executors (apply merge-with
concat
(for [[executor [_ port]] my-executors]
{port [executor]}
))]
;; 返回port->LocalAssignment对象的map，LocalAssignment包含两个属性：topology-id和executor-id集合
(into {} (for [[port executors] port-executors]
;; need to cast to int b/c it might be a long (due to how yaml parses things)
;; doall is to avoid serialization/deserialization problems with lazy seqs
[(Integer. port) (LocalAssignment. storm-id (doall executors))]
))))
download-storm-code函数定义如下：
download-storm-code函数
(defmulti download-storm-code cluster-mode)
download-storm-code函数是一个"多重函数"，根据cluster-mode函数的返回值决定调用哪个函数，cluster-mode函数可能返回关键字:distributed和:local，如果返回:distributed，那么会调用下面这个函数。
(defmethod download-storm-code
;; master-code-dir绑定storm-id的代码jar包在nimbus服务器上的路径
:distributed [conf storm-id master-code-dir]
;; Downloading to permanent location is atomic
;; tmproot绑定supervisor本地路径"{storm.local.dir}/supervisor/tmp/{uuid}"，临时存放从nimbus上下载的代码jar包
(let [tmproot (str (supervisor-tmp-dir conf) file-path-separator (uuid))
;; stormroot绑定该storm-id的代码jar包在supervisor上的路径"{storm.local.dir}/supervisor/stormdist/{storm-id}"
stormroot (supervisor-stormdist-root conf storm-id)]
;; 创建临时目录tmproot
(FileUtils/forceMkdir (File. tmproot))
;; 将nimbus服务器上的"{storm.local.dir}/nimbus/stormdist/{storm-id}/stormjar.jar"文件下载到supervisor服务器的tmproot目录中，stormjar.jar包含这个topology所有代码
(Utils/downloadFromMaster conf (master-stormjar-path master-code-dir) (supervisor-stormjar-path tmproot))
;; 将nimbus服务器上的"{storm.local.dir}/nimbus/stormdist/{storm-id}/stormcode.ser"文件下载到supervisor服务器的tmproot目录中，stormcode.ser是这个topology对象的序列化
(Utils/downloadFromMaster conf (master-stormcode-path master-code-dir) (supervisor-stormcode-path tmproot))
;; 将nimbus服务器上的"{storm.local.dir}/nimbus/stormdist/{storm-id}/stormconf.ser"文件下载到supervisor服务器的tmproot目录中，stormconf.ser包含运行这个topology的配置
(Utils/downloadFromMaster conf (master-stormconf-path master-code-dir) (supervisor-stormconf-path tmproot))
;; RESOURCES-SUBDIR值为字符串"resources"，extract-dir-from-jar函数主要作用就是将jar包解压，然后将jar包中路径以"resources"开头的文件解压到"{tmproot}/resources/......"目录
(extract-dir-from-jar (supervisor-stormjar-path tmproot) RESOURCES-SUBDIR tmproot)
;; 将临时目录tmproot中的文件剪切到stormroot目录中，这样"{storm.local.dir}/nimbus/stormdist/{storm-id}/"目录中将包括resources目录，stormjar.jar文件，stormcode.ser文件，stormconf.ser文件
(FileUtils/moveDirectory (File. tmproot) (File. stormroot))
))
extract-dir-from-jar函数定义如下：
extract-dir-from-jar函数
;; jarpath标识jar路径，dir标识"resources"，destdir标识"{tmproot}"路径
(defn extract-dir-from-jar [jarpath dir destdir]
(try-cause
;; 使用类ZipFile来解压jar包，jarpath绑定ZipFile对象
(with-open [jarpath (ZipFile. jarpath)]
;; 调用entries方法，返回一个枚举对象，然后调用enumeration-seq函数获取文件的ZIP条目对象
(let [entries (enumeration-seq (.entries jarpath))]
;; 遍历entries中路径以"resources"开头的文件
(doseq [file (filter (fn [entry](and (not (.isDirectory entry)) (.startsWith (.getName entry) dir))) entries)]
;; 在"tmproot"目录中创建文件的完整父路径
(.mkdirs (.getParentFile (File. destdir (.getName file))))
;; 将文件复制到"{tmproot}/{在压缩文件中的路径}"
(with-open [out (FileOutputStream. (File. destdir (.getName file)))]
(io/copy (.getInputStream jarpath file) out)))))
(catch IOException e
(log-message "Could not extract " dir " from " jarpath))))
以上就是storm启动supervisor的完整流程，启动supervisor的工作主要是在mk-supervisor函数中进行的，所以阅读该部分源码时，要首先从该函数入手，然后依次分析在该函数中所调用的其他函数，根据函数的控制流程分析每个函数。
storm启动supervisor源码分析-supervisor.clj
标签：des style http color io os 使用 java ar
原文地址：http://www.cnblogs.com/ierbar0604/p/3991488.html
