深入学习Heritrix解析Frontier(链接工厂)

Frontier是Heritrix最核心的组成部分之一,也是最复杂的组成部分.它主要功能是为处理链接的线程提供URL,并负责链接处理完成后的一些后续调度操作.并且为了提高效率,它在内部使用了Berkeley DB.本节将对它的内部机理进行详细解剖.

在Heritrix的官方文档上有一个Frontier的例子,虽然很简单,但是它却解释Frontier实现的基本原理.在这里就不讨论,有兴趣的读者可以参考相应文档.但是不得不提它的三个核心方法:
(1)next(int timeout):为处理线程提供一个链接.Heritrix的所有处理线程(ToeThread)都是通过调用该方法获取链接的.

(2)schedule(CandidateURI caURI):调度待处理的链接.

(3)finished(CrawlURI cURI):完成一个已处理的链接.

整体结构如下:

BdbMultipleWorkQueues:

它是对Berkeley DB的简单封装.在内部有一个Berkeley Database,存放所有待处理的链接.

package org.archive.crawler.frontier;
public class BdbMultipleWorkQueues 
{
//存放所有待处理的URL的数据库
    private Database pendingUrisDB = null;


 //由key获取一个链接
    public CrawlURI get(DatabaseEntry headKey)
    throws DatabaseException {
        DatabaseEntry result = new DatabaseEntry();
        
        // From Linda Lee of sleepycat:
        // "You want to check the status returned from Cursor.getSearchKeyRange
        // to make sure that you have OperationStatus.SUCCESS. In that case,
        // you have found a valid data record, and result.getData()
        // (called by internally by the binding code, in this case) will be
        // non-null. The other possible status return is
        // OperationStatus.NOTFOUND, in which case no data record matched
        // the criteria. "
        //由key获取相应的链接
        OperationStatus status = getNextNearestItem(headKey, result);
        CrawlURI retVal = null;
        if (status != OperationStatus.SUCCESS) {
            LOGGER.severe("See '1219854 NPE je-2.0 "
                    + "entryToObject'. OperationStatus "
                    + " was not SUCCESS: "
                    + status
                    + ", headKey "
                    + BdbWorkQueue.getPrefixClassKey(headKey.getData()));
            return null;
        }
        try {
            retVal = (CrawlURI)crawlUriBinding.entryToObject(result);
        } catch (RuntimeExceptionWrapper rw) {
            LOGGER.log(
                Level.SEVERE,
                "expected object missing in queue " +
                BdbWorkQueue.getPrefixClassKey(headKey.getData()),
                rw);
            return null; 
        }
        retVal.setHolderKey(headKey);
        return retVal;//返回链接
    }

    //从等处理列表获取一个链接
    protected OperationStatus getNextNearestItem(DatabaseEntry headKey,
            DatabaseEntry result) throws DatabaseException {
        Cursor cursor = null;
        OperationStatus status;
        try {
            //打开游标
            cursor = this.pendingUrisDB.openCursor(null, null);
            // get cap; headKey at this point should always point to 
            // a queue-beginning cap entry (zero-length value)
            status = cursor.getSearchKey(headKey, result, null);
            if(status!=OperationStatus.SUCCESS || result.getData().length > 0) {
                // cap missing
                throw new DatabaseException("bdb queue cap missing");
            }
            // get next item (real first item of queue)
            status = cursor.getNext(headKey,result,null);
        } finally { 
            if(cursor!=null) {
                cursor.close();
            }
        }
        return status;
    }

 /** *//**
     * Put the given CrawlURI in at the appropriate place. 
     * 添加URL到数据库
     * @param curi
     * @throws DatabaseException
     */
    public void put(CrawlURI curi, boolean overwriteIfPresent) 
    throws DatabaseException {
        DatabaseEntry insertKey = (DatabaseEntry)curi.getHolderKey();
        if (insertKey == null) {
            insertKey = calculateInsertKey(curi);
            curi.setHolderKey(insertKey);
        }
        DatabaseEntry value = new DatabaseEntry();
        crawlUriBinding.objectToEntry(curi, value);
        // Output tally on avg. size if level is FINE or greater.
        if (LOGGER.isLoggable(Level.FINE)) {
            tallyAverageEntrySize(curi, value);
        }
        OperationStatus status; 
        if(overwriteIfPresent) {
            //添加
            status = pendingUrisDB.put(null, insertKey, value);
        } else {
            status = pendingUrisDB.putNoOverwrite(null, insertKey, value);
        }
        if(status!=OperationStatus.SUCCESS) {
            LOGGER.severe("failed; "+status+ " "+curi);
        }
    }
    
}

 BdbWorkQueue:

代表一个链接队列,该队列中所有的链接都具有相同的键值.它实际上是通过调用BdbMultipleWorkQueues的get方法从等处理链接数据库中取得一个链接的.

package org.archive.crawler.frontier;

public class BdbWorkQueue extends WorkQueue
implements Comparable, Serializabl
{
//获取一个URL
    protected CrawlURI peekItem(final WorkQueueFrontier frontier)
    throws IOException {
        /**
         * 关键:从BdbFrontier中返回pendingUris
         */
        final BdbMultipleWorkQueues queues = ((BdbFrontier) frontier)
            .getWorkQueues();
        
        DatabaseEntry key = new DatabaseEntry(origin);
        CrawlURI curi = null;
        int tries = 1;
        while(true) {
            try {
                //获取链接
                curi = queues.get(key);
            } catch (DatabaseException e) {
                LOGGER.log(Level.SEVERE,"peekItem failure; retrying",e);
            }
         
    return curi;
    }
}

 WorkQueueFrontier:

实现了最核心的三个方法.

public CrawlURI next()
    throws InterruptedException, EndedException {
        while (true) { // loop left only by explicit return or exception
            long now = System.currentTimeMillis();

            // Do common checks for pause, terminate, bandwidth-hold
            preNext(now);
            
            synchronized(readyClassQueues) {
                int activationsNeeded = targetSizeForReadyQueues() - readyClassQueues.size();
                while(activationsNeeded > 0 && !inactiveQueues.isEmpty()) {
                    activateInactiveQueue();
                    activationsNeeded--;
                }
            }
                   
            WorkQueue readyQ = null;
            Object key = readyClassQueues.poll(DEFAULT_WAIT,TimeUnit.MILLISECONDS);
            if (key != null) {
                readyQ = (WorkQueue)this.allQueues.get(key);
            }
            if (readyQ != null) {
                while(true) { // loop left by explicit return or break on empty
                    CrawlURI curi = null;
                    synchronized(readyQ) {
                        /**取出一个URL,最终从子类BdbFrontier的
                         * pendingUris中取出一个链接
                         */
                        curi = readyQ.peek(this);                     
                        if (curi != null) {
                            // check if curi belongs in different queue
                            String currentQueueKey = getClassKey(curi);
                            if (currentQueueKey.equals(curi.getClassKey())) {
                                // curi was in right queue, emit
                                noteAboutToEmit(curi, readyQ);
                                //加入正在处理队列中
                                inProcessQueues.add(readyQ);
                                return curi; //返回
                            }
                            // URI's assigned queue has changed since it
                            // was queued (eg because its IP has become
                            // known). Requeue to new queue.
                            curi.setClassKey(currentQueueKey);
                            readyQ.dequeue(this);//出队列
                            decrementQueuedCount(1);
                            curi.setHolderKey(null);
                            // curi will be requeued to true queue after lock
                            //  on readyQ is released, to prevent deadlock
                        } else {
                            // readyQ is empty and ready: it's exhausted
                            // release held status, allowing any subsequent 
                            // enqueues to again put queue in ready
                            readyQ.clearHeld();
                            break;
                        }
                    }
                    if(curi!=null) {
                        // complete the requeuing begun earlier
                        sendToQueue(curi);
                    }
                }
            } else {
                // ReadyQ key wasn't in all queues: unexpected
                if (key != null) {
                    logger.severe("Key "+ key +
                        " in readyClassQueues but not allQueues");
                }
            }

            if(shouldTerminate) {
                // skip subsequent steps if already on last legs
                throw new EndedException("shouldTerminate is true");
            }
                
            if(inProcessQueues.size()==0) {
                // Nothing was ready or in progress or imminent to wake; ensure 
                // any piled-up pending-scheduled URIs are considered
                this.alreadyIncluded.requestFlush();
            }    
        }
    }


//将URL加入待处理队列
    public void schedule(CandidateURI caUri) {
        // Canonicalization may set forceFetch flag.  See
        // #canonicalization(CandidateURI) javadoc for circumstance.
        String canon = canonicalize(caUri);
        if (caUri.forceFetch()) {
            alreadyIncluded.addForce(canon, caUri);
        } else {
            alreadyIncluded.add(canon, caUri);
        }
    }

BdbFrontier:

继承了WorkQueueFrontier,是Heritrix唯一个具有实际意义的链接工厂.

package org.archive.crawler.frontier;
public class BdbFrontier extends WorkQueueFrontier implements Serializable 
{
    /** 所有待抓取的链接*/
    protected transient BdbMultipleWorkQueues pendingUris;

    //初始化pendingUris,父类为抽象方法
    protected void initQueue() throws IOException {
        try {
            this.pendingUris = createMultipleWorkQueues();
        } catch(DatabaseException e) {
            throw (IOException)new IOException(e.getMessage()).initCause(e);
        }
    }

   private BdbMultipleWorkQueues createMultipleWorkQueues()
    throws DatabaseException {
        return new BdbMultipleWorkQueues(this.controller.getBdbEnvironment(),
            this.controller.getBdbEnvironment().getClassCatalog(),
            this.controller.isCheckpointRecover());
    }
    protected BdbMultipleWorkQueues getWorkQueues() {
        return pendingUris;
    }


}

BdbUriUniqFilter:
实际上是一个过滤器,它用来检查一个要进入等待队列的链接是否已经被抓取过.

//添加URL
    protected boolean setAdd(CharSequence uri) {
        DatabaseEntry key = new DatabaseEntry();
        LongBinding.longToEntry(createKey(uri), key);
        long started = 0;
        
        OperationStatus status = null;
        try {
            if (logger.isLoggable(Level.INFO)) {
                started = System.currentTimeMillis();
            }
            //添加到数据库
            status = alreadySeen.putNoOverwrite(null, key, ZERO_LENGTH_ENTRY);
            if (logger.isLoggable(Level.INFO)) {
                aggregatedLookupTime +=
                    (System.currentTimeMillis() - started);
            }
        } catch (DatabaseException e) {
            logger.severe(e.getMessage());
        }
        if (status == OperationStatus.SUCCESS) {
            count++;
            if (logger.isLoggable(Level.INFO)) {
                final int logAt = 10000;
                if (count > 0 && ((count % logAt) == 0)) {
                    logger.info("Average lookup " +
                        (aggregatedLookupTime / logAt) + "ms.");
                    aggregatedLookupTime = 0;
                }
            }
        }
        //如果存在,返回false
        if(status == OperationStatus.KEYEXIST) {
            return false; // not added
        } else {
            return true;
        }
    }