SegmentReader Nutch 21 Mar 2014
前面我们看了一下Generate的流程，它是为Fetch产生相应的fetchlist，这里想介绍一下Segment的查看工具SegmentReader类。
1. 命令介绍
$ bin/nutch readseg
Usage: SegmentReader (-dump ... | -list ... | -get ...) [general options]General options:
-nocontent ignore content directory
-nofetch ignore crawl_fetch directory
-nogenerate ignore crawl_generate directory
-noparse ignore crawl_parse directory
-noparsedata ignore parse_data directory
-noparsetext ignore parse_text directory
这里用于下载segment的内容，把其转换成文本格式，后面可以加General options参数，看是不是过滤相应的目录
SegmentReader -dump <segment_dir> <output> [general options]
Dumps content of a <segment_dir> as a text file to <output>.
<segment_dir> name of the segment directory.
<output> name of the (non-existent) output directory.列出相应的segment信息
SegmentReader -list (<segment_dir1> ... | -dir <segments>) [general options]
List a synopsis of segments in specified directories, or all segments in
a directory <segments>, and print it on System.out
<segment_dir1> ... list of segment directories to process
-dir <segments> directory that contains multiple segments得到相对应的url的信息
SegmentReader -get <segment_dir> <keyValue> [general options]
Get a specified record from a segment, and print it on System.out.
<segment_dir> name of the segment directory.
<keyValue> value of the key (url).
Note: put double-quotes around strings with spaces.2. 导出segments
$ bin/nutch readseg -dump在本地运行其命令的结果如下：
$ bin/nutch readseg -dump db/segments/20110822105243/ output
SegmentReader: dump segment: db/segments/20110822105243
SegmentReader: done下载目录浏览
$ ls output
dump输出一部分下载信息
$ head output/dump
Recno:: 0
URL:: http://baike.baidu.com/
CrawlDatum::
Version: 7
Status: 67 (linked)
Fetch time: Mon Aug 22 10:58:21 EDT 2011
Modified time: Wed Dec 31 19:00:00 EST 1969
Retries since fetch: 0我们来看一下其源代码是怎么写的，这个shell命令最终是调用org.apache.nutch.segment.SegmentReader中的dump方法，如下是这个方法的主要源代码：
// 这里生成一个MP任务
JobConf job = createJobConf();
job.setJobName("read " + segment); // 查看General Options的参数，是否过滤相应的目录
if (ge) FileInputFormat.addInputPath(job, new Path(segment, CrawlDatum.GENERATE_DIR_NAME));
if (fe) FileInputFormat.addInputPath(job, new Path(segment, CrawlDatum.FETCH_DIR_NAME));
if (pa) FileInputFormat.addInputPath(job, new Path(segment, CrawlDatum.PARSE_DIR_NAME));
if (co) FileInputFormat.addInputPath(job, new Path(segment, Content.DIR_NAME));
if (pd) FileInputFormat.addInputPath(job, new Path(segment, ParseData.DIR_NAME));
if (pt) FileInputFormat.addInputPath(job, new Path(segment, ParseText.DIR_NAME)); // 输入的目录文件格式，这里是SequenceFileInputFormat
job.setInputFormat(SequenceFileInputFormat.class); // 相应的Map与Reducer操作
job.setMapperClass(InputCompatMapper.class); // 这里主要是把key转成UTF8格式，
job.setReducerClass(SegmentReader.class); // 把相应的value对象反序列化成Text类型
Path tempDir = new Path(job.get("hadoop.tmp.dir", "/tmp") + "/segread-" + new java.util.Random().nextInt());
fs.delete(tempDir, true);
FileOutputFormat.setOutputPath(job, tempDir); // 输出目录
job.setOutputFormat(TextOutputFormat.class); // output text
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(NutchWritable.class); // 输出的value类型，这里要注意一下，因为上面Reducer是SegmentReader，
// 其输出的<key,value>类型为<Text,Text>,而这里的value类型为NutchWritable，这里使用了强制类型转换。不知道这么做是为什么？
JobClient.runJob(job);
// concatenate the output
Path dumpFile = new Path(output, job.get("segment.dump.dir", "dump"));
// remove the old file
fs.delete(dumpFile, true);
FileStatus[] fstats = fs.listStatus(tempDir, HadoopFSUtil.getPassAllFilter());
Path[] files = HadoopFSUtil.getPaths(fstats);
PrintWriter writer = null;
int currentRecordNumber = 0;
// 这里主要是合并上面的临时文件到正式的目录文件中output/dump
// 并且加一些格式信息，使用append方法
if (files.length > 0) {
// create print writer with format
writer = new PrintWriter(new BufferedWriter(new OutputStreamWriter(fs.create(dumpFile))));
try {
for (int i = 0; i < files.length; i++) { // read tmp files
Path partFile = (Path) files[i];
try {
currentRecordNumber = append(fs, job, partFile, writer, currentRecordNumber);
} catch (IOException exception) {
if (LOG.isWarnEnabled()) {
LOG.warn("Couldn't copy the content of " + partFile.toString() + " into " + dumpFile.toString());
LOG.warn(exception.getMessage());
}
}
}
} finally {
writer.close();
}
}
fs.delete(tempDir); // 删除临时目录
if (LOG.isInfoEnabled()) { LOG.info("SegmentReader: done"); }3. 列出相应的segment信息
$ bin/nutch readseg -list在本地的运行结果如下：
列出单个Segment的信息
$ bin/nutch readseg -list db/segments/20110822105243/
NAME GENERATED FETCHER START FETCHER END FETCHED PARSED
20110822105243 1 2011-08-22T10:56:09 2011-08-22T10:56:09 1 1列出多个Segment的信息
$ bin/nutch readseg -list -dir db/segments/
NAME GENERATED FETCHER START FETCHER END FETCHED PARSED
20110822105243 1 2011-08-22T10:56:09 2011-08-22T10:56:09 1 1
20110825112318 9 ? ? ? ?
20110825112320 10 ? ? ? ?下面来看一下其源代码，调用的是SegmentReader中的list方法，主要代码如下：
这个list方法调用了另外一个getStats方法，得到单个Segment的信息
// 得到一个文件的读取器
SequenceFile.Reader[] readers = SequenceFileOutputFormat.getReaders(getConf(), new Path(segment, CrawlDatum.GENERATE_DIR_NAME));
long cnt = 0L;
Text key = new Text(); // 计算这个Segment有多少url
for (int i = 0; i < readers.length; i++) {
while (readers[i].next(key)) cnt++;
readers[i].close();
}
stats.generated = cut; // get generated url count(cnt)
// parse fetch dir 解析fetch目录，计算Fetch的开始与结束时间和fetch网页个数
// 主要这里的fetch目录的文件格式为MapFileOutputFormat
Path fetchDir = new Path(segment, CrawlDatum.FETCH_DIR_NAME);
if (fs.exists(fetchDir) && fs.getFileStatus(fetchDir).isDir()) {
cnt = 0L;
long start = Long.MAX_VALUE;
long end = Long.MIN_VALUE;
CrawlDatum value = new CrawlDatum();
MapFile.Reader[] mreaders = MapFileOutputFormat.getReaders(fs, fetchDir, getConf());
for (int i = 0; i < mreaders.length; i++) {
while (mreaders[i].next(key, value)) {
cnt++;
if (value.getFetchTime() < start) start = value.getFetchTime();
if (value.getFetchTime() > end) end = value.getFetchTime();
}
mreaders[i].close();
}
stats.start = start;
stats.end = end;
stats.fetched = cnt;
}
// parse parsed dir 解析parsed目录，得到解析成功与出错的网页个数
Path parseDir = new Path(segment, ParseData.DIR_NAME);
if (fs.exists(fetchDir) && fs.getFileStatus(fetchDir).isDir()) {
cnt = 0L;
long errors = 0L;
ParseData value = new ParseData();
MapFile.Reader[] mreaders = MapFileOutputFormat.getReaders(fs, parseDir, getConf());
for (int i = 0; i < mreaders.length; i++) {
while (mreaders[i].next(key, value)) {
cnt++;
if (!value.getStatus().isSuccess()) errors++;
}
mreaders[i].close();
}
stats.parsed = cnt;
stats.parseErrors = errors;
}4. 得到相对应的url的信息
bin/nutch readseg -get本机运行结果
$ bin/nutch readseg -get db/segments/20110822105243/ http://hi.baidu.com/
SegmentReader: get 'http://hi.baidu.com/'
Crawl Parse:: Version: 7
Status: 67 (linked)
Fetch time: Mon Aug 22 10:58:21 EDT 2011
Modified time: Wed Dec 31 19:00:00 EST 1969
Retries since fetch: 0
Retry interval: 2592000 seconds (30 days)
Score: 0.045454547
Signature: null
Metadata:下面我们来看一下它的源代码，它是调用SegmentReader中的get方法，主要代码如下：
public void get(final Path segment, final Text key, Writer writer, final Map<String, List<Writable>> results) throws Exception {
LOG.info("SegmentReader: get '" + key + "'"); // 这里使用的inner class来实现对于不同目录的异步读取
ArrayList<Thread> threads = new ArrayList<Thread>();
if (co) threads.add(new Thread() {
public void run() {
try { // 从MapFileOutputFormat格式的文件中找到相应的key的value值
List<Writable> res = getMapRecords(new Path(segment, Content.DIR_NAME), key);
// NOTE:有没有注意到这个results是一个HashMap，而在Java中这个HashMap不是线程安全的
// 在极端的情况下，会出现多个线程同时put数据，这里是不是应该把result改成线程安全的
// 如 Map result = Collections.synchronizedMap(new HashMap(...));
results.put("co", res);
} catch (Exception e) {
e.printStackTrace(LogUtil.getWarnStream(LOG));
}
}
});]
.....
if (pt) threads.add(new Thread() {
public void run() {
try {
List<Writable> res = getMapRecords(new Path(segment, ParseText.DIR_NAME), key);
results.put("pt", res);
} catch (Exception e) {
e.printStackTrace(LogUtil.getWarnStream(LOG));
}
}
});
Iterator<Thread> it = threads.iterator();
while (it.hasNext()) it.next().start(); // 运行所有线程
int cnt; // 这里用一个循环还查看线程是否运行结束
do {
cnt = 0;
try {
Thread.sleep(5000);
} catch (Exception e) {};
it = threads.iterator();
while (it.hasNext()) {
if (it.next().isAlive()) cnt++;
}
if ((cnt > 0) && (LOG.isDebugEnabled())) {
LOG.debug("(" + cnt + " to retrieve)");
}
} while (cnt > 0); // 把收集的结果输出到终端
for (int i = 0; i < keys.length; i++) {
List<Writable> res = results.get(keys[i][0]);
if (res != null && res.size() > 0) {
for (int k = 0; k < res.size(); k++) {
writer.write(keys[i][1]);
writer.write(res.get(k) + "\n");
}
}
writer.flush();
}
}
}5. 总结
这里大概介绍了一下bin/nutch readseg的使用，主要是来查看前面generate和后面的fetch,parse的结束是否有问题。