Storm-源码分析-Topology Submit-Task-TopologyContext (backtype.storm.task)

1. GeneralTopologyContext

记录了Topology的基本信息, 包含StormTopology, StormConf
已经从他们推导出的, task和component, component的streams, input/output信息

public class GeneralTopologyContext implements JSONAware {
    private StormTopology _topology; 
    private Map<Integer, String> _taskToComponent;
    private Map<String, List<Integer>> _componentToTasks;
    private Map<String, Map<String, Fields>> _componentToStreamToFields; //ComponentCommon.streams, map<string, StreamInfo>
    private String _stormId;   ;;topology id
    protected Map _stormConf;  

}

StormTopology, worker从磁盘stormcode.ser中读出

struct StormTopology {
  //ids must be unique across maps
  // #workers to use is in conf
  1: required map<string, SpoutSpec> spouts;
  2: required map<string, Bolt> bolts;
  3: required map<string, StateSpoutSpec> state_spouts;
}

StormConf, worker从磁盘stormconf.ser中读出

taskToComponent, componentToTasks, task和component的对应关系

componentToStreamToFields, component包含哪些streams, 每个stream包含哪些fields

除了显而易见的操作以外, 还有如下操作以获得component的输入和输出

    /**
     * Gets the declared inputs to the specified component.
     *
     * @return A map from subscribed component/stream to the grouping subscribed with.
     */
    public Map<GlobalStreamId, Grouping> getSources(String componentId) {
        return getComponentCommon(componentId).get_inputs();  //ComponentCommon.inputs,map<GlobalStreamId, Grouping>
    }
    /**
     * Gets information about who is consuming the outputs of the specified component,
     * and how.
     *
     * @return Map from stream id to component id to the Grouping used.
     */
    public Map<String, Map<String, Grouping>> getTargets(String componentId) {
        Map<String, Map<String, Grouping>> ret = new HashMap<String, Map<String, Grouping>>();
        for(String otherComponentId: getComponentIds()) {  //对所有components的id
            Map<GlobalStreamId, Grouping> inputs = getComponentCommon(otherComponentId).get_inputs();  //取出component的inputs
            for(GlobalStreamId id: inputs.keySet()) {  //对inputs里面的每个stream-id
                if(id.get_componentId().equals(componentId)) {    //判断stream的源component是否是该component
                    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; // [steamid, [target-componentid, grouping]]
    }

这里面的getComponentCommon和getComponentIds, 来自ThriftTopologyUtils类
不要误解, 不是通过thriftAPI去nimbus获取信息, 只是从StormTopology里面读信息, 而StormTopology类本身是generated by thrift
thrift产生的class, 是有metaDataMap的, 所以实现如下

    public static Set<String> getComponentIds(StormTopology topology) {
        Set<String> ret = new HashSet<String>();
        for(StormTopology._Fields f: StormTopology.metaDataMap.keySet()) {
            Map<String, Object> componentMap = (Map<String, Object>) topology.getFieldValue(f);
            ret.addAll(componentMap.keySet());
        }
        return ret;
    }
通过metaDataMap读出StormTopology里面有哪些field, spouts,bolts,state_spouts, 然后遍历getFieldValue, 将value中的keyset返回
这样做的好处是, 动态, 当StormTopology发生变化时, 代码不用改, 对于普通java class应该无法实现这样的功能, 但是对于python这样的动态语言, 就简单了
当然这里其实也可以不用ThriftTopologyUtils, 直接写死从StormTopology.spouts…中去读

 

从storm.thrift里面看看ComponentCommon的定义, 上面两个函数就很好理解了
getTargets的实现, 需要看看, 因为是从inputs去推出outputs
因为在ComponentCommon只记录了output的streamid以及fields, 但无法知道这个stream发往哪个component
但对于input, streamid是GlobalStreamId类型, GlobalStreamId里面不但包含streamid,还有源component的componentid
所以从这个可以反推, 只要源component是当前component, 那么说明该component是源component的target component

struct ComponentCommon {
  1: required map<GlobalStreamId, Grouping> inputs;
  2: required map<string, StreamInfo> streams; //key is stream id, outputs
  3: optional i32 parallelism_hint; //how many threads across the cluster should be dedicated to this component
  4: optional string json_conf;
}

struct SpoutSpec {
  1: required ComponentObject spout_object;
  2: required ComponentCommon common;
  // can force a spout to be non-distributed by overriding the component configuration
  // and setting TOPOLOGY_MAX_TASK_PARALLELISM to 1
}

struct Bolt {
  1: required ComponentObject bolt_object;
  2: required ComponentCommon common;
}

 

2. WorkerTopologyContext

WorkerTopologyContext封装了些worker相关信息

public class WorkerTopologyContext extends GeneralTopologyContext {
    public static final String SHARED_EXECUTOR = "executor";
    
    private Integer _workerPort;         ;;worker进程的port
    private List<Integer> _workerTasks;  ;;worker包含的taskids
    private String _codeDir;             ;;supervisor上的代码目录, stormdist/stormid
    private String _pidDir;              ;;记录worker运行进程(可能多个)的pids的目录,workid/pids 
    Map<String, Object> _userResources;
    Map<String, Object> _defaultResources;

}

 

3. TopologyContext

看注释, TopologyContext会作为bolt和spout的prepare(or open)函数的参数
所以用openOrPrepareWasCalled, 表示该TopologyContext是否被prepare调用过

registerMetric, 可以用于往_registeredMetrics中注册metics
注册的结构, [timeBucketSizeInSecs, [taskId, [name, metric]]]

_hooks, 用于注册task hook

/**
 * A TopologyContext is given to bolts and spouts in their "prepare" and "open"
 * methods, respectively. This object provides information about the component's
 * place within the topology, such as task ids, inputs and outputs, etc.
 *
 * <p>The TopologyContext is also used to declare ISubscribedState objects to
 * synchronize state with StateSpouts this object is subscribed to.</p>
 */
public class TopologyContext extends WorkerTopologyContext implements IMetricsContext {
    private Integer _taskId;
    private Map<String, Object> _taskData = new HashMap<String, Object>();
    private List<ITaskHook> _hooks = new ArrayList<ITaskHook>();
    private Map<String, Object> _executorData;
    private Map<Integer,Map<Integer, Map<String, IMetric>>> _registeredMetrics;
    private clojure.lang.Atom _openOrPrepareWasCalled;
    public TopologyContext(StormTopology topology, Map stormConf,
            Map<Integer, String> taskToComponent, Map<String, List<Integer>> componentToSortedTasks,
            Map<String, Map<String, Fields>> componentToStreamToFields,
            String stormId, String codeDir, String pidDir, Integer taskId,
            Integer workerPort, List<Integer> workerTasks, Map<String, Object> defaultResources,
            Map<String, Object> userResources, Map<String, Object> executorData, Map registeredMetrics,
            clojure.lang.Atom openOrPrepareWasCalled) {
        super(topology, stormConf, taskToComponent, componentToSortedTasks,
                componentToStreamToFields, stormId, codeDir, pidDir,
                workerPort, workerTasks, defaultResources, userResources);
        _taskId = taskId;
        _executorData = executorData;
        _registeredMetrics = registeredMetrics;
        _openOrPrepareWasCalled = openOrPrepareWasCalled;
    }

 

4. 使用

mk-task-data, 创建每个task的topology context

user-context (user-topology-context (:worker executor-data) executor-data task-id)
(defn user-topology-context [worker executor-data tid]
  ((mk-topology-context-builder
    worker
    executor-data
    (:topology worker))
   tid))

(defn mk-topology-context-builder [worker executor-data topology]
  (let [conf (:conf worker)]
    #(TopologyContext.
      topology
      (: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 (:storm-id worker)))
      (worker-pids-root conf (:worker-id worker))
      (int %)
      (:port worker)
      (:task-ids worker)
      (:default-shared-resources worker)
      (:user-shared-resources worker)
      (:shared-executor-data executor-data)
      (:interval->task->metric-registry executor-data)
      (:open-or-prepare-was-called? executor-data))))

posted on 2013-07-26 16:26  fxjwind  阅读(1949)  评论(4编辑  收藏  举报