发现一个hdfs空文件导致任务失败问题

问题所在方法：

> com.alibaba.datax.plugin.reader.hdfsreader.DFSUtil.getHDFSAllFilesNORegex(String path, FileSystem hdfs)

方法内容：

private HashSet getHDFSAllFilesNORegex(String path, FileSystem hdfs) throws IOException {

Path listFiles = new Path(path);

// If the network disconnected, this method will retry 45 times
// each time the retry interval for 20 seconds
FileStatus stats[] = hdfs.listStatus(listFiles);

for (FileStatus f : stats) {
if (f.isDirectory()) {
LOG.info(String.format("[%s] 是目录, 递归获取该目录下的文件", f.getPath().toString()));
getHDFSAllFilesNORegex(f.getPath().toString(), hdfs);
} else if (f.isFile()) {
//这里没有对空文件作判断，会导致后续校验ORCFile时报数组越界错误，导致整个任务失败
addSourceFileByType(f.getPath().toString());
} else {
String message = String.format("该路径[%s]文件类型既不是目录也不是文件，插件自动忽略。",
f.getPath().toString());
LOG.info(message);
}
}
return sourceHDFSAllFilesList;
}

建议在上面判断中加入如下代码：

> if (f.getLen() == 0){
> String message = String.format("文件[%s]长度为0，将会跳过不作处理！", path);
> LOG.warn(message);
> continue;
> }

Start with com.alibaba.datax.plugin.reader.hdfsreader.DFSUtil.getHDFSAllFilesNORegex(String path, FileSystem hdfs) and trace how discovered files reach ORCFile validation. Verify the behavior for a zero-length HDFS file, then confirm that it is logged and skipped while non-empty files continue through normal processing.

Written by the indexing model from the issue text.