1. 引言
在上一篇中,我们简单的了解了爬虫的工作流程,也简单的实现了一个爬虫,并且在文末简单分析了目前存在的问题。这一篇博客将会对上一篇分析出的问题,给出改进方法。我们将从以下几个方面加以改进。
2. 改进
我们首先利用Bloom Filet来改进UrlQueue中的visitedSet。
在上一篇中,我们使用visitedSet(HashSet)来存放已经访问过的url。之所以使用HashSet是因为我们需要不断的插入url到visitedSet中,并且还需要频繁地判断某个url是否在其中,而采用Hash Table,在平均情况下,所有的字典操作在O(1)时间内都能完成(具体分析请看散列表(hash table)——算法导论(13))。但不足之处在于我们需要花费大量的内存空间去维护hash table,我们是否可以减小它的空间复杂度呢?
从visitedSet的作用入手,它只是用来判断某个url是否被包含在它内部,仅此而已。因此完全没有必要保存每个url的完整信息,保存指纹信息即可。这时我们可想到常用的md5和sha1摘要算法。但尽管对url做了压缩,我们还是需要去保存压缩后的信息。我们还有没有更好的方法呢?这时,Bloom Filter就派上用场了(关于Bloom Filter的介绍及实现在散列表(hash table)——算法导论(13)中的最后一小节)。
我们再来使用Berkeley DB改进我们UrlQueue中的unvisitedList。
Berkeley DB是一个嵌入式数据库系统,简单、小巧、性能高(简单小巧没得说,至于性能,没验证过)。关于Berkeley DB的下载和使用请到其官网:http://www.oracle.com/technetwork/database/database-technologies/berkeleydb/overview/index.html
使用Berkeley DB后,我们会将从页面解析出的url直接存入DB中,而unvisitedList只是作为从DB中取url时的缓冲池。即我们会开启一个线程以一定的频率从DB中读取一定数量的url到unvisitedList中,执行页面请求的线程还是从unvisitedList读取url。
最后我们引入多线程来提高爬虫的效率。
多线程的关键在于同步与通信。这些内容请自行百度。
改进后的整个结构图如下:
3. 实现
最后我们给出改进后的代码:
① 首先是改进后的UrlQueue.java,我们重命名为BloomQueue.java(其中的BloomFilter类,在散列表(hash table)——算法导论(13)中可找到)
public class BloomQueue<T> { private BloomFilter<T> bloomFilter; private LinkedBlockingQueue<T> visitedList; private AtomicInteger flowedCount; private int queueCapacity; public BloomQueue() { this(0.000001, 10000000, 500); } public BloomQueue(double falsePositiveProbability, int filterCapacity, int queueCapacity) { this.queueCapacity = queueCapacity; bloomFilter = new BloomFilter<>(falsePositiveProbability, filterCapacity); visitedList = new LinkedBlockingQueue<>(queueCapacity); flowedCount = new AtomicInteger(0); } /** * 入队(当无法入队时,默认阻塞3秒) * * @param t * @return */ public boolean enqueue(T t) { return enqueue(t, 3000); } /** * 入队 * * @param t * @param timeout * 单位为毫秒 */ public boolean enqueue(T t, long timeout) { try { boolean result = visitedList.offer(t, timeout, TimeUnit.MILLISECONDS); if (result) { bloomFilter.add(t); flowedCount.getAndIncrement(); } return result; } catch (InterruptedException e) { e.printStackTrace(); } return false; } /** * 出队(当队列为空时,默认会阻塞3秒) * * @return */ public T dequeue() { return dequeue(3000); } /** * 出队 * * @param timeout * 单位为毫秒 * @return */ public T dequeue(long timeout) { try { return visitedList.poll(timeout, TimeUnit.MILLISECONDS); } catch (InterruptedException e) { } return null; } /** * 当前是否包含 * * @return */ public boolean contains(T t) { return visitedList.contains(t); } /** * 曾经是否包含 * * @param t * @return */ public boolean contained(T t) { return bloomFilter.contains(t); } public boolean isEmpty() { return visitedList.isEmpty(); } public boolean isFull() { return visitedList.size() == queueCapacity; } public int size() { return visitedList.size(); } public int flowedCount() { return flowedCount.get(); } @Override public String toString() { return visitedList.toString(); } }
② 然后我们对Berkeley DB做一个简单的封装,便于使用。
public class DBHelper<T> { public static final String DEFAULT_DB_DIR = "C:/Users/Administrator/Desktop/db/"; public static final String DEFAULT_Entity_Store = "EntityStore"; public Environment myEnv; public EntityStore store; public PrimaryIndex<Long, T> primaryIndex; public DBHelper(Class<T> clazz) { this(clazz, DEFAULT_DB_DIR, DEFAULT_Entity_Store, false); } public DBHelper(Class<T> clazz, String dbDir, String storeName, boolean isRead) { File dir = new File(dbDir); if (!dir.exists()) { dir.mkdirs(); } EnvironmentConfig envConfig = new EnvironmentConfig(); envConfig.setAllowCreate(!isRead); // Environment myEnv = new Environment(dir, envConfig); // StoreConfig StoreConfig storeConfig = new StoreConfig(); storeConfig.setAllowCreate(!isRead); // store store = new EntityStore(myEnv, storeName, storeConfig); // PrimaryIndex primaryIndex = store.getPrimaryIndex(Long.class, clazz); } public void put(T t) { primaryIndex.put(t); store.sync(); myEnv.sync(); } public EntityCursor<T> entities() { return primaryIndex.entities(); } public T get(long key) { return primaryIndex.get(key); } public void close() { if (store != null) { store.close(); } if (myEnv != null) { myEnv.cleanLog(); myEnv.close(); } } }
③ 接着我们写一个url的entity,便于存储。
import com.sleepycat.persist.model.Entity; import com.sleepycat.persist.model.PrimaryKey; @Entity public class Url { @PrimaryKey(sequence = "Sequence_Namespace") private long id; private String url; public Url() { } public Url(String url) { super(); this.url = url; } public long getId() { return id; } public void setId(long id) { this.id = id; } public String getUrl() { return url; } public void setUrl(String url) { this.url = url; } @Override public String toString() { return url; } public boolean isEmpty() { return url == null || url.isEmpty(); } }
④最后是我们的核心类CrawlerEngine。该类有两个内部类:Feeder和Fetcher。Feeder意为饲养员、进料器,负责向urlQueue中添加url;Fetcher意为抓取者,负责从urlQueue中取出url,进行请求,解析。其中用到的JsoupDownLoader类和上一篇一样,保持不变。
public class CrawlerEngine { public static final String DEFAULT_SAVE_DIR = "C:/Users/Administrator/Desktop/html/"; private static final long FEEDER_SLEEP_TIME = 10; private static final long FEEDER_MAX_WAIT_TIME = 3 * 1000;// 当DB中取不到url时,feeder最长等待时间(即如果等待该时间后,DB还是为空,则feeder结束工作) private static final int FEEDER_MAX_WAIT_COUNT = (int) (FEEDER_MAX_WAIT_TIME / FEEDER_SLEEP_TIME);// 当DB中取不到url时,feeder最长等待时间(即如果等待该时间后,DB还是为空,则feeder结束工作) private static final boolean LOG = false; private BloomQueue<Url> urlQueue; private ExecutorService fetcherPool; private int fetcherCount; private boolean running; private DBHelper<Url> dbHelper; private JsoupDownloader downloader; private String parseRegex; private String saveRegex; private String saveDir; private String saveName; private long maxCount = 1000; private long startTime; private long endTime = Long.MAX_VALUE; public CrawlerEngine() { this(20, DEFAULT_SAVE_DIR, null); } public CrawlerEngine(int fetcherCount, String saveDir, String saveName) { this.fetcherCount = fetcherCount; urlQueue = new BloomQueue<>(); fetcherPool = Executors.newFixedThreadPool(fetcherCount); dbHelper = new DBHelper<>(Url.class); downloader = JsoupDownloader.getInstance(); this.saveDir = saveDir; this.saveName = saveName; } public void startUp(String[] seeds) { if (running) { return; } running = true; startTime = System.currentTimeMillis(); for (String seed : seeds) { Url url = new Url(seed); urlQueue.enqueue(url); } for (int i = 0; i < fetcherCount; i++) { fetcherPool.execute(new Fetcher()); } new Feeder().start(); } public void shutdownNow() { running = false; fetcherPool.shutdown(); } public void shutdownAtTime(long time) { if (time > startTime) { endTime = time; } } public void shutdownDelayed(long delayed) { shutdownAtTime(startTime + delayed); } public void shutdownAtCount(long count) { maxCount = count; } private boolean isEnd() { return urlQueue.flowedCount() > maxCount || System.currentTimeMillis() > endTime; } private long currId = 1; private int currWaitCount; /** * 饲养员 * <p> * 从DB中获取一定数量的url到queue * </p> * * @author D.K * */ private class Feeder extends Thread { @Override public void run() { while (!isEnd() && running && currWaitCount != FEEDER_MAX_WAIT_COUNT) { try { sleep(FEEDER_SLEEP_TIME); if (urlQueue.isFull()) { log("Feeder", "队列已满"); continue; } Url url = dbHelper.get(currId); if (url == null) { currWaitCount++; log("Feeder", "url为null,currWaitCount = " + currWaitCount); } else { while (urlQueue.contained(url)) { currId++; url = dbHelper.get(currId); } if (url != null) { log("Feeder", "url准备入队"); urlQueue.enqueue(url); currId++; log("Feeder", "url已经入队,currId = " + currId); currWaitCount = 0; } } } catch (Exception e) { e.printStackTrace(); } } log("Feeder", "执行结束..."); while (true) { try { sleep(100); log("Feeder", "等待Fetcher结束..."); } catch (InterruptedException e) { } if (urlQueue.isEmpty()) { shutdownNow(); System.out.println(">>>>>>>>>>>>爬取结束,共请求了" + urlQueue.flowedCount() + "个页面,用时" + (System.currentTimeMillis() - startTime) + "毫秒<<<<<<<<<<<<"); return; } } } } /** * 抓取者 * <p> * 从queue中取出url,下载页面,解析页面,并把解析出的新的url添加到DB中 * </p> * * @author D.K * */ private class Fetcher implements Runnable { @Override public void run() { while (!isEnd() && (running || !urlQueue.isEmpty())) { log("Fetcher", "开始从队列获取url,size=" + urlQueue.size()); Url url = urlQueue.dequeue(); if (url == null) { log("Fetcher", "url为null"); continue; } log("Fetcher", "取出了url"); Document doc = downloader.downloadPage(url.getUrl()); Set<String> urlSet = downloader.parsePage(doc, parseRegex); downloader.savePage(doc, saveDir, saveName, saveRegex); for (String str : urlSet) { Url u = new Url(str); if (!urlQueue.contained(u)) { dbHelper.put(u); } } } } } private void log(String talker, String content) { if (LOG) { System.out.println("[" + talker + "] " + content); } } public String getParseRegex() { return parseRegex; } public void setParseRegex(String parseRegex) { this.parseRegex = parseRegex; } public String getSaveRegex() { return saveRegex; } public void setSaveRegex(String saveRegex) { this.saveRegex = saveRegex; } public void setSavePath(String saveDir, String saveName) { this.saveDir = saveDir; this.saveName = saveName; } }
我们采用上一篇的测试例子来做同样的测试,以检验我们优化后的效果。下面是测试代码:
public class Test { public static void main(String[] args) throws InterruptedException { CrawlerEngine crawlerEngine = new CrawlerEngine(); crawlerEngine.setParseRegex("(http://www.cnblogs.com/artech/p|http://www.cnblogs.com/artech/default|http://www.cnblogs.com/artech/archive/\d{4}/\d{2}/\d{2}/).*"); crawlerEngine.setSaveRegex("(http://www.cnblogs.com/artech/p|http://www.cnblogs.com/artech/archive/\d{4}/\d{2}/\d{2}/).*"); crawlerEngine.startUp(new String[] { "http://www.cnblogs.com/artech/" }); crawlerEngine.shutdownAtCount(1000); } }
下面是运行结果:
(4) 总结
对比我们上一篇中的测试时间61s,改进后用时14s,效率有明显的提升。
在下一篇中,我们要对整个代码再次进行小的优化,完善一些细节,如对请求状态码的处理,抽取出一些接口以降低代码之间的耦合度,增强灵活性。