在mapreduce中读取DistributedCache中的HAR文件

Question

我已经编写了一个oozie工作流，它创建HAR归档，然后运行需要从该归档读取数据的MR-job。1.创建归档文件2.作业运行时，mapper在分布式缓存中看到归档文件。3.？我怎样才能读懂这篇文章呢？从这个归档中逐行读取数据的API是什么(我的har是一批多个以新行分隔的文本文件)。注意:当我处理存储在DistirubtedCache中的常用文件(不是HAR存档)时，它工作得很好。我在尝试从HAR读取数据时遇到问题。

下面是一个代码片段：

    InputStream inputStream;
    String cachedDatafileName = System.getProperty(DIST_CACHE_FILE_NAME);
    LOG.info(String.format("Looking for[%s]=[%s] in DistributedCache",DIST_CACHE_FILE_NAME, cachedDatafileName));

    URI[] uris = DistributedCache.getCacheArchives(getContext().getConfiguration());
    URI uriToCachedDatafile = null;
    for(URI uri : uris){
        if(uri.toString().endsWith(cachedDatafileName)){
            uriToCachedDatafile = uri;
            break;
        }
    }
    if(uriToCachedDatafile == null){
        throw new RuntimeConfigurationException(String.format("Looking for[%s]=[%s] in DistributedCache failed. There is no such file",
                DIST_CACHE_FILE_NAME, cachedDatafileName));
    }

    Path pathToFile = new Path(uriToCachedDatafile);
    LOG.info(String.format("[%s] has been found. Uri is: [%s]. The path is:[%s]",cachedDatafileName, uriToCachedDatafile, pathToFile));

    FileSystem fileSystem =  pathToFile.getFileSystem(getContext().getConfiguration());
    HarFileSystem harFileSystem = new HarFileSystem(fileSystem);
    inputStream = harFileSystem.open(pathToFile); //NULL POINTER EXCEPTION IS HERE!
    return inputStream;

Answer 1

protected InputStream getInputStreamToDistCacheFile() throws IOException{
        InputStream inputStream;
        String cachedDatafileName = System.getProperty(DIST_CACHE_FILE_NAME);
        LOG.info(String.format("Looking for[%s]=[%s] in DistributedCache",DIST_CACHE_FILE_NAME, cachedDatafileName));

        URI[] uris = DistributedCache.getCacheArchives(getContext().getConfiguration());
        URI uriToCachedDatafile = null;
        for(URI uri : uris){
            if(uri.toString().endsWith(cachedDatafileName)){
                uriToCachedDatafile = uri;
                break;
            }
        }
        if(uriToCachedDatafile == null){
            throw new RuntimeConfigurationException(String.format("Looking for[%s]=[%s] in DistributedCache failed. There is no such file",
                    DIST_CACHE_FILE_NAME, cachedDatafileName));
        }

        //Path pathToFile = new Path(uriToCachedDatafile +"/stf/db_bts_stf.txt");
        Path pathToFile = new Path("har:///"+"home/ssa/devel/megalabs/kyc-solution/kyc-mrjob/target/test-classes/GSMCellSubscriberHomeIntersectionJobDescriptionClusterMRTest/in/gsm_cell_location_stf.har" +"/stf/db_bts_stf.txt");
        //Path pathToFile = new Path(("har://home/ssa/devel/megalabs/kyc-solution/kyc-mrjob/target/test-classes/GSMCellSubscriberHomeIntersectionJobDescriptionClusterMRTest/in/gsm_cell_location_stf.har"));

        LOG.info(String.format("[%s] has been found. Uri is: [%s]. The path is:[%s]",cachedDatafileName, uriToCachedDatafile, pathToFile));
        FileSystem harFileSystem = pathToFile.getFileSystem(context.getConfiguration());
        FSDataInputStream fin = harFileSystem.open(pathToFile);
        LOG.info("fin: " + fin);
//        FileSystem fileSystem =  pathToFile.getFileSystem(getContext().getConfiguration());
//        HarFileSystem harFileSystem = new HarFileSystem(fileSystem);
//        harFileSystem.exists(new Path("har://home/ssa/devel/mycompany/my-solution/my-mrjob/target/test-classes/HomeJobDescriptionClusterMRTest/in/locations.har"));
//        LOG.info("harFileSystem.exists(pathToFile):"+ harFileSystem.exists(pathToFile));
//        harFileSystem.initialize(uriToCachedDatafile, context.getConfiguration());



        FileStatus[] statuses = harFileSystem.listStatus(new Path("har:///"+"har://home/ssa/devel/mycompany/my-solution/my-mrjob/target/test-classes/HomeJobDescriptionClusterMRTest/in/locations.har"));
        for(FileStatus fileStatus : statuses){
            LOG.info("fileStatus isDir"+fileStatus.isDirectory() +" len:" + fileStatus.getLen());
        }

//        String tmpPathToFile = "har:///"+pathToFile.toString(); //+"/stf/db_bts_stf.txt";
//        Path tmpPath = new Path(tmpPathToFile);
//        LOG.info("KILL ME PATH TO FILE IN ARCHIVE: " +tmpPath);
//        inputStream = harFileSystem.open(tmpPath);
//        return inputStream;
        return fin;
    }

如你所见，这太可怕了。您已经手动读取了存储在归档中的索引文件，并使用索引文件元数据重建路径。如果您知道存档中存储的文件的确切名称(如我的示例中所示)，则可以手动构建路径。

这并不方便，我确实期望像Zip->zipEntry这样的东西，当你可以在不知道它的结构的情况下迭代存档的条目。