HBase源码分析之HRegion上compact流程分析（三）

        在《HBase源码分析之HRegion上compact流程分析（二）》一文中，我们没有讲解真正执行合并的CompactionContext的compact()方法。现在我们来分析下它的具体实现。

        首先，CompactionContext表示合并的上下文信息，它只是一个抽象类，其compact()并没有实现，代码如下：

/**
   * Runs the compaction based on current selection. select/forceSelect must have been called.
   * @return The new file paths resulting from compaction.
   */
  public abstract List<Path> compact() throws IOException;

        那么，我们来找下它的实现类。它一共有两种实现类：DefaultCompactionContext和StripeCompaction，今天我们以DefaultCompactionContext为例来讲解。

        首先看下DefaultCompactionContext中compact()方法的实现：

    @Override
    public List<Path> compact() throws IOException {
      return compactor.compact(request);
    }

        这个compactor可以根据参数hbase.hstore.defaultengine.compactor.class配置，但是默认实现为DefaultCompactor。那么，接下来，我们看下它的实现：

  /**
   * Do a minor/major compaction on an explicit set of storefiles from a Store.
   * 在一个Store中明确的storefiles集合中执行一个minor或者major合并
   */
  public List<Path> compact(final CompactionRequest request) throws IOException {
    
	  // 从请求中获取文件详情fd，fd是FileDetails类型
	  FileDetails fd = getFileDetails(request.getFiles(), request.isAllFiles());
    
	  // 构造合并过程追踪器CompactionProgress
	  this.progress = new CompactionProgress(fd.maxKeyCount);

    // Find the smallest read point across all the Scanners.
	  // 找到scanners中的最小的可读点，实际上就是找到最小能够读取数据的点
    long smallestReadPoint = getSmallestReadPoint();

    List<StoreFileScanner> scanners;
    Collection<StoreFile> readersToClose;
    
    // 根据参数hbase.regionserver.compaction.private.readers确定是否使用私有readers
    if (this.conf.getBoolean("hbase.regionserver.compaction.private.readers", false)) {
      
      // clone all StoreFiles, so we'll do the compaction on a independent copy of StoreFiles,
      // HFileFiles, and their readers
      // 克隆所有的StoreFiles，以便我们将在StoreFiles、HFileFiles以及它们的readers等一个独立的副本上执行合并
      
      
      // 根据请求中待合并文件的数目创建一个StoreFile列表：readersToClose
      readersToClose = new ArrayList<StoreFile>(request.getFiles().size());
      
      // 将待合并文件复制一份加入readersToClose列表
      for (StoreFile f : request.getFiles()) {
        readersToClose.add(new StoreFile(f));
      }
      
      // 根据readersToClose列表，即待合并文件的副本创建文件浏览器FileScanners
      scanners = createFileScanners(readersToClose, smallestReadPoint);
    } else {
      // 创建空的列表readersToClose
      readersToClose = Collections.emptyList();
      
      // 根据实际请求中的待合并文件列表创建文件浏览器FileScanners
      scanners = createFileScanners(request.getFiles(), smallestReadPoint);
    }

    StoreFile.Writer writer = null;
    List<Path> newFiles = new ArrayList<Path>();
    boolean cleanSeqId = false;
    IOException e = null;
    try {
      InternalScanner scanner = null;
      try {
        /* Include deletes, unless we are doing a compaction of all files */
        
    	// 确定scan类型scanType：
    	// 如果compact请求是MAJOR或ALL_FILES合并，则scanType为COMPACT_DROP_DELETES；
    	// 如果compact请求是MINOR合并，则scanType为COMPACT_RETAIN_DELETES。
    	ScanType scanType =
            request.isAllFiles() ? ScanType.COMPACT_DROP_DELETES : ScanType.COMPACT_RETAIN_DELETES;
        
    	// 如果有协处理器，调用协处理器的preCreateCoprocScanner()方法
    	scanner = preCreateCoprocScanner(request, scanType, fd.earliestPutTs, scanners);
        if (scanner == null) {
          // 如果协处理器中未创建scanner，调用createScanner()方法创建一个
          scanner = createScanner(store, scanners, scanType, smallestReadPoint, fd.earliestPutTs);
        }
        
        // 如果有协处理器，调用协处理器的preCompact()方法
        scanner = postCreateCoprocScanner(request, scanType, scanner);
        if (scanner == null) {
          // NULL scanner returned from coprocessor hooks means skip normal processing.
          return newFiles;
        }
        // Create the writer even if no kv(Empty store file is also ok),
        // because we need record the max seq id for the store file, see HBASE-6059
        
        // 确定最小读取点smallestReadPoint
        if(fd.minSeqIdToKeep > 0) {
          smallestReadPoint = Math.min(fd.minSeqIdToKeep, smallestReadPoint);
          cleanSeqId = true;
        }

        // When all MVCC readpoints are 0, don't write them.
        // See HBASE-8166, HBASE-12600, and HBASE-13389.
        // 调用HStore的createWriterInTmp()方法，获取writer
        writer = store.createWriterInTmp(fd.maxKeyCount, this.compactionCompression, true,
          fd.maxMVCCReadpoint > 0, fd.maxTagsLength > 0);
        
        // 调用performCompaction()方法，执行合并
        boolean finished = performCompaction(scanner, writer, smallestReadPoint, cleanSeqId);
        
        // 如果没有完成合并
        if (!finished) {
          // 关闭writer
          writer.close();
          
          // 删除writer中的临时文件