IBatis的分页研究

本文是在参阅了http://ivanl.iteye.com/blog/24739基础上完成的
在看JPetStore的代码时，发现它的分页处理主要是通过返回PaginatedList对象来完成的。如：在CatalogService类中

public PaginatedList getProductListByCategory(String categoryId) { 
    return productDao.getProductListByCategory(categoryId); 
  } 

分页是操作数据库型系统常遇到的问题。分页实现方法很多，但效率的差异就很大了。iBatis是通过什么方式来实现这个分页的了。查看它的实现部分：

返回的PaginatedList实际上是个接口，实现这个接口的是PaginatedDataList类的对象，查看PaginatedDataList类发现，每次翻页的时候最后都会调用下面这段函数

private List getList(int idx, int localPageSize) throws SQLException { 
    return sqlMapExecutor.queryForList(statementName, parameterObject, (idx) * pageSize, localPageSize); 
  }

由于

public interface SqlMapClient extends SqlMapExecutor, SqlMapTransactionManager {……} 

所以实际的调用次序如下：

SqlMapClientImpl.queryForPaginatedList
->SqlMapSessionImpl.queryForPaginatedList 
->SqlMapExecutorDelegate.queryForPaginatedList
->GeneralStatement.executeQueryForList 
->GeneralStatment.executeQueryWithCallback
->GeneralStatment.executeQueryWithCallback 
->SqlExecutor.executeQuery->SqlExecutor.handleMultipleResults()
->SqlExecutor.executeQuery-> handleResults 

分页处理的函数如下

private void handleResults(RequestScope request, ResultSet rs, int skipResults, int maxResults, RowHandlerCallback callback) throws SQLException { 
    try { 
      request.setResultSet(rs); 
      ResultMap resultMap = request.getResultMap(); 
      if (resultMap != null) { 
        // Skip Results 
        if (rs.getType() != ResultSet.TYPE_FORWARD_ONLY) { 
          if (skipResults > 0) { 
            rs.absolute(skipResults); 
          } 
        } else { 
          for (int i = 0; i < skipResults; i++) { 
            if (!rs.next()) { 
              return; 
            } 
          } 
        } 

        // Get Results 
        int resultsFetched = 0; 
        while ((maxResults == SqlExecutor.NO_MAXIMUM_RESULTS || resultsFetched < maxResults) && rs.next()) { 
          Object[] columnValues = resultMap.resolveSubMap(request, rs).getResults(request, rs); 
          callback.handleResultObject(request, columnValues, rs); 
          resultsFetched++; 
        } 
      } 
    } finally { 
      request.setResultSet(null); 
    } 
  } 

由此可见，iBatis的分页主要依赖于jdbcdriver的如何实现以及是否支持rs.absolute(skipResults)。它并不是一个好的分页方式。它先要取出所有的符合条件的记录存入ResultSet对象，然后用absolute方法进行定位，来实现分页。当记录数较大（比如十万条）时，整体的查询速度将会变得很慢。
所以分页还是要考虑采用直接操作sql语句来完成。当然小批量的可以采用iBatis的分页模式。一般分页的sql语句与数据库的具体实现有关

mysql: 
select * from A limit startRow,endRow 
oracle: 
select b.* from (select a.*,rownum as linenum from (select * from A) a where rownum <= endRow) b where linenum >= startRow 

Hibernate的Oracle分页采用的就是是拼凑RowNum的Sql语句来完成的。参考代码如下：

 public String createOraclePagingSql(String sql, int pageIndex, int pageSize){ 
            int m = pageIndex * pageSize; 
            int n = m + pageSize; 
            return "select * from ( select row_.*, rownum rownum_ from ( " + sql 
                    + " ) row_ where rownum <= " + n  
                    + ") where rownum_ > " + m; 
        } 

综上，小批量（<2w）可以采用ibatis自带的分页类，大批量的还是直接操纵sql，当然也可以将这些sql自己进行封装，或在包中封装都可以。包封装的示例代码如下：
一个封装了分页功能的Oracle Package

create or replace package body FMW_FY_HELPER is
PROCEDURE GET_DATA(pi_sql in varchar,pi_whichpage in integer,pi_rownum in integer,
po_cur_data out cur_DATA,po_allrownum out integer,pio_succeed in out integer)
as 
v_cur_data cur_DATA;
v_cur_temp cur_TEMP;
v_temp integer;
v_sql varchar(5000);
v_temp1 integer;
v_temp2 integer;
begin
pio_succeed := 1;
v_sql := 'select count(''a'') from ( ' || pi_sql || ')';
execute immediate v_sql into v_temp;

po_allrownum:=ceil(v_temp/pi_rownum);

v_sql := '';
v_temp :=pi_whichpage*pi_rownum + 1;
v_temp1:=(pi_whichpage-1)*pi_rownum + 1;
v_temp2:=pi_whichpage*pi_rownum;
v_sql:= 'select * from (select rownum as rn,t.* from (' || pi_sql ||') t where rownum<' || to_char(v_temp) || ')  where rn between ' || to_char(v_temp1) || ' and ' || to_char(v_temp2);
open v_cur_data for v_sql;
if v_cur_data %notfound
then
pio_succeed:=-1;
return;
end if;
po_cur_DATA := v_cur_data;
end;