首页
学习
活动
专区
工具
TVP
发布
社区首页 >专栏 >聊聊storm WindowTridentProcessor的FreshCollector

聊聊storm WindowTridentProcessor的FreshCollector

作者头像
code4it
发布2018-12-13 17:59:14
3570
发布2018-12-13 17:59:14
举报

本文主要研究一下storm WindowTridentProcessor的FreshCollector

实例

        TridentTopology topology = new TridentTopology();
        topology.newStream("spout1", spout)
                .partitionBy(new Fields("user"))
                .window(windowConfig,windowsStoreFactory,new Fields("user","score"),new UserCountAggregator(),new Fields("aggData"))
                .parallelismHint(1)
                .each(new Fields("aggData"), new PrintEachFunc(),new Fields());
  • 这个实例在window操作之后跟了一个each操作

WindowTridentProcessor

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/windowing/WindowTridentProcessor.java

public class WindowTridentProcessor implements TridentProcessor {
    
    private FreshCollector collector;

    //......

    public void prepare(Map stormConf, TopologyContext context, TridentContext tridentContext) {
        this.topologyContext = context;
        List<TridentTuple.Factory> parents = tridentContext.getParentTupleFactories();
        if (parents.size() != 1) {
            throw new RuntimeException("Aggregation related operation can only have one parent");
        }

        Long maxTuplesCacheSize = getWindowTuplesCacheSize(stormConf);

        this.tridentContext = tridentContext;
        collector = new FreshCollector(tridentContext);
        projection = new TridentTupleView.ProjectionFactory(parents.get(0), inputFields);

        windowStore = windowStoreFactory.create(stormConf);
        windowTaskId = windowId + WindowsStore.KEY_SEPARATOR + topologyContext.getThisTaskId() + WindowsStore.KEY_SEPARATOR;
        windowTriggerInprocessId = getWindowTriggerInprocessIdPrefix(windowTaskId);

        tridentWindowManager = storeTuplesInStore ?
                new StoreBasedTridentWindowManager(windowConfig, windowTaskId, windowStore, aggregator, tridentContext.getDelegateCollector(), maxTuplesCacheSize, inputFields)
                : new InMemoryTridentWindowManager(windowConfig, windowTaskId, windowStore, aggregator, tridentContext.getDelegateCollector());

        tridentWindowManager.prepare();
    }

    public void finishBatch(ProcessorContext processorContext) {

        Object batchId = processorContext.batchId;
        Object batchTxnId = getBatchTxnId(batchId);

        LOG.debug("Received finishBatch of : [{}] ", batchId);
        // get all the tuples in a batch and add it to trident-window-manager
        List<TridentTuple> tuples = (List<TridentTuple>) processorContext.state[tridentContext.getStateIndex()];
        tridentWindowManager.addTuplesBatch(batchId, tuples);

        List<Integer> pendingTriggerIds = null;
        List<String> triggerKeys = new ArrayList<>();
        Iterable<Object> triggerValues = null;

        if (retriedAttempt(batchId)) {
            pendingTriggerIds = (List<Integer>) windowStore.get(inprocessTriggerKey(batchTxnId));
            if (pendingTriggerIds != null) {
                for (Integer pendingTriggerId : pendingTriggerIds) {
                    triggerKeys.add(triggerKey(pendingTriggerId));
                }
                triggerValues = windowStore.get(triggerKeys);
            }
        }

        // if there are no trigger values in earlier attempts or this is a new batch, emit pending triggers.
        if(triggerValues == null) {
            pendingTriggerIds = new ArrayList<>();
            Queue<StoreBasedTridentWindowManager.TriggerResult> pendingTriggers = tridentWindowManager.getPendingTriggers();
            LOG.debug("pending triggers at batch: [{}] and triggers.size: [{}] ", batchId, pendingTriggers.size());
            try {
                Iterator<StoreBasedTridentWindowManager.TriggerResult> pendingTriggersIter = pendingTriggers.iterator();
                List<Object> values = new ArrayList<>();
                StoreBasedTridentWindowManager.TriggerResult triggerResult = null;
                while (pendingTriggersIter.hasNext()) {
                    triggerResult = pendingTriggersIter.next();
                    for (List<Object> aggregatedResult : triggerResult.result) {
                        String triggerKey = triggerKey(triggerResult.id);
                        triggerKeys.add(triggerKey);
                        values.add(aggregatedResult);
                        pendingTriggerIds.add(triggerResult.id);
                    }
                    pendingTriggersIter.remove();
                }
                triggerValues = values;
            } finally {
                // store inprocess triggers of a batch in store for batch retries for any failures
                if (!pendingTriggerIds.isEmpty()) {
                    windowStore.put(inprocessTriggerKey(batchTxnId), pendingTriggerIds);
                }
            }
        }

        collector.setContext(processorContext);
        int i = 0;
        for (Object resultValue : triggerValues) {
            collector.emit(new ConsList(new TriggerInfo(windowTaskId, pendingTriggerIds.get(i++)), (List<Object>) resultValue));
        }
        collector.setContext(null);
    }
}
  • WindowTridentProcessor在prepare的时候创建了FreshCollector
  • finishBatch的时候,调用FreshCollector.emit将窗口的aggregate的结果集传递过去
  • 传递的数据结构为ConsList,其实是个AbstractList的实现,由Object类型的first元素,以及List<Object>结构的_elems组成

FreshCollector

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/planner/processor/FreshCollector.java

public class FreshCollector implements TridentCollector {
    FreshOutputFactory _factory;
    TridentContext _triContext;
    ProcessorContext context;
    
    public FreshCollector(TridentContext context) {
        _triContext = context;
        _factory = new FreshOutputFactory(context.getSelfOutputFields());
    }
                
    public void setContext(ProcessorContext pc) {
        this.context = pc;
    }

    @Override
    public void emit(List<Object> values) {
        TridentTuple toEmit = _factory.create(values);
        for(TupleReceiver r: _triContext.getReceivers()) {
            r.execute(context, _triContext.getOutStreamId(), toEmit);
        }            
    }

    @Override
    public void reportError(Throwable t) {
        _triContext.getDelegateCollector().reportError(t);
    } 

    public Factory getOutputFactory() {
        return _factory;
    }    
}
  • FreshCollector在构造器里头根据context的selfOutputFields(第一个field固定为_task_info,之后的几个field为用户在window方法定义的functionFields)构造FreshOutputFactory
  • emit方法,首先使用FreshOutputFactory根据outputFields构造TridentTupleView,之后获取TupleReceiver,调用TupleReceiver的execute方法把TridentTupleView传递过去
  • 这里的TupleReceiver有ProjectedProcessor、PartitionPersistProcessor

TridentTupleView.FreshOutputFactory

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/tuple/TridentTupleView.java

    public static class FreshOutputFactory  implements Factory {
        Map<String, ValuePointer> _fieldIndex;
        ValuePointer[] _index;

        public FreshOutputFactory(Fields selfFields) {
            _fieldIndex = new HashMap<>();
            for(int i=0; i<selfFields.size(); i++) {
                String field = selfFields.get(i);
                _fieldIndex.put(field, new ValuePointer(0, i, field));
            }
            _index = ValuePointer.buildIndex(selfFields, _fieldIndex);
        }
        
        public TridentTuple create(List<Object> selfVals) {
            return new TridentTupleView(PersistentVector.EMPTY.cons(selfVals), _index, _fieldIndex);
        }

        @Override
        public Map<String, ValuePointer> getFieldIndex() {
            return _fieldIndex;
        }

        @Override
        public int numDelegates() {
            return 1;
        }
        
        @Override
        public List<String> getOutputFields() {
            return indexToFieldsList(_index);
        }        
    }
  • FreshOutputFactory是TridentTupleView的一个静态类,其构造方法主要是计算index以及fieldIndex
  • fieldIndex是一个map,key是field字段,value是ValuePointer,记录其delegateIndex(这里固定为0)、index及field信息;第一个field为task_info,index为0;之后的fields为用户在window方法定义的functionFields
  • 这里的create方法主要是构造TridentTupleView,其构造器第一个值为IPersistentVector,第二个值为index,第三个值为fieldIndex

ValuePointer

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/tuple/ValuePointer.java

public class ValuePointer {
    public static Map<String, ValuePointer> buildFieldIndex(ValuePointer[] pointers) {
        Map<String, ValuePointer> ret = new HashMap<String, ValuePointer>();
        for(ValuePointer ptr: pointers) {
            ret.put(ptr.field, ptr);
        }
        return ret;        
    }

    public static ValuePointer[] buildIndex(Fields fieldsOrder, Map<String, ValuePointer> pointers) {
        if(fieldsOrder.size()!=pointers.size()) {
            throw new IllegalArgumentException("Fields order must be same length as pointers map");
        }
        ValuePointer[] ret = new ValuePointer[pointers.size()];
        for(int i=0; i<fieldsOrder.size(); i++) {
            ret[i] = pointers.get(fieldsOrder.get(i));
        }
        return ret;
    }    
    
    public int delegateIndex;
    protected int index;
    protected String field;
    
    public ValuePointer(int delegateIndex, int index, String field) {
        this.delegateIndex = delegateIndex;
        this.index = index;
        this.field = field;
    }

    @Override
    public String toString() {
        return ToStringBuilder.reflectionToString(this);
    }    
}
  • 这里的buildIndex,主要是根据selfOutputFields的顺序返回ValuePointer数组

ProjectedProcessor

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/planner/processor/ProjectedProcessor.java

public class ProjectedProcessor implements TridentProcessor {
    Fields _projectFields;
    ProjectionFactory _factory;
    TridentContext _context;
    
    public ProjectedProcessor(Fields projectFields) {
        _projectFields = projectFields;
    }
    
    @Override
    public void prepare(Map conf, TopologyContext context, TridentContext tridentContext) {
        if(tridentContext.getParentTupleFactories().size()!=1) {
            throw new RuntimeException("Projection processor can only have one parent");
        }
        _context = tridentContext;
        _factory = new ProjectionFactory(tridentContext.getParentTupleFactories().get(0), _projectFields);
    }

    @Override
    public void cleanup() {
    }

    @Override
    public void startBatch(ProcessorContext processorContext) {
    }

    @Override
    public void execute(ProcessorContext processorContext, String streamId, TridentTuple tuple) {
        TridentTuple toEmit = _factory.create(tuple);
        for(TupleReceiver r: _context.getReceivers()) {
            r.execute(processorContext, _context.getOutStreamId(), toEmit);
        }
    }

    @Override
    public void finishBatch(ProcessorContext processorContext) {
    }

    @Override
    public Factory getOutputFactory() {
        return _factory;
    }
}
  • ProjectedProcessor在prepare的时候,创建了ProjectionFactory,其_projectFields就是window方法定义的functionFields,这里还使用tridentContext.getParentTupleFactories().get(0)提取了parent的第一个Factory,由于是FreshCollector传递过来的,因而这里是TridentTupleView.FreshOutputFactory
  • execute的时候,首先调用ProjectionFactory.create方法,对TridentTupleView进行字段提取操作,toEmit就是根据window方法定义的functionFields重新提取的TridentTupleView
  • execute方法之后对_context.getReceivers()挨个调用execute操作,将toEmit传递过去,这里的receiver就是window操作之后的各种processor了,比如EachProcessor

TridentTupleView.ProjectionFactory

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/tuple/TridentTupleView.java

public static class ProjectionFactory implements Factory {
        Map<String, ValuePointer> _fieldIndex;
        ValuePointer[] _index;
        Factory _parent;

        public ProjectionFactory(Factory parent, Fields projectFields) {
            _parent = parent;
            if(projectFields==null) projectFields = new Fields();
            Map<String, ValuePointer> parentFieldIndex = parent.getFieldIndex();
            _fieldIndex = new HashMap<>();
            for(String f: projectFields) {
                _fieldIndex.put(f, parentFieldIndex.get(f));
            }            
            _index = ValuePointer.buildIndex(projectFields, _fieldIndex);
        }
        
        public TridentTuple create(TridentTuple parent) {
            if(_index.length==0) return EMPTY_TUPLE;
            else return new TridentTupleView(((TridentTupleView)parent)._delegates, _index, _fieldIndex);
        }

        @Override
        public Map<String, ValuePointer> getFieldIndex() {
            return _fieldIndex;
        }

        @Override
        public int numDelegates() {
            return _parent.numDelegates();
        }

        @Override
        public List<String> getOutputFields() {
            return indexToFieldsList(_index);
        }
    }
  • ProjectionFactory是TridentTupleView的静态类,它在构造器里头根据projectFields构造index及fieldIndex,这样create方法就能根据所需的字段创建TridentTupleView

EachProcessor

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/planner/processor/EachProcessor.java

public class EachProcessor implements TridentProcessor {
    Function _function;
    TridentContext _context;
    AppendCollector _collector;
    Fields _inputFields;
    ProjectionFactory _projection;
    
    public EachProcessor(Fields inputFields, Function function) {
        _function = function;
        _inputFields = inputFields;
    }
    
    @Override
    public void prepare(Map conf, TopologyContext context, TridentContext tridentContext) {
        List<Factory> parents = tridentContext.getParentTupleFactories();
        if(parents.size()!=1) {
            throw new RuntimeException("Each operation can only have one parent");
        }
        _context = tridentContext;
        _collector = new AppendCollector(tridentContext);
        _projection = new ProjectionFactory(parents.get(0), _inputFields);
        _function.prepare(conf, new TridentOperationContext(context, _projection));
    }

    @Override
    public void cleanup() {
        _function.cleanup();
    }    

    @Override
    public void execute(ProcessorContext processorContext, String streamId, TridentTuple tuple) {
        _collector.setContext(processorContext, tuple);
        _function.execute(_projection.create(tuple), _collector);
    }

    @Override
    public void startBatch(ProcessorContext processorContext) {
    }

    @Override
    public void finishBatch(ProcessorContext processorContext) {
    }

    @Override
    public Factory getOutputFactory() {
        return _collector.getOutputFactory();
    }    
}
  • EachProcessor的execute方法,首先设置collector的context为processorContext,然后调用function.execute方法
  • 这里调用了projection.create(tuple)来提取字段,主要是根据function定义的inputFields来提取
  • 这里传递给_function的collector为AppendCollector

AppendCollector

storm-core-1.2.2-sources.jar!/org/apache/storm/trident/planner/processor/AppendCollector.java

public class AppendCollector implements TridentCollector {
    OperationOutputFactory _factory;
    TridentContext _triContext;
    TridentTuple tuple;
    ProcessorContext context;
    
    public AppendCollector(TridentContext context) {
        _triContext = context;
        _factory = new OperationOutputFactory(context.getParentTupleFactories().get(0), context.getSelfOutputFields());
    }
                
    public void setContext(ProcessorContext pc, TridentTuple t) {
        this.context = pc;
        this.tuple = t;
    }

    @Override
    public void emit(List<Object> values) {
        TridentTuple toEmit = _factory.create((TridentTupleView) tuple, values);
        for(TupleReceiver r: _triContext.getReceivers()) {
            r.execute(context, _triContext.getOutStreamId(), toEmit);
        }
    }

    @Override
    public void reportError(Throwable t) {
        _triContext.getDelegateCollector().reportError(t);
    } 
    
    public Factory getOutputFactory() {
        return _factory;
    }
}
  • AppendCollector在构造器里头创建了OperationOutputFactory,其emit方法也是提取OperationOutputFields,然后挨个调用triContext.getReceivers()的execute方法;如果each之后没有其他操作,那么AppendCollector的triContext.getReceivers()就为空

小结

  • WindowTridentProcessor里头使用的是FreshCollector,WindowTridentProcessor在finishBatch的时候,会从TridentWindowManager提取window创建的pendingTriggers(提取之后会将其数据从pendingTriggers移除),里头包含了窗口累积的数据,然后使用FreshCollector发射这些数据,默认第一个value为TriggerInfo,第二个value就是窗口累积发射的values
  • FreshCollector的emit方法首先使用TridentTupleView.FreshOutputFactory根据selfOutputFields(第一个field固定为_task_info,之后的几个field为用户在window方法定义的functionFields)构建TridentTupleView,然后挨个调用_triContext.getReceivers()的execute方法
  • 后续的receivers中有一个ProjectedProcessor,用于根据window方法定义的functionFields重新提取的TridentTupleView,它的execute方法也类似FreshCollector.emit方法,先提取所需字段构造TridentTupleView,然后挨个调用_triContext.getReceivers()的execute方法(比如EachProcessor.execute)
  • EachProcessor使用的collector为AppendCollector,它的emit方法也类似FreshCollector的emit方法,先进行字段提取构造TridentTupleView,然后挨个调用_triContext.getReceivers()的execute方法
  • FreshCollector的emit方法与ProjectedProcessor的execute方法以及AppendCollector的emit方法都非常类似,首先是使用Factory提取所需字段构建TridentTupleView,然后挨个调用triContext.getReceivers()的execute方法;当一个triContext没有receiver的时候,tuple的传递也就停止了

doc

  • Windowing Support in Core Storm
本文参与 腾讯云自媒体分享计划,分享自微信公众号。
原始发表:2018-11-14,如有侵权请联系 cloudcommunity@tencent.com 删除

本文分享自 码匠的流水账 微信公众号,前往查看

如有侵权,请联系 cloudcommunity@tencent.com 删除。

本文参与 腾讯云自媒体分享计划  ,欢迎热爱写作的你一起参与!

评论
登录后参与评论
0 条评论
热度
最新
推荐阅读
目录
  • 实例
  • WindowTridentProcessor
  • FreshCollector
    • TridentTupleView.FreshOutputFactory
      • ValuePointer
      • ProjectedProcessor
        • TridentTupleView.ProjectionFactory
          • EachProcessor
            • AppendCollector
            • 小结
            • doc
            领券
            问题归档专栏文章快讯文章归档关键词归档开发者手册归档开发者手册 Section 归档