Hbase region split源代码阅读笔记

客户端 
1. HbaseAdmin.split(final byte [] tableNameOrRegionName, 
      final byte [] splitPoint) 

这个方法首先判断参数是regionName还是tableName；如果是regionName则只分裂该region,如果是tableName则分裂该表下的所有region 

if (isRegionName(tableNameOrRegionName)) {//如果是regionName 
        // Its a possible region name. 
        Pair<HRegionInfo, HServerAddress> pair = 
          MetaReader.getRegion(ct, tableNameOrRegionName);//读出regionInfo和RegionServerAddress 
        if (pair == null || pair.getSecond() == null) { 
          LOG.info("No server in .META. for " + 
            Bytes.toStringBinary(tableNameOrRegionName) + "; pair=" + pair); 
        } else { 
          split(pair.getSecond(), pair.getFirst(), splitPoint);//调用方法分裂该region 
        } 
      } else {{//如果是tableName 
        List<Pair<HRegionInfo, HServerAddress>> pairs = 
          MetaReader.getTableRegionsAndLocations(ct, 
              Bytes.toString(tableNameOrRegionName));//读出table下的一批region 
        for (Pair<HRegionInfo, HServerAddress> pair: pairs) { 
          // May not be a server for a particular row 
          if (pair.getSecond() == null) continue; 
          HRegionInfo r = pair.getFirst(); 
          // check for parents 
          if (r.isSplitParent()) continue; 
          if (splitPoint != null) { 
            // if a split point given, only split that particular region 
            if (!r.containsRow(splitPoint)) continue; 
          } 
          split(pair.getSecond(), pair.getFirst(), splitPoint);调用方法分裂这批region 

        } 

2. HbaseAdmin.split(final HServerAddress hsa, final HRegionInfo hri, 
      byte[] splitPoint) 

这只是个代理方法，调起regionserver的远程接口 
    HRegionInterface rs = this.connection.getHRegionConnection(hsa); 
    rs.splitRegion(hri, splitPoint); 

RegionServer服务端 
3.HRegionServer.splitRegion(HRegionInfo regionInfo, byte[] splitPoint) 

    checkOpen();//检查regionserver是否停止 
    HRegion region = getRegion(regionInfo.getRegionName());//拿到在线region 
    region.flushcache();//写memstore到hfile 
    region.forceSplit(splitPoint);//设置了两个值this.splitRequest=true    和 splitPoint = sp 
    // force a compaction, split will be side-effect 
    // TODO: flush/compact/split refactor will make it trivial to do this 
    // sync/async (and won't require us to do a compaction to split!) 
    compactSplitThread.requestCompaction(region, "User-triggered split", 

        CompactSplitThread.PRIORITY_USER);//异步发起一个region的compact操作

转自：http://uestzengting.iteye.com/blog/1294550