代码如下，详细请参见注释。

class Transaction { //class Transaction
  //结构体Write
  struct Write {
    Row row;        //行
    Column col;     //列
    string value;   //列值
  };//Write Struct 
  vector<Write> writes_;//数据缓存Write
  int start_ts_;//事务开始时间

  Transaction() : start_ts_(oracle.GetTimestamp()) {} //构造函数，初始化变量start_ts_
  
  /*
    输入：Write结构体
    输出：无
    实现：简单的把Write对象（列值）push到Vector中
  */
  void Set(Write w) {//Set函数
    writes_.push_back(w);   
  }
  
  /*
    输入：Row-行标识，Column-列标识
    输出：value-列值，成功返回true，失败（如没有获取值）返回false
  */
  bool Get(Row row, Column c, string* value) {
    while (true) {
      //Bigtable提供的行级事务
      bigtable::Txn T = bigtable::StartRowTransaction(row);
      // Check for locks that signal concurrent writes.
      // 检查是否存在并发事务在写数据
      // 注：SI的特点是写不阻塞读，读不阻塞写，但在这里却需要等待？
      //     原因是SI保证读到的是事务开始(start_ts)之前已提交的数据，
      //     存在锁意味着写操作未完成且该操作的commit_ts可能在事务开始之前，
      //     但需要在写入之后才能知道是否在start_ts之前，因此需要等待
      if (T.Read(row, c+"lock", [0, start_ts_])) { //判断[0, start_ts_]内是否存在lock？
        // There is a pending lock; try to clean it and wait
        // 仍存在lock，等待
        BackoffAndMaybeCleanupLock(row, c);
        continue;
      }
      // Find the latest write below our start timestamp.
      //读取最近已提交的数据版本
      latest write = T.Read(row, c+"write", [0, start_ts_]);
      if (!latest_write.found())
        //没有数据，返回false
        return false; // no data
      //从Column+write中获取start_ts
      int data_ts = latest_write.start_timestamp();
      //获取真正的数据：Row+Column(column+"data")+start_ts
      *value = T.Read(row, c+"data", [data_ts, data_ts]);
      return true;
    }
  }
  
  // Prewrite tries to lock cell w, returning false in case of conflict.
  // 预写入（理论基础：通过意向表缓存数据，执行延迟更新）
  /*    
    输入：Write结构体，Write主节点
    输出：成功返回true，失败返回false
  */
  bool Prewrite(Write w, Write primary) {
    //获取列
    Column c = w.col;
    //启动Bigtable行事务
    bigtable::Txn T = bigtable::StartRowTransaction(w.row);
    // Abort on writes after our start timestamp ...
    // 存在比事务启动时间start_ts更大的值,存在ww冲突,按照FUW原则,本事务回滚
    if (T.Read(w.row, c+"write", [start_ts_, ∞]))
      return false;
    // ... or locks at any timestamp.
    // 存在锁,说明未完成的写,即存在ww冲突,且其他事务比本事务更"早"获得锁,本事务回滚
    if (T.Read(w.row, c+"lock", [0, ∞]))
      return false;

    //校验完毕,可以写数据
    //写入数据:key=Row+Column(data)+start_ts,value=需写入的值
    T.Write(w.row, c+"data", start_ts_, w.value);
    //上锁,key=Row+Column(lock)+start_ts,value=主节点的行&列
    T.Write(w.row, c+"lock", start_ts_,
      {primary.row, primary.col}); // The primary’s location.
    //执行提交操作
    return T.Commit();
  }

  //提交操作
  /*
    输入：无
    输出：成功返回true，失败返回false
  */
  bool Commit() {
    // The primary’s location.
    // 数组writes_的第一个元素为主节点
    Write primary = writes_[0];
    // 除第一个元素外，其他为从节点
    vector<Write> secondaries(writes_.begin()+1, writes_.end());
    //主节点预写入失败
    if (!Prewrite(primary, primary))
      return false;
    //遍历从节点，执行预写入，一个节点不成功则全部失败
    for (Write w : secondaries)
      if (!Prewrite(w, primary))
        return false;

    //获取事务提交时间戳
    int commit_ts = oracle_.GetTimestamp();
    // Commit primary first.
    // 主节点首先提交
    Write p = primary;
    //启动Bigtable事务
    bigtable::Txn T = bigtable::StartRowTransaction(p.row);
    //谨慎起见，判断是否存在锁（本事务,start_ts唯一），避免重复写入
    if (!T.Read(p.row, p.col+"lock", [start_ts_, start_ts_]))
      return false; // aborted while working
    //写入：key=Row+Column(write)+commit_ts,value=start_ts,实际的值在key=Row+Column(data)+start_ts中
    T.Write(p.row, p.col+"write", commit_ts, start_ts_); // Pointer to data written at start_ts_.
    //删除锁
    T.Erase(p.row, p.col+"lock", commit_ts);
    //Bigtable事务提交
    if (!T.Commit())
      return false; // commit point
    // Second phase: write out write records for secondary cells.
    //遍历从节点，写key=Row+Column(write)+commit_ts,value=start_ts,同时删除锁
    for (Write w : secondaries) {
      bigtable::Write(w.row, w.col+"write", commit_ts, start_ts_);
      bigtable::Erase(w.row, w.col+"lock", commit_ts);
    }
    return true;
  }
} // class Transaction