T-SQL Recipes之Organizing and Archiving Data

The Problem

当我们处理存档数据或内存数据时，我们想要自定义命名表名，数据库，架构加上日期，时间，或者应用名时，用标准的TSQL来实现是比较困难的。

假设我们有一张日志表，增长速度异常快。但是你不需要超过一个星期的数据，应该怎么做呢？

表分区当然是最佳选择了，但只有企业版里面有这项功能，难道我们就没有其他选择了吗？当然是有的，先来准备点数据吧。

准备数据：

USE AdventureWorks2014;
GO
IF OBJECT_ID('dbo.Database_Log') IS NOT NULL
BEGIN
    DROP TABLE dbo.Database_Log
END

CREATE TABLE dbo.Database_Log
    (
      log_id INT NOT NULL
                 IDENTITY(1, 1)
                 CONSTRAINT PK_Database_Log PRIMARY KEY CLUSTERED ,
      Log_Time DATETIME ,
      Log_Data NVARCHAR(1000)
    );

DECLARE @datetime DATETIME = CURRENT_TIMESTAMP;
DECLARE @datediff TABLE
    (
      previous_hour SMALLINT
    );

DECLARE @count SMALLINT = 0;
WHILE @count <= 360
    BEGIN
        INSERT  INTO @datediff
                ( previous_hour )
                SELECT  @count;
        SELECT  @count = @count + 1
    END
SELECT  @count = 0;
WHILE @count <= 1000
    BEGIN
        INSERT  INTO Database_Log
                ( Log_Time ,
                  Log_Data
                )
                SELECT  DATEADD(HOUR, -1 * previous_hour, CURRENT_TIMESTAMP) ,
                        CAST(DATEADD(HOUR, -1 * previous_hour,
                                     CURRENT_TIMESTAMP) AS NVARCHAR)
                FROM    @datediff;
        SELECT  @count = @count + 1;
    END

DECLARE @year_offset TINYINT = 5;
WHILE @year_offset > 0
    BEGIN
        INSERT  INTO dbo.Database_Log
                ( Log_Time ,
                  Log_Data
                )
                SELECT TOP 10000
                        DATEADD(YEAR, -1 * @year_offset, Log_Time) ,
                        CAST(DATEADD(YEAR, -1 * @year_offset, Log_Time) AS NVARCHAR)
                FROM    Database_Log
        SELECT  @year_offset = @year_offset - 1;
    END

The Solution

假设我们存档数据格式是，每年为数据库，每星期为一个表。如：dbo.database_log_2016为数据库，其中 dbo.database_log_2016_32为表。

我们动态SQL应该如何实现呢？ 先整理一下思路

• 获取日志记录通过时间段
• 创建数据库或表（如果他们不存在）
• 进行插入操作
• 删除原始数据

代码：

DECLARE @sql_command NVARCHAR(MAX);
DECLARE @parameter_list NVARCHAR(MAX) = '@start_of_week DATETIME, @end_of_week DATETIME';
DECLARE @min_datetime DATETIME;

SELECT @min_datetime = MIN(Log_Time)
FROM Database_Log;

DECLARE @previous_min_time DATETIME = '1/1/1900';
DECLARE @start_of_week DATETIME = CAST(DATEADD(dd, -1 * (DATEPART(dw, @min_datetime) - 1), @min_datetime) AS DATE);
DECLARE @end_of_week DATETIME = DATEADD(WEEK, 1, @start_of_week);
DECLARE @current_year SMALLINT;
DECLARE @current_week TINYINT;
DECLARE @database_name NVARCHAR(128);
DECLARE @table_name NVARCHAR(128);

WHILE (@previous_min_time <> @min_datetime)
BEGIN
    SELECT @current_year = DATEPART(YEAR, @start_of_week);
    SELECT @current_week = DATEPART(WEEK, @start_of_week);
    SELECT @database_name = 'Database_Log_' + CAST(@current_year AS NVARCHAR);

    -- Create the yearly database if it does not already exist
    SELECT @table_name = 'Database_Log_' + CAST(@current_year AS NVARCHAR) + '_' + CAST(@current_week AS NVARCHAR)
    IF NOT EXISTS(SELECT *FROM sys.databases WHERE databases.name = @database_name)
    BEGIN
        SELECT @sql_command = 'CREATE DATABASE [' + @database_name + ']';
        EXEC sp_executesql @sql_command;

    END 

    -- Create the weekly table if it does not already exist
    SELECT @sql_command = '
        USE [' + @database_name + '];
        IF NOT EXISTS (SELECT * FROM sys.tables WHERE tables.name = ''' + @table_name + ''')
        BEGIN
            CREATE TABLE [dbo].[' + @table_name + ']
                (Log_Id INT NOT NULL CONSTRAINT PK_Database_Log_' + CAST(@current_year AS NVARCHAR) + '_' + CAST(@current_week AS NVARCHAR) + ' PRIMARY KEY CLUSTERED,
                Log_Time DATETIME,
                Log_Data NVARCHAR(1000));
        END' 

    EXEC sp_executesql @sql_command;

    SELECT @sql_command = '
    INSERT INTO [' + @database_name + '].[dbo].[' + @table_name + ']
        (Log_Id, Log_Time, Log_Data)

    SELECT
            Log_Id,
            Log_Time,
            Log_Data
    FROM    AdventureWorks2014.dbo.Database_Log
    WHERE
            Log_Time >= @start_of_week
            AND Log_Time <= @end_of_week
            AND Log_Time < DATEADD(WEEK, -1, CURRENT_TIMESTAMP);

    DELETE
    FROM    AdventureWorks2014.dbo.Database_Log
    WHERE
            Log_Time >= @start_of_week
            AND Log_Time <= @end_of_week
            AND Log_Time < DATEADD(WEEK, -1, CURRENT_TIMESTAMP);' 

    EXEC sp_executesql @sql_command,@parameter_list,@start_of_week,@end_of_week

    SELECT @previous_min_time = @min_datetime;

    SELECT    @min_datetime = MIN(Log_Time)
    FROM    Database_Log;

    SELECT @start_of_week = CAST(DATEADD(dd, -1 * (DATEPART(dw, @min_datetime) - 1), @min_datetime) AS DATE);

    SELECT @end_of_week = DATEADD(WEEK, 1, @start_of_week);

 END

结语

善使用动态TSQL会让你在工作中的心情变得更美好，当然也可能更恶劣。以后还会讲到如何使用动态TSQL来备份数据库，索引管理。