環境
1.sql server 數據庫
2.使用mybatis-plus分頁插件
需求
有兩種表分別是電腦表、電腦配套表。兩張表的關系是一對多。表數據如下：
電腦表
電腦配套表
需求是查詢那些電腦擁有特殊配套。
正常查詢
image.png
過濾只返回電腦信息
image.png
mapper xml的代碼
<mapper namespace="com.xxxx.xxxx.business.dao.ComputerDao">
<select id="getList"
resultType="com.xxxx.xxxx.business.entity.Computer" >
select DISTINCT c.* from computer c LEFT join computer_part cp on c.id = cp.computer_id
where cp.type ='特殊配套'
</select>
</mapper>
swagger調用返回結果
swagger
很簡單的功能可以看到total是2,但是數據內容是4。分別有一條重復。
原因在于上面的正常查詢，確實是4條數據。但是mapper里面明明已經配置了distinct為什么還會重復呢？
因為使用了mybatis-plus分頁插件，所以實際查詢語句是被起包裝過的，具體，通過控制臺可以找到實際sql執行語句。如下：
image.png
WITH selectTemp AS (SELECT DISTINCT TOP 100 PERCENT ROW_NUMBER() OVER (ORDER BY CURRENT_TIMESTAMP) as __row_number__, c.* from computer c LEFT join computer_part cp on c.id = cp.computer_id where cp.type ='特殊配套') SELECT * FROM selectTemp WHERE __row_number__ BETWEEN 1 AND 10 ORDER BY __row_number__
sql語句拷貝到數據庫連接工具執行：
sql執行結果
確實是四條。
到此真相大白，確實是插件的鍋。查看插件源碼：
public String buildPaginationSql(String originalSql, long offset, long limit) {
StringBuilder pagingBuilder = new StringBuilder();
String orderby = getOrderByPart(originalSql);
String distinctStr = "";
String loweredString = originalSql.toLowerCase();
String sqlPartString = originalSql;
if (loweredString.trim().startsWith("select")) {
int index = 6;
if (loweredString.startsWith("select distinct")) {
distinctStr = "DISTINCT ";
index = 15;
}
sqlPartString = originalSql.substring(index);
}
pagingBuilder.append(sqlPartString);
if (StringUtils.isEmpty(orderby)) {
orderby = "ORDER BY CURRENT_TIMESTAMP";
}
StringBuilder sql = new StringBuilder();
sql.append("WITH selectTemp AS (SELECT ").append(distinctStr).append("TOP 100 PERCENT ").append(" ROW_NUMBER() OVER (").append(orderby).append(") as __row_number__, ").append(pagingBuilder).append(") SELECT * FROM selectTemp WHERE __row_number__ BETWEEN ").append(offset + 1L).append(" AND ").append(offset + limit).append(" ORDER BY __row_number__");
return sql.toString();
}
可以看到它確實是把我自帶的distinct給提出去，再跟ROW_NUMBER函數拼接，導致去重失敗。也可以看到源碼中select distinct要是不只間隔一個空格，它還識別不出來。經過嘗試多個空格還是導致報錯。到此，基本鍋已定。我這個版本是3.0.3。github上最新的是3.4.0。源碼如下
public DialectModel buildPaginationSql(String originalSql, long offset, long limit) {
StringBuilder pagingBuilder = new StringBuilder();
String orderby = getOrderByPart(originalSql);
String distinctStr = StringPool.EMPTY;
String loweredString = originalSql.toLowerCase();
String sqlPartString = originalSql;
if (loweredString.trim().startsWith("select")) {
int index = 6;
if (loweredString.startsWith("select distinct")) {
distinctStr = "DISTINCT ";
index = 15;
}
sqlPartString = sqlPartString.substring(index);
}
pagingBuilder.append(sqlPartString);
// if no ORDER BY is specified use fake ORDER BY field to avoid errors
if (StringUtils.isBlank(orderby)) {
orderby = "ORDER BY CURRENT_TIMESTAMP";
}
long firstParam = offset + 1;
long secondParam = offset + limit;
String sql = "WITH selectTemp AS (SELECT " + distinctStr + "TOP 100 PERCENT " +
" ROW_NUMBER() OVER (" + orderby + ") as __row_number__, " + pagingBuilder +
") SELECT * FROM selectTemp WHERE __row_number__ BETWEEN " +
//FIX#299：原因：mysql中limit 10(offset,size) 是從第10開始（不包含10）,；而這里用的BETWEEN是兩邊都包含，所以改為offset+1
firstParam + " AND " + secondParam + " ORDER BY __row_number__";
return new DialectModel(sql);
}
一看終究還是錯付了。實際嘗試還是會存在同樣的問題。
我暫時沒想通為啥作者要先處理 select 或者select distinct，而不是直接把原始sql整段保留，然后在拼接ROW_NUMBER函數進行分頁處理。
解決
1.等待作者處理
2.手動分頁
3.子查詢的方式，避免連表查詢
最終處理方案3，修改mapper代碼如下：
<mapper namespace="com.xxxx.xxxx.business.dao.ComputerDao">
<select id="getList"
resultType="com.xxxx.xxxx.business.entity.Computer" >
select * from computer c
where id in (select computer_id from computer_part where type = '特殊配套')
</select>
</mapper>
swagger執行結果
期望的結果
結語
本文到此結束。
偶爾分享一點，記得點贊加關注。
我也是mybatis-plus的小白使用者，本文如有不正確之處，望各種留情。
