记一次mapreduce读取不到输入文件的问题

hdfs上输入文件所在包含两个目录，分别是： /20170503/shoplast/
/20170503/shop/
但是我想过滤掉shop，只把shoplast作为输入
故我实现了过滤器如下：

public  static class FileNameFilter implements PathFilter {

@Override
public boolean accept(Path path) {
if (path.getName().endsWith("last")) {
return true;
}
else {
return false;
}

}

}

然后mapreduce的输入设为 /20170503/*， 开始执行。。。
结果 Total input paths to process : 0 输入文件数为0！ 这什么鬼！

-----------------------------
看了源码之后感觉自己是个弱智啊。源码如下：

protected List<FileStatus> listStatus(JobContext job
) throws IOException {
List<FileStatus> result = new ArrayList<FileStatus>();
Path[] dirs = getInputPaths(job);
if (dirs.length == 0) {
throw new IOException("No input paths specified in job");
}

// get tokens for all the required FileSystems..
TokenCache.obtainTokensForNamenodes(job.getCredentials(), dirs,
job.getConfiguration());

// Whether we need to recursive look into the directory structure
boolean recursive = getInputDirRecursive(job);

List<IOException> errors = new ArrayList<IOException>();

// creates a MultiPathFilter with the hiddenFileFilter and the
// user provided one (if any).
List<PathFilter> filters = new ArrayList<PathFilter>();
filters.add(hiddenFileFilter);
PathFilter jobFilter = getInputPathFilter(job);
if (jobFilter != null) {
filters.add(jobFilter);
}
PathFilter inputFilter = new MultiPathFilter(filters);

for (int i=0; i < dirs.length; ++i) {
Path p = dirs[i];
FileSystem fs = p.getFileSystem(job.getConfiguration());
FileStatus[] matches = fs.globStatus(p, inputFilter);
if (matches == null) {
errors.add(new IOException("Input path does not exist: " + p));
} else if (matches.length == 0) {
errors.add(new IOException("Input Pattern " + p + " matches 0 files"));
} else {
for (FileStatus globStat: matches) {
if (globStat.isDirectory()) {
RemoteIterator<LocatedFileStatus> iter =
fs.listLocatedStatus(globStat.getPath());
while (iter.hasNext()) {
LocatedFileStatus stat = iter.next();
if (inputFilter.accept(stat.getPath())) {
if (recursive && stat.isDirectory()) {
addInputPathRecursively(result, fs, stat.getPath(),
inputFilter);
} else {
result.add(stat);
}
}
}
} else {
result.add(globStat);
}
}
}
}

if (!errors.isEmpty()) {
throw new InvalidInputException(errors);
}
LOG.info("Total input paths to process : " + result.size());
return result;
}

仔细看这段

for (FileStatus globStat: matches) {
if (globStat.isDirectory()) {
RemoteIterator<LocatedFileStatus> iter =
fs.listLocatedStatus(globStat.getPath());
while (iter.hasNext()) {
LocatedFileStatus stat = iter.next();
if (inputFilter.accept(stat.getPath())) {
if (recursive && stat.isDirectory()) {
addInputPathRecursively(result, fs, stat.getPath(),
inputFilter);
} else {
result.add(stat);
}
}
}
} else {
result.add(globStat);
}
}

以为过滤器是针对最终的输入文件名。如果输入的路径为目录，它会跟进里面的文件的
好吧，接下来修改下过滤器吧。

public  static class FileNameFilter implements PathFilter {

@Override
public boolean accept(Path path) {
if (path.getParent().getName().endsWith("last")) {
return true;
}
else {
return false;
}

}

}

再次运行，当当当

cause:org.apache.hadoop.mapreduce.lib.input.InvalidInputException: Input Pattern hdfs://20170503/* matches 0 files
Exception in thread "main" org.apache.hadoop.mapreduce.lib.input.InvalidInputException: Input Pattern hdfs://20170503/* matches 0 files

还是读取不到输入文件，这是什么原因呢？看源码吧，还是上面贴出的那个方法。

Path p = dirs[i];
FileSystem fs = p.getFileSystem(job.getConfiguration());
FileStatus[] matches = fs.globStatus(p, inputFilter);
if (matches == null) {
errors.add(new IOException("Input path does not exist: " + p));
} else if (matches.length == 0) {
errors.add(new IOException("Input Pattern " + p + " matches 0 files"));
}

不继续跟进globStatus方法了，想了解的自己去看源码吧。总之，这里是针对父目录的也应用了过滤器

结论：过滤器不光针对最终的文件，输入路径的父目录也要应用过滤器。