Hbase源碼研究（四）------put（2）

   在前面的文章裏粗略地分析了hbase的put的源碼，但是對提交那塊分析的不夠深入.........現在把提交那塊拿出來再分析一下。

   書接上文，前面說了在HconnectionManager 其中有一步是提交請求，調用瞭如下代碼

    private <R> Callable<MultiResponse> createCallable(final HRegionLocation loc,
        final MultiAction<R> multi, final byte [] tableName) {
      final HConnection connection = this;
      return new Callable<MultiResponse>() {
       public MultiResponse call() throws IOException {
         return getRegionServerWithoutRetries(
             new ServerCallable<MultiResponse>(connection, tableName, null) {
               public MultiResponse call() throws IOException {
                 return server.multi(multi);
               }
               @Override
               public void connect(boolean reload) throws IOException {
                 server =
                   connection.getHRegionConnection(loc.getHostname(), loc.getPort());
               }
             }
         );
       }
     };
   }

跟進這個server.multi(multi); 方法，來到HRegionServer的multi這個方法

public <R> MultiResponse multi(MultiAction<R> multi) throws IOException {
    checkOpen();
    MultiResponse response = new MultiResponse();
    for (Map.Entry<byte[], List<Action<R>>> e : multi.actions.entrySet()) {
      byte[] regionName = e.getKey();
      List<Action<R>> actionsForRegion = e.getValue();
      // sort based on the row id - this helps in the case where we reach the
      // end of a region, so that we don't have to try the rest of the
      // actions in the list.
      Collections.sort(actionsForRegion);
      Row action;
      List<Action<R>> puts = new ArrayList<Action<R>>();
      for (Action<R> a : actionsForRegion) {
        action = a.getAction();
        int originalIndex = a.getOriginalIndex();

        try {
          if (action instanceof Delete) {
            delete(regionName, (Delete) action);
            response.add(regionName, originalIndex, new Result());
          } else if (action instanceof Get) {
            response.add(regionName, originalIndex, get(regionName, (Get) action));
          } else if (action instanceof Put) {
            puts.add(a);  // wont throw.
          } else if (action instanceof Exec) {
            ExecResult result = execCoprocessor(regionName, (Exec)action);
            response.add(regionName, new Pair<Integer, Object>(
                a.getOriginalIndex(), result.getValue()
            ));
          } else {
            LOG.debug("Error: invalid Action, row must be a Get, Delete, " +
                "Put or Exec.");
            throw new DoNotRetryIOException("Invalid Action, row must be a " +
                "Get, Delete or Put.");
          }
        } catch (IOException ex) {
          response.add(regionName, originalIndex, ex);
        }
      }

      // We do the puts with result.put so we can get the batching efficiency
      // we so need. All this data munging doesn't seem great, but at least
      // we arent copying bytes or anything.
      if (!puts.isEmpty()) {
        try {
          HRegion region = getRegion(regionName);

          if (!region.getRegionInfo().isMetaTable()) {
            this.cacheFlusher.reclaimMemStoreMemory();
          }

          List<Pair<Put,Integer>> putsWithLocks =
              Lists.newArrayListWithCapacity(puts.size());
          for (Action<R> a : puts) {
            Put p = (Put) a.getAction();

            Integer lock;
            try {
              lock = getLockFromId(p.getLockId());
            } catch (UnknownRowLockException ex) {
              response.add(regionName, a.getOriginalIndex(), ex);
              continue;
            }
            putsWithLocks.add(new Pair<Put, Integer>(p, lock));
          }

          this.requestCount.addAndGet(puts.size());

          OperationStatus[] codes =
              region.put(putsWithLocks.toArray(new Pair[]{}));

          for( int i = 0 ; i < codes.length ; i++) {
            OperationStatus code = codes[i];

            Action<R> theAction = puts.get(i);
            Object result = null;

            if (code.getOperationStatusCode() == OperationStatusCode.SUCCESS) {
              result = new Result();
            } else if (code.getOperationStatusCode()
                == OperationStatusCode.BAD_FAMILY) {
              result = new NoSuchColumnFamilyException(code.getExceptionMsg());
            }
            // FAILURE && NOT_RUN becomes null, aka: need to run again.

            response.add(regionName, theAction.getOriginalIndex(), result);
          }
        } catch (IOException ioe) {
          // fail all the puts with the ioe in question.
          for (Action<R> a: puts) {
            response.add(regionName, a.getOriginalIndex(), ioe);
          }
        }
      }
    }
    return response;
  }

又是一個冗長的方法（囧......）

首先處理action 還是按region來處理的（一個region 一個 region順序來處理的）。

在一個region範圍內，首先把所有action 按照row-id 排序，並排除掉不屬於該region的action。

接下來判斷 action 的類型 ， 如果是put類型的話 就放到新new的一個數組裏。

程序再往後，對裝有put的list做處理。

首先

List<Pair<Put,Integer>> putsWithLocks = Lists.newArrayListWithCapacity(puts.size());

先new 了一個存有put和鎖的數組，針對於每一個put都去調用 getLockFromId 這個方法，獲得鎖（Integer類型），放入list中。

然後在requestcout裏計數

this.requestCount.addAndGet(puts.size());

再往後就看到了提交層.......

  OperationStatus[] codes =
              region.put(putsWithLocks.toArray(new Pair[]{}));

接下來的工作就簡單了，針對返回的codes 逐一進行遍歷，

if (code.getOperationStatusCode() == OperationStatusCode.SUCCESS) {
              result = new Result();
            } else if (code.getOperationStatusCode()
                == OperationStatusCode.BAD_FAMILY) {
              result = new NoSuchColumnFamilyException(code.getExceptionMsg());
            }
            // FAILURE && NOT_RUN becomes null, aka: need to run again.

            response.add(regionName, theAction.getOriginalIndex(), result);

如果成功了就new一個新的result放進去，如果是BAD_FAMILY這個錯，就new一個Exception ，否則的話就是null。 這樣上面這層就能按照返回類型進行重試判斷了。

最核心的提交層又沒分析........等下次吧！