nutch源码分析---1

nutch源码分析—inject

本章开始分析nutch 1.12版本的源码，nutch在爬取网页时分为inject、generate、fetch、parse、updatedb五个步骤，本章先来看inject命令，nutch官网教程给出的实例如下，
bin/nutch inject crawl/crawldb urls
urls目录中的文件seed.txt包含了其实的url地址。
编译nutch源码后，在目录runtime/local/bin/的nutch脚本里可以看到如下一段代码，

...

elif [ "$COMMAND" = "inject" ] ; then
  CLASS=org.apache.nutch.crawl.Injector
elif [ "$COMMAND" = "generate" ] ; then
  CLASS=org.apache.nutch.crawl.Generator
elif [ "$COMMAND" = "fetch" ] ; then
  CLASS=org.apache.nutch.fetcher.Fetcher
elif [ "$COMMAND" = "parse" ] ; then
  CLASS=org.apache.nutch.parse.ParseSegment
elif [ "$COMMAND" = "updatedb" ] ; then
  CLASS=org.apache.nutch.crawl.CrawlDb

...

exec "${EXEC_CALL[@]}" $CLASS "$@"

EXEC_CALL是执行Java程序的命令，因此对于inject命令，最终执行org.apache.nutch.crawl.Injector类的main函数。

Injector::main

  public static void main(String[] args) throws Exception {
    int res = ToolRunner.run(NutchConfiguration.create(), new Injector(), args);
    System.exit(res);
  }

ToolRunner是hadoop的一个工具，该段代码最终会调用Injector类的run函数，

Injector::main->Injector::run

  public int run(String[] args) throws Exception {

      ...

      inject(new Path(args[0]), new Path(args[1]), overwrite, update);

      ...
  }

  public void inject(Path crawlDb, Path urlDir, boolean overwrite,
      boolean update) throws IOException, ClassNotFoundException, InterruptedException {

    ...

    Configuration conf = getConf();
    conf.setLong("injector.current.time", System.currentTimeMillis());
    conf.setBoolean("db.injector.overwrite", overwrite);
    conf.setBoolean("db.injector.update", update);
    conf.setBoolean("mapreduce.fileoutputcommitter.marksuccessfuljobs", false);

    FileSystem fs = FileSystem.get(conf);
    Path current = new Path(crawlDb, CrawlDb.CURRENT_NAME);
    if (!fs.exists(current))
      fs.mkdirs(current);

    Path tempCrawlDb = new Path(crawlDb,
        "crawldb-" + Integer.toString(new Random().nextInt(Integer.MAX_VALUE)));

    Path lock = new Path(crawlDb, CrawlDb.LOCK_NAME);
    LockUtil.createLockFile(fs, lock, false);

    Job job = Job.getInstance(conf, "inject " + urlDir);
    job.setJarByClass(Injector.class);
    job.setMapperClass(InjectMapper.class);
    job.setReducerClass(InjectReducer.class);
    job.setOutputFormatClass(MapFileOutputFormat.class);
    job.setOutputKeyClass(Text.class);
    job.setOutputValueClass(CrawlDatum.class);
    job.setSpeculativeExecution(false);

    MultipleInputs.addInputPath(job, current, SequenceFileInputFormat.class);
    MultipleInputs.addInputPath(job, urlDir, KeyValueTextInputFormat.class);
    FileOutputFormat.setOutputPath(job, tempCrawlDb);

    job.waitForCompletion(true);
    CrawlDb.install(job, crawlDb);

  }

传入的参数crawlDb为crawl/crawldb
创建hadoop的Configuration，作相应的设置。
在crawl/crawldb下创建“current”、“crawldb-随机数”和“.locked”目录，其中“crawldb-随机数”为临时目录，后面会删除。
再接下来创建Job，设置Mapper和Reducer的处理类，并添加数据源为current目录里的数据和urls文件下的文本数据，然后调用其waitForCompletion函数被hadoop框架调用。
最后执行CrawlDb的install函数，替换old和current目录，并删除锁文件。

Job提交到hadoop框架后，会首先调用InjectMapper的map函数处理。

InjectMapper::map

    public void map(Text key, Writable value, Context context)
        throws IOException, InterruptedException {
      if (value instanceof Text) {
        String url = key.toString().trim();
        url = filterNormalize(url);
        if (url == null) {
          context.getCounter("injector", "urls_filtered").increment(1);
        } else {
          CrawlDatum datum = new CrawlDatum();
          datum.setStatus(CrawlDatum.STATUS_INJECTED);
          datum.setFetchTime(curTime);
          datum.setScore(scoreInjected);
          datum.setFetchInterval(interval);

          String metadata = value.toString().trim();
          if (metadata.length() > 0)
            processMetaData(metadata, datum, url);

          key.set(url);
          scfilters.injectedScore(key, datum);

          context.getCounter("injector", "urls_injected").increment(1);
          context.write(key, datum);
        }
      } else if (value instanceof CrawlDatum) {
        CrawlDatum datum = (CrawlDatum) value;
        String url = filterNormalize(key.toString());
        key.set(url);
        context.write(key, datum);
      }
    }

根据前面的分析，inject函数向hadoop框架注册了两个数据源，因此map函数分两种情况处理，map函数的参数key是对应的url地址，value则是url地址后面跟着的url信息。
当value是Text类型时，表示数据源是urls文件夹下的seed.txt文件，这种情况下，首先读取url地址，并调用filterNormalize函数对url规范化，得到统一的格式，接下来创建CrawlDatum，并调用processMetaData函数处理url信息，scfilters的类型为ScoringFilters，其injectedScore用来为url打分，再往下就调用hadoop的Context的write函数交由Reducer继续处理。
当value的类型是CrawlDatum时，表示之前已经对该url进行了处理，此时仅对url规范化，就继续交由Reducer处理了。
因此，无论数据源为何类型，map函数最终返回key为url地址，value为CrawlDatum的数据交由Reducer继续处理。

InjectMapper::map->processMetaData

    private void processMetaData(String metadata, CrawlDatum datum,
        String url) {

      String[] splits = metadata.split(TAB_CHARACTER);
      for (String split : splits) {
        int indexEquals = split.indexOf(EQUAL_CHARACTER);

        String metaname = split.substring(0, indexEquals);
        String metavalue = split.substring(indexEquals + 1);

        if (metaname.equals(nutchScoreMDName)) {
          datum.setScore(Float.parseFloat(metavalue));
        } else if (metaname.equals(nutchFetchIntervalMDName)) {
          datum.setFetchInterval(Integer.parseInt(metavalue));
        } else if (metaname.equals(nutchFixedFetchIntervalMDName)) {
          int fixedInterval = Integer.parseInt(metavalue);
          if (fixedInterval > -1) {
            datum.getMetaData().put(Nutch.WRITABLE_FIXED_INTERVAL_KEY,
                new FloatWritable(fixedInterval));
            datum.setFetchInterval(fixedInterval);
          }
        } else {
          datum.getMetaData().put(new Text(metaname), new Text(metavalue));
        }
      }
    }

TAB_CHARACTER的默认值是“\t”，EQUAL_CHARACTER的默认值是“=”，processMetaData函数根据TAB_CHARACTER提取每组url信息，每组url信息又通过等号划分属性名metaname和属性值metavalue ，然后将其设置进CrawlDatum中。

map函数处理完，hadoop框架继而调用InjectReducer的reduce函数继续处理，

InjectReducer::reduce

    public void reduce(Text key, Iterable<CrawlDatum> values, Context context)
        throws IOException, InterruptedException {

      for (CrawlDatum val : values) {
        if (val.getStatus() == CrawlDatum.STATUS_INJECTED) {
          injected.set(val);
          injected.setStatus(CrawlDatum.STATUS_DB_UNFETCHED);
          injectedSet = true;
        } else {
          old.set(val);
          oldSet = true;
        }
      }

      CrawlDatum result;
      if (injectedSet && (!oldSet || overwrite)) {
        result = injected;
      } else {
        result = old;

        if (injectedSet && update) {
          old.putAllMetaData(injected);
          old.setScore(injected.getScore() != scoreInjected
              ? injected.getScore() : old.getScore());
          old.setFetchInterval(injected.getFetchInterval() != interval
              ? injected.getFetchInterval() : old.getFetchInterval());
        }
      }
      context.write(key, result);
    }

reduce函数简而言之，要么覆盖之前某个url对应的CrawlDatum结构，要么只是通过putAllMetaData、setScore和setFetchInterval设置CrawlDatum中的对应信息，并不重写。

reduce函数执行成功后，就要向HDFS文件系统（前面注册的tempCrawlDb目录）中写入处理结果了。这里简单看一下CrawlDatum是如何写入的，CrawlDatum实现了hadoop的WritableComparable的write函数。

CrawlDatum::write

  public void write(DataOutput out) throws IOException {
    out.writeByte(CUR_VERSION); // store current version
    out.writeByte(status);
    out.writeLong(fetchTime);
    out.writeByte(retries);
    out.writeInt(fetchInterval);
    out.writeFloat(score);
    out.writeLong(modifiedTime);
    if (signature == null) {
      out.writeByte(0);
    } else {
      out.writeByte(signature.length);
      out.write(signature);
    }
    if (metaData != null && metaData.size() > 0) {
      out.writeBoolean(true);
      metaData.write(out);
    } else {
      out.writeBoolean(false);
    }
  }

再回头看CrawlDb的install函数，当hadoop处理完数据后，就会调用该函数进行最后的处理，

  public static void install(Job job, Path crawlDb) throws IOException {
    Configuration conf = job.getConfiguration();
    boolean preserveBackup = conf.getBoolean("db.preserve.backup", true);
    FileSystem fs = FileSystem.get(conf);
    Path old = new Path(crawlDb, "old");
    Path current = new Path(crawlDb, CURRENT_NAME);
    Path tempCrawlDb = org.apache.hadoop.mapreduce.lib.output.FileOutputFormat
        .getOutputPath(job);
    FSUtils.replace(fs, old, current, true);
    FSUtils.replace(fs, current, tempCrawlDb, true);
    Path lock = new Path(crawlDb, LOCK_NAME);
    LockUtil.removeLockFile(fs, lock);
    if (!preserveBackup && fs.exists(old)) {
      fs.delete(old, true);
    }
  }

  public static void replace(FileSystem fs, Path current, Path replacement,
      boolean removeOld) throws IOException {

    Path old = new Path(current + ".old");
    if (fs.exists(current)) {
      fs.rename(current, old);
    }

    fs.rename(replacement, current);
    if (fs.exists(old) && removeOld) {
      fs.delete(old, true);
    }
  }

  public static boolean removeLockFile(FileSystem fs, Path lockFile)
      throws IOException {
    return fs.delete(lockFile, false);
  }

install函数将原来的old目录替换为current目录，将current目录替换为最新的tempCrawlDb即“crawldb-随机数”目录，然后删除锁文件。