MySQL - SQL代码优化

Question

我对编写 SQL 还很陌生，刚刚构建了几个过程来将数据添加到我的 MySQL 数据库中。问题是由于大量查询，它非常慢。我现在所做的是遍历包含未排序的原始数据的表中的每条记录，然后获取该数据点并添加到数据库中。这变得很复杂，因为我必须处理许多 FK。

你能帮我优化一下吗？

例如，要添加我执行的指定表:CALL add_table1(112,15);

添加数据的过程
-- 初级程序

CREATE PROCEDURE `add_table1`(
    IN c_id INT UNSIGNED;
    IN t_id INT UNSIGNED;
)
BEGIN
    -- Table variables
    DECLARE r_id INT UNSIGNED;
    DECLARE dh_name VARCHAR(50);
    DECLARE d_value DECIMAL(20,10);

    -- Loop variables
    DECLARE done BOOLEAN;

    -- Cursor for measurement table
    DECLARE m_cur CURSOR FOR 
        SELECT Run_ID, DataHeader_Name, Data_Value 
        FROM `measurements`.`measurement_20131029_152902`;

    -- Handlers for exceptions
    DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;

    -- Set start time
    UPDATE `measurements`.`queue` 
        SET Start_Time = NOW() 
        WHERE Experiment_ID = 112 AND Procedure_Name = 'add_table1';

    -- Loop through measurement table
    OPEN m_cur;
    m_loop: LOOP
        FETCH m_cur INTO r_id, dh_name, d_value;
        IF done THEN
            CLOSE m_cur;
            LEAVE m_loop;
        END IF;
        CALL add_measurement(dh_name, d_value, t_id, c_id, r_id);
    END LOOP m_loop;
END

添加测量的过程
-- 辅助过程，从 add_table1

调用

CREATE PROCEDURE `add_measurement`(
    IN measurement_header VARCHAR(50),
    IN measurement_value DECIMAL(20,10),
    IN tool_id_var INT UNSIGNED,
    IN config_id_var INT UNSIGNED,
    IN run_id_var INT UNSIGNED
)
BEGIN
    -- Variables representing FKs
    DECLARE data_header_id INT UNSIGNED;
    DECLARE tool_header_link_id INT UNSIGNED;
    DECLARE tool_data_id INT UNSIGNED;
    DECLARE tool_data_link_id INT UNSIGNED;

    -- Add header
    INSERT IGNORE INTO data_headers(DataHeader_Name) 
        VALUES(measurement_header);
    SET data_header_id = (SELECT DataHeader_ID 
        FROM data_headers WHERE DataHeader_Name = measurement_header);

    -- Link header to tool
    INSERT IGNORE INTO tool_header_link(DataHeader_ID, Tool_ID)
        VALUES(data_header_id, tool_id_var);
    SET tool_header_link_id = (SELECT ToolHeaderLink_ID 
        FROM tool_header_link
        WHERE DataHeader_ID = data_header_id AND Tool_ID = tool_id_var);

    -- Add measurement
    INSERT IGNORE INTO tool_data(Data_Value) VALUES(measurement_value);
    SET tool_data_id = (SELECT ToolData_ID
        FROM tool_data WHERE Data_Value = measurement_value);

    -- Link measurement to header and configuration
    INSERT IGNORE INTO 
        tool_data_link(ToolHeaderLink_ID, ToolData_ID, Run_ID)
        VALUES(tool_header_link_id, tool_data_id, run_id_var);
    SET tool_data_link_id = (SELECT ToolDataLink_ID FROM tool_data_link
        WHERE ToolHeaderLink_ID = tool_header_link_id
        AND ToolData_ID = tool_data_id AND Run_ID = run_id_var);

    -- Link measurement to experiment configuration
    INSERT IGNORE INTO tool_link(ToolDataLink_ID, Config_ID)
        VALUES(tool_data_link_id, config_id_var);
END

当前解决方案
我偶然发现 this关于类似问题的解决方案。我将代码的内容封装在 TRANSACTION 中，并立即注意到速度有了巨大的提高。查询的估计完成时间不是大约 36 小时，而是实际完成时间减少到大约 5 分钟!我还对数据库进行了轻微的设计更改并删除了不必要的 FK。如果有人看到进一步改进此代码的方法，我仍然很感兴趣。我的性能在我们的应用程序可接受的范围内，但我一直有兴趣让事情变得更好。

显示更改:

    START TRANSACTION;

    -- Loop through measurement table
    OPEN m_cur;
    m_loop: LOOP
        FETCH m_cur INTO r_id, dh_name, d_value;
        IF done THEN
            CLOSE m_cur;
            LEAVE m_loop;
        END IF;
        CALL add_measurement(dh_name, d_value, t_id, c_id, r_id);
    END LOOP m_loop;

    COMMIT;

替代解决方案
基于下面的答案，我能够将我的新解决方案更新为下面的解决方案。从我的测试来看，这个新解决方案似乎正在按预期运行。它也比以前的解决方案快两倍以上。使用此例程，我可以在大约 2.5 分钟内添加一百万条唯一数据!

谢谢大家的帮助!

CREATE PROCEDURE `add_table`(
    IN config_id_var INT UNSIGNED
)
BEGIN
    START TRANSACTION;

    -- Add header
    INSERT IGNORE INTO data_headers(DataHeader_Name)
        SELECT DataHeader_Name
        FROM `measurements`.`measurement_20131114_142402`;

    -- Add measurement
    INSERT IGNORE INTO tool_data(Data_Value)
        SELECT Data_Value
        FROM `measurements`.`measurement_20131114_142402`;

    -- Link measurement to header and configuration
        -- INSERT Non-Unique Values
    INSERT IGNORE INTO tool_data_link(DataHeader_ID, ToolData_ID, Run_ID)
        SELECT h.DataHeader_ID, d.ToolData_ID, m.Run_ID
        FROM `measurements`.`measurement_20131114_142402` AS m
        JOIN data_headers AS h ON h.DataHeader_Name = m.DataHeader_Name
        JOIN tool_data AS d ON d.Data_Value = m.Data_Value;
        -- INSERT Unique Values
    INSERT IGNORE INTO tool_data_link(DataHeader_ID, ToolData_ID, Run_ID)
        SELECT h.DataHeader_ID, d.ToolData_ID, m.Run_ID
        FROM `measurements`.`measurement_20131114_142402` AS m
        LEFT OUTER JOIN data_headers AS h ON h.DataHeader_Name = m.DataHeader_Name
        LEFT OUTER JOIN tool_data AS d ON d.Data_Value = m.Data_Value
        WHERE ((h.DataHeader_Name IS NULL) OR (d.Data_Value IS NULL));

    -- Link measurement to experiment configuration
        -- INSERT Non-Unique Values
    INSERT IGNORE INTO tool_link(ToolDataLink_ID, Config_ID)
        SELECT tdl.ToolDataLink_ID, config_id_var
        FROM tool_data_link AS tdl
        JOIN data_headers AS h ON h.DataHeader_ID = tdl.DataHeader_ID
        JOIN tool_data AS d ON d.ToolData_ID = tdl.ToolData_ID;
        -- INSERT Unique Values
    INSERT IGNORE INTO tool_link(ToolDataLink_ID, Config_ID)
        SELECT tdl.ToolDataLink_ID, config_id_var
        FROM tool_data_link AS tdl
        LEFT OUTER JOIN data_headers AS h ON h.DataHeader_ID = tdl.DataHeader_ID
        LEFT OUTER JOIN tool_data AS d ON d.ToolData_ID = tdl.ToolData_ID
        WHERE ((h.DataHeader_ID IS NULL) OR (d.ToolData_ID IS NULL));

    COMMIT;
END

结论
我对不使用游标的解决方案进行了更多测试。最初肯定更快；然而，当数据库的大小增长时，执行时间会急剧增加。

我在数据库中添加了几百万个数据点。然后我尝试添加一个包含大约几百个数据点的小数据集。它比游标解决方案花费了将近 400 倍的时间。我相信这是因为游标只查看所需的数据点，而连接必须查看所有数据。

根据这些结果，游标解决方案似乎更适合我的应用程序。

Answer 1

数据库将使用基于集合的逻辑。尝试在没有游标的情况下编写相同的查询。

What your code does:
-1: Reads the records from measurement_xxxx.... table and for each one it executes 5 insert.

How this can be writen with set logic: 
-1: make the first insert for all the records from measurement_XXXX....
-2: make the second insert for all the records from measurement_XXX....
.......

This will change the code like the following: ( I did not test the code, so it is a guideline ) 

CREATE PROCEDURE `using_set_logic`(
    IN c_id INT UNSIGNED;
    IN t_id INT UNSIGNED;
)
begin
-- the first insert 
insert into data_headers(DataHeader_Name) 
select DataHeader_Name FROM `measurements`.`measurement_20131029_152902
-- the second insert if the dataheadername is uniqueu
insert into tool_header_link(DataHeader_ID, Tool_ID)
select DataHeader_ID, t_id from data_headers where DataHeader_Name in (
    select DataHeader_Name FROM `measurements`.`measurement_20131029_152902
)
--the second insert if the dataheadername isnot unique.
--take the last id for the dataheaders before the first insert 
--take the last id for the dataheaders after the insert
--use those values to bound the data header id.
insert into tool_header_link(DataHeader_ID, Tool_ID)
select DataHeader_ID, @t_id from data_headers 
where DataHeader_id between @beforeFirstInsert and @afterFirstInsert

end