HBase源码分析 -- HBase Region 拆分(split)

代码版本：hbase-1.2.6

工程：hbase-server

类：org.apache.hadoop.hbase.regionserver.HRegion

需要解决的问题：

1、什么时候触发拆分？

2、拆分的策略是什么？

1、判断是否需要切分

方法： checkSplit

返回值： splitpoint

做了一些判断后，其实是调用:

byte[] ret = splitPolicy.getSplitPoint();

2、切分策略

org.apache.hadoop.hbase.regionserver.RegionSplitPolicy

/**
   * @return the key at which the region should be split, or null
   * if it cannot be split. This will only be called if shouldSplit
   * previously returned true.
   */
  protected byte[] getSplitPoint() {
    byte[] explicitSplitPoint = this.region.getExplicitSplitPoint();
    if (explicitSplitPoint != null) {
      return explicitSplitPoint;
    }
    List stores = region.getStores();

    byte[] splitPointFromLargestStore = null;
    long largestStoreSize = 0;
    for (Store s : stores) {
      byte[] splitPoint = s.getSplitPoint();
      long storeSize = s.getSize();
      if (splitPoint != null && largestStoreSize < storeSize) {
        splitPointFromLargestStore = splitPoint;
        largestStoreSize = storeSize;
      }
    }

    return splitPointFromLargestStore;
  }

从上边代码看如果 explicitSplitPoint不为空，则使用这个，再往上查是 forceSplit赋值的

如果explicitSplitPoint为空，则region.getStores() ，根据storeSize找到splitPoint

调用的是HStore的getSplitPoint方法：

@Override
  public byte[] getSplitPoint() {
    this.lock.readLock().lock();
    try {
      // Should already be enforced by the split policy!
      assert !this.getRegionInfo().isMetaRegion();
      // Not split-able if we find a reference store file present in the store.
      if (hasReferences()) {
        return null;
      }
      return this.storeEngine.getStoreFileManager().getSplitPoint();
    } catch(IOException e) {
      LOG.warn("Failed getting store size for " + this, e);
    } finally {
      this.lock.readLock().unlock();
    }
    return null;
  }

DefaultStoreFileManager

@Override
  public final byte[] getSplitPoint() throws IOException {
    if (this.storefiles.isEmpty()) {
      return null;
    }
    return StoreUtils.getLargestFile(this.storefiles).getFileSplitPoint(this.kvComparator);
  }

然后到SotreFile的：

 /**
   * Gets the approximate mid-point of this file that is optimal for use in splitting it.
   * @param comparator Comparator used to compare KVs.
   * @return The split point row, or null if splitting is not possible, or reader is null.
   */
  @SuppressWarnings("deprecation")
  byte[] getFileSplitPoint(KVComparator comparator) throws IOException {
    if (this.reader == null) {
      LOG.warn("Storefile " + this + " Reader is null; cannot get split point");
      return null;
    }
    // Get first, last, and mid keys.  Midkey is the key that starts block
    // in middle of hfile.  Has column and timestamp.  Need to return just
    // the row we want to split on as midkey.
    byte [] midkey = this.reader.midkey();
    if (midkey != null) {
      KeyValue mk = KeyValue.createKeyValueFromKey(midkey, 0, midkey.length);
      byte [] fk = this.reader.getFirstKey();
      KeyValue firstKey = KeyValue.createKeyValueFromKey(fk, 0, fk.length);
      byte [] lk = this.reader.getLastKey();
      KeyValue lastKey = KeyValue.createKeyValueFromKey(lk, 0, lk.length);
      // if the midkey is the same as the first or last keys, we cannot (ever) split this region.
      if (comparator.compareRows(mk, firstKey) == 0 || comparator.compareRows(mk, lastKey) == 0) {
        if (LOG.isDebugEnabled()) {
          LOG.debug("cannot split because midkey is the same as first or last row");
        }
        return null;
      }
      return mk.getRow();
    }
    return null;
  }