My-Oracle数据库优化-with as 分析优化

WITH AS (CTE) 优化改写原理

WITH AS 即公用表表达式(CTE, Common Table Expression) ，核心原理是：将子查询结果集临时存储在内存或临时段中，可被主查询多次引用。

一、核心工作原理

sql

复制代码

-- 原始写法：子查询执行2次
SELECT * FROM (SELECT * FROM orders WHERE status='ACTIVE') a
UNION ALL
SELECT * FROM (SELECT * FROM orders WHERE status='ACTIVE') b;

-- CTE写法：只执行1次，结果复用
WITH active_orders AS (
    SELECT * FROM orders WHERE status='ACTIVE'
)
SELECT * FROM active_orders
UNION ALL
SELECT * FROM active_orders;

执行流程：

物化阶段：执行CTE中的查询，将结果集存入临时工作区
引用阶段：主查询中每次引用CTE，直接读取临时结果集
清理阶段：主查询执行完毕后，释放临时空间

二、关键优化场景

场景1：多表重复扫描

sql

复制代码

-- ❌ 低效：同一张表扫描3次
SELECT 
    (SELECT COUNT(*) FROM sales WHERE year=2024) AS total,
    (SELECT AVG(amount) FROM sales WHERE year=2024) AS avg_amount,
    (SELECT SUM(amount) FROM sales WHERE year=2024) AS sum_amount;

-- ✅ 高效：只扫描1次
WITH sales_2024 AS (
    SELECT amount FROM sales WHERE year=2024
)
SELECT 
    (SELECT COUNT(*) FROM sales_2024) AS total,
    (SELECT AVG(amount) FROM sales_2024) AS avg_amount,
    (SELECT SUM(amount) FROM sales_2024) AS sum_amount;

场景2：递归查询（CTE独有功能）

sql

复制代码

-- 组织架构树查询
WITH RECURSIVE org_tree AS (
    -- 锚点：根节点
    SELECT id, name, parent_id, 1 AS level
    FROM organization WHERE parent_id IS NULL
    UNION ALL
    -- 递归：向下遍历
    SELECT o.id, o.name, o.parent_id, t.level + 1
    FROM organization o
    INNER JOIN org_tree t ON o.parent_id = t.id
)
SELECT * FROM org_tree;

场景3：复杂分页优化

sql

复制代码

-- 先CTE过滤和排序，再分页
WITH ranked_data AS (
    SELECT t.*, ROW_NUMBER() OVER (ORDER BY create_time DESC) AS rn
    FROM huge_table t
    WHERE status = 'ACTIVE'
)
SELECT * FROM ranked_data WHERE rn BETWEEN 1 AND 100;

三、优化原理深度解析

优化机制	说明	适用场景
结果物化	将中间结果存入临时表，避免重复计算	子查询被多次引用
查询合并	优化器可将CTE与外层查询合并执行计划	单次引用的简单CTE
递归迭代	支持树形/图形数据的遍历查询	组织架构、BOM物料清单
逻辑封装	将复杂逻辑拆解为多个语义清晰的步骤	长SQL可维护性改造

四、Oracle优化器特殊行为

sql

复制代码

-- Oracle中可控制物化行为
WITH /*+ MATERIALIZE */ cache_data AS (  -- 强制物化
    SELECT * FROM large_table WHERE condition
)
SELECT * FROM cache_data;

WITH /*+ INLINE */ cache_data AS (  -- 强制不物化，直接合并
    SELECT * FROM large_table WHERE condition
)
SELECT * FROM cache_data;

Oracle优化器决策规则：

默认情况下，CTE只被引用一次 → 可能不物化（直接合并）
CTE被引用多次 → 自动物化
CTE结果集很大 → 物化到临时段
CTE结果集很小 → 物化到内存（PGA）

五、针对您原SQL的CTE改写示例

sql

复制代码

WITH base_data AS (
    -- 核心关联数据：一次性获取所有基础字段
    SELECT 
        a.FLOW_NO, a.FLOW_TITLE, a.SERVICE_TYPE, a.END_TIME,
        a.SEND_DEPT, a.SEND_MAN, a.SEND_TIME, a.FLOW_ID,
        b.DISMANTLE_TYPE, b.TASK_DESC, b.START_TIME, b.REPLY_TIME,
        b.DEAL_TIME, b.SERVICE_TYPE_ID, b.SPECIAL_TYPE_ID,
        b.APP_ORDER_NUMBER, b.SERVICE_COMPANY, b.BELONG_UNIT,
        b.PROJECT_NAME, b.SPECIAL_TYPE, b.ISAUTO,
        b.JC_KH_SCORE, b.JC_KHSCORE_FEE, b.PLANTIME, b.FINISHTIME,
        b.DEGREE, b.PRICE, b.TOTAL, b.MONEY, b.BUDGET_DEPT, b.DUTY_DEPT,
        b.ISJK, b.SERVICE_COMPANY_ID
    FROM t_BPM_FORM_INFO a
    INNER JOIN T_BNS_SERVICE_BYNUM_jc b 
        ON a.FLOW_ID = b.FLOW_ID 
        AND a.FLOW_NO = b.APP_ORDER_NUMBER
    WHERE a.SEND_TIME >= TO_DATE('2021/01/01 00:00:00', 'yyyy/mm/dd hh24:mi:ss')
      AND a.APP_TYPE = 'netMaintain'
      AND a.CURRENT_STATE <> '4'
      AND b.SPECIAL_TYPE_ID <> '009'
      AND b.SERVICE_COMPANY_ID <> '2521'
      AND b.IS_ZH IS NULL
),
-- 代维回单时间子查询（只执行1次）
deal_time_info AS (
    SELECT 
        DISORDERNUMBER,
        TO_CHAR(MAX(CREATIONTIME), 'yyyy-mm-dd hh24:mi:ss') AS max_creation_time
    FROM authdb.ufp_general_info_table
    WHERE NODECODE = 'clz'
    GROUP BY DISORDERNUMBER
),
-- 现场完成时间子查询
field_time_info AS (
    SELECT FLOW_ID, TO_CHAR(ETIME, 'yyyy-mm-dd hh24:mi:ss') AS etime_str
    FROM df_task
),
-- 当前环节子查询
current_step_info AS (
    SELECT JOBCODE, ACTDEFNAME
    FROM authdb.ufp_pending_worke
    WHERE (JOBCODE, ACTDEFNAME) IN (
        SELECT JOBCODE, MIN(ACTDEFNAME) 
        FROM authdb.ufp_pending_worke 
        GROUP BY JOBCODE
    )
)
-- 主查询：直接JOIN CTE结果
SELECT 
    b.FLOW_NO AS 工单编号,
    b.FLOW_TITLE AS 工单主题,
    CASE WHEN b.SERVICE_TYPE IN ('室分退网', '基站退网') THEN b.DISMANTLE_TYPE ELSE '' END AS 整站拆除或整逻辑站拆除,
    b.TASK_DESC AS 派单描述,
    TO_DATE(b.START_TIME, 'yyyy/MM/dd hh24:mi:ss') AS 派单时间,
    TO_DATE(b.REPLY_TIME, 'yyyy/MM/dd hh24:mi:ss') AS 处理时限,
    COALESCE(b.DEAL_TIME, d.max_creation_time) AS 代维回单时间,
    CASE 
        WHEN b.SERVICE_TYPE_ID = '14' THEN f.etime_str
        WHEN b.SPECIAL_TYPE_ID = '002' AND b.SERVICE_TYPE_ID = '13' THEN b.DEAL_TIME
        ELSE '--'
    END AS 现场完成时间,
    -- ... 其他字段
    COALESCE(c.ACTDEFNAME, '已归档') AS 当前环节,
    CASE 
        WHEN (b.DEAL_TIME IS NULL AND b.REPLY_TIME < TO_CHAR(SYSDATE, 'yyyy-MM-dd hh24:mi:ss') AND b.ISJK IS NULL)
          OR (b.DEAL_TIME IS NOT NULL AND b.REPLY_TIME < b.DEAL_TIME)
        THEN '是' ELSE '否'
    END AS 是否超时
FROM base_data b
LEFT JOIN deal_time_info d ON b.APP_ORDER_NUMBER = d.DISORDERNUMBER
LEFT JOIN field_time_info f ON b.APP_ORDER_NUMBER = f.FLOW_ID
LEFT JOIN current_step_info c ON b.FLOW_NO = c.JOBCODE
ORDER BY b.FLOW_NO;

六、性能对比总结

维度	原SQL（标量子查询）	CTE改写后
`ufp_general_info_table` 扫描次数	每行1次	1次（全表+聚合）
`df_task` 扫描次数	每行1次	1次（全表）
`ufp_pending_worke` 扫描次数	每行1次	1次（全表）
适用数据量	< 1000行	> 10000行
可读性	较低	高（逻辑分层清晰）

核心结论： CTE通过空间换时间 的方式，将多次重复的子查询扫描转化为单次物化+多次读取，适合子查询被频繁引用且结果集可复用的场景。