284
Storm-源碼分析-Topology Submit-Task
mk-task, 比較簡單, 因為task隻是概念上的結構, 不象其他worker, executor都需要創建進程或線程
所以其核心其實就是mk-task-data,
1. 創建TopologyContext對象, 其實就是把之前的topology對象和worker-data混合到一起, 便於task在執行時可以取到需要的topology信息.
2. 創建task-object, spout-object或bolt-object, 封裝相應的邏輯, 如nextTuple, execute
3. 生成tasks-fn, 名字起的不好,讓人誤解執行了task的功能, 其實就是做些emit之間的準備工作, 其中最重要的就是調用grouper去產生targets task, 當然還包含些metrics, hooks的調用.
說白了其實mk-tasks, 沒做啥事
(defn mk-task [executor-data task-id]
(let [task-data (mk-task-data executor-data task-id) ;;1 mk-task-data
storm-conf (:storm-conf executor-data)]
(doseq [klass (storm-conf TOPOLOGY-AUTO-TASK-HOOKS)] ;; add預定義的hooks
(.addTaskHook ^TopologyContext (:user-context task-data) (-> klass Class/forName .newInstance)))
;; when this is called, the threads for the executor haven't been started yet,
;; so we won't be risking trampling on the single-threaded claim strategy disruptor queue
(send-unanchored task-data SYSTEM-STREAM-ID ["startup"]) ;;向SYSTEM-STREAM, 發送startup通知,誰會接收SYSTEM-STREAM…?
task-data
))
1 mk-task-data
(defn mk-task-data [executor-data task-id]
(recursive-map
:executor-data executor-data
:task-id task-id
:system-context (system-topology-context (:worker executor-data) executor-data task-id)
:user-context (user-topology-context (:worker executor-data) executor-data task-id)
:builtin-metrics (builtin-metrics/make-data (:type executor-data))
:tasks-fn (mk-tasks-fn <>)
:object (get-task-object (.getRawTopology ^TopologyContext (:system-context <>)) (:component-id executor-data))))1.1 TopologyContext
Storm-源碼分析-Topology Submit-Task-TopologyContext
:system-context, :user-context, 隻是context中的topology對象不同, system為system-topology!
1.2 builtin-metrics/make-data
這裏的builtin-metrics用來記錄spout或bolt的執行狀況的metrics
1.3 mk-tasks-fn
返回tasks-fn, 這個函數主要用於做emit之前的準備工作, 返回target tasks list
1. 調用grouper, 產生target tasks
2. 執行emit hook
3. 滿足sampler條件時, 更新stats和buildin-metrics
task-fn, 兩種不同參數版本
[^String stream ^List values], 這個版本好理解些, 就是將stream對應的component的target tasks都算上(一個stream可能有多個out component, 一份數據需要發到多個bolt處理)
[^Integer out-task-id ^String stream ^List values], 指定out-task-id, 即direct grouping
這裏對out-task-id做了驗證
out-task-id (if grouping out-task-id), 即out-task-id->component->grouper不為nil(為:direct?), 即驗證這個stream確實有到該out-task-id對應component
如果驗證失敗, 將out-task-id置nil
(defn mk-tasks-fn [task-data]
(let [task-id (:task-id task-data)
executor-data (:executor-data task-data)
component-id (:component-id executor-data)
^WorkerTopologyContext worker-context (:worker-context executor-data)
storm-conf (:storm-conf executor-data)
emit-sampler (mk-stats-sampler storm-conf)
stream->component->grouper (:stream->component->grouper executor-data) ;;Storm-源碼分析-Streaming Grouping
user-context (:user-context task-data)
executor-stats (:stats executor-data)
debug? (= true (storm-conf TOPOLOGY-DEBUG))]
(fn ([^Integer out-task-id ^String stream ^List values]
(when debug?
(log-message "Emitting direct: " out-task-id "; " component-id " " stream " " values))
(let [target-component (.getComponentId worker-context out-task-id)
component->grouping (get stream->component->grouper stream)
grouping (get component->grouping target-component)
out-task-id (if grouping out-task-id)]
(when (and (not-nil? grouping) (not= :direct grouping))
(throw (IllegalArgumentException. "Cannot emitDirect to a task expecting a regular grouping")))
(apply-hooks user-context .emit (EmitInfo. values stream task-id [out-task-id]))
(when (emit-sampler)
(builtin-metrics/emitted-tuple! (:builtin-metrics task-data) executor-stats stream)
(stats/emitted-tuple! executor-stats stream)
(if out-task-id
(stats/transferred-tuples! executor-stats stream 1)
(builtin-metrics/transferred-tuple! (:builtin-metrics task-data) executor-stats stream 1)))
(if out-task-id [out-task-id])
))
([^String stream ^List values]
(when debug?
(log-message "Emitting: " component-id " " stream " " values))
(let [out-tasks (ArrayList.)]
(fast-map-iter [[out-component grouper] (get stream->component->grouper stream)]
(when (= :direct grouper)
;; TODO: this is wrong, need to check how the stream was declared
(throw (IllegalArgumentException. "Cannot do regular emit to direct stream")))
(let [comp-tasks (grouper task-id values)] ;;執行grouper, 產生target tasks
(if (or (sequential? comp-tasks) (instance? Collection comp-tasks))
(.addAll out-tasks comp-tasks)
(.add out-tasks comp-tasks)
)))
(apply-hooks user-context .emit (EmitInfo. values stream task-id out-tasks)) ;;執行事先注冊的emit hook
(when (emit-sampler) ;;滿足抽樣條件時, 更新stats和buildin-metrics中的emitted和transferred metric
(stats/emitted-tuple! executor-stats stream)
(builtin-metrics/emitted-tuple! (:builtin-metrics task-data) executor-stats stream)
(stats/transferred-tuples! executor-stats stream (count out-tasks))
(builtin-metrics/transferred-tuple! (:builtin-metrics task-data) executor-stats stream (count out-tasks)))
out-tasks)))
))
1.4 get-task-object
取出component的對象,
比如對於Spout, 取出SpoutSpec中的ComponentObject spout_object, 包含了spout的邏輯, 比如nextTuple()
(defn- get-task-object [^TopologyContext topology component-id] (let [spouts (.get_spouts topology) bolts (.get_bolts topology) state-spouts (.get_state_spouts topology) obj (Utils/getSetComponentObject (cond (contains? spouts component-id) (.get_spout_object ^SpoutSpec (get spouts component-id)) (contains? bolts component-id) (.get_bolt_object ^Bolt (get bolts component-id)) (contains? state-spouts component-id) (.get_state_spout_object ^StateSpoutSpec (get state-spouts component-id)) true (throw-runtime "Could not find " component-id " in " topology))) obj (if (instance? ShellComponent obj) (if (contains? spouts component-id) (ShellSpout. obj) (ShellBolt. obj)) obj ) obj (if (instance? JavaObject obj) (thrift/instantiate-java-object obj) obj )] obj ))
本文章摘自博客園，原文發布日期：2013-07-31
最後更新：2017-05-19 10:25:18