聊聊storm WindowTridentProcessor的FreshCollector

序

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