SQL Server GUID 排序算法。为什么？

Question

UniqueIdentifiers 的问题

我们有一个现有的数据库，它广泛使用唯一标识符（不幸的是！）作为主键和某些表的一些可空列。我们遇到了这样一种情况，在这些表上运行的一些报告对这些唯一标识符进行排序，因为表中没有其他列可以提供有意义的排序（这不是很讽刺！）。目的是排序，以便按插入顺序显示项目，但未使用 NewSequentialId() 插入它们 - 因此浪费时间。

排序算法简介

无论如何，考虑到 SQL Server 根据字节组对唯一标识符进行排序，该字节组从结束的第 5 个字节组（6 个字节）开始向第 1 个字节组（4 个字节）移动，将第 3 个字节组（2 个字节）的顺序从从右到左到左到右，

我的问题

我很想知道这种类型是否对现实生活有帮助。

SQL Server 如何在内部存储唯一标识符，这可能有助于深入了解 为什么它有这种古怪的排序算法？

参考：

Alberto Ferrari's discovery of the SQL Server GUID sort

示例

当您在具有以下数据的唯一标识符列上使用排序依据时，唯一标识符的排序如下所示。

请注意，以下数据按升序排序，最高排序偏好是从第 5 个字节组到第 1 个字节组（向后）。

-- 1st byte group of 4 bytes sorted in the reverse (left-to-right) order below -- 

01000000-0000-0000-0000-000000000000
10000000-0000-0000-0000-000000000000
00010000-0000-0000-0000-000000000000
00100000-0000-0000-0000-000000000000
00000100-0000-0000-0000-000000000000
00001000-0000-0000-0000-000000000000
00000001-0000-0000-0000-000000000000
00000010-0000-0000-0000-000000000000

-- 2nd byte group of 2 bytes sorted in the reverse (left-to-right) order below -- 

00000000-0100-0000-0000-000000000000
00000000-1000-0000-0000-000000000000
00000000-0001-0000-0000-000000000000
00000000-0010-0000-0000-000000000000

-- 3rd byte group of 2 bytes sorted in the reverse (left-to-right) order below -- 

00000000-0000-0100-0000-000000000000
00000000-0000-1000-0000-000000000000
00000000-0000-0001-0000-000000000000
00000000-0000-0010-0000-000000000000

-- 4th byte group of 2 bytes sorted in the straight (right-to-left) order below -- 

00000000-0000-0000-0001-000000000000
00000000-0000-0000-0010-000000000000
00000000-0000-0000-0100-000000000000
00000000-0000-0000-1000-000000000000

-- 5th byte group of 6 bytes sorted in the straight (right-to-left) order below -- 

00000000-0000-0000-0000-000000000001
00000000-0000-0000-0000-000000000010
00000000-0000-0000-0000-000000000100
00000000-0000-0000-0000-000000001000
00000000-0000-0000-0000-000000010000
00000000-0000-0000-0000-000000100000
00000000-0000-0000-0000-000001000000
00000000-0000-0000-0000-000010000000
00000000-0000-0000-0000-000100000000
00000000-0000-0000-0000-001000000000
00000000-0000-0000-0000-010000000000
00000000-0000-0000-0000-100000000000

代码：

Alberto 的代码扩展为表示排序是在字节上而不是在单个位上。

With Test_UIDs As (--                     0 1 2 3  4 5  6 7  8 9  A B C D E F
            Select ID =  1, UID = cast ('00000000-0000-0000-0000-100000000000' as uniqueidentifier)
    Union   Select ID =  2, UID = cast ('00000000-0000-0000-0000-010000000000' as uniqueidentifier)
    Union   Select ID =  3, UID = cast ('00000000-0000-0000-0000-001000000000' as uniqueidentifier)
    Union   Select ID =  4, UID = cast ('00000000-0000-0000-0000-000100000000' as uniqueidentifier)
    Union   Select ID =  5, UID = cast ('00000000-0000-0000-0000-000010000000' as uniqueidentifier)
    Union   Select ID =  6, UID = cast ('00000000-0000-0000-0000-000001000000' as uniqueidentifier)
    Union   Select ID =  7, UID = cast ('00000000-0000-0000-0000-000000100000' as uniqueidentifier)
    Union   Select ID =  8, UID = cast ('00000000-0000-0000-0000-000000010000' as uniqueidentifier)
    Union   Select ID =  9, UID = cast ('00000000-0000-0000-0000-000000001000' as uniqueidentifier)
    Union   Select ID = 10, UID = cast ('00000000-0000-0000-0000-000000000100' as uniqueidentifier)
    Union   Select ID = 11, UID = cast ('00000000-0000-0000-0000-000000000010' as uniqueidentifier)
    Union   Select ID = 12, UID = cast ('00000000-0000-0000-0000-000000000001' as uniqueidentifier)
    Union   Select ID = 13, UID = cast ('00000000-0000-0000-0001-000000000000' as uniqueidentifier)
    Union   Select ID = 14, UID = cast ('00000000-0000-0000-0010-000000000000' as uniqueidentifier)
    Union   Select ID = 15, UID = cast ('00000000-0000-0000-0100-000000000000' as uniqueidentifier)
    Union   Select ID = 16, UID = cast ('00000000-0000-0000-1000-000000000000' as uniqueidentifier)
    Union   Select ID = 17, UID = cast ('00000000-0000-0001-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 18, UID = cast ('00000000-0000-0010-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 19, UID = cast ('00000000-0000-0100-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 20, UID = cast ('00000000-0000-1000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 21, UID = cast ('00000000-0001-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 22, UID = cast ('00000000-0010-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 23, UID = cast ('00000000-0100-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 24, UID = cast ('00000000-1000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 25, UID = cast ('00000001-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 26, UID = cast ('00000010-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 27, UID = cast ('00000100-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 28, UID = cast ('00001000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 29, UID = cast ('00010000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 30, UID = cast ('00100000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 31, UID = cast ('01000000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = 32, UID = cast ('10000000-0000-0000-0000-000000000000' as uniqueidentifier)
)
Select * From Test_UIDs Order By UID, ID

Answer 1

SQL Server 人员在此处记录了该算法：How are GUIDs compared in SQL Server 2005? 我在此处引用（因为这是一篇可能在几年后就永远消失的旧文章）

一般来说，相等比较很有意义 唯一标识符值。但是，如果您发现自己需要通用 排序，那么您可能正在查看错误的数据类型，应该 考虑各种整数类型。

如果经过深思熟虑，您决定使用唯一标识符进行排序 列，你可能会对你得到的结果感到惊讶。

鉴于这两个唯一标识符值：

@g1= '55666BEE-B3A0-4BF5-81A7-86FF976E763F' @g2 = '8DD5BCA5-6ABE-4F73-B4B7-393AE6BBB849'

很多人认为@g1 小于@g2，因为 '55666BEE' 是 肯定小于“8DD5BCA5”。但是，这不是 SQL Server 的方式 2005 比较 uniqueidentifier 值。

通过从右到左查看字节“组”来进行比较，并且 在一个字节“组”内从左到右。字节组是分隔的 由 '-' 字符。从技术上讲，我们查看字节 {10 到 15} 首先，然后是 {8-9}，然后是 {6-7}，然后是 {4-5}，最后是 {0 到 3}。

在这个具体示例中，我们将从比较“86FF976E763F”开始 带有“393AE6BBB849”。我们立即看到@g2 确实更大 比@g1。

请注意，在 .NET 语言中，Guid 值具有不同的默认排序 顺序比在 SQL Server 中。如果您发现需要订购阵列或 使用 SQL Server 比较语义的 Guid 列表，您可以使用 SqlGuid 的数组或列表，它在 符合SQL Server语义的方式。

另外，排序遵循字节组字节序（参见此处：Globally unique identifier）。组 10-15 和 8-9 存储为大端（对应于维基百科文章中的 Data4），因此将它们作为大端进行比较。其他组使用 little endian 进行比较。

Answer 2

为那些发现接受的答案有点模糊的人提供的特殊服务。代码不言自明；神奇的部分是：

System.Guid g
g.ToByteArray();
int[] m_byteOrder = new int[16] // 16 Bytes = 128 Bit 
    {10, 11, 12, 13, 14, 15, 8, 9, 6, 7, 4, 5, 0, 1, 2, 3};


public int Compare(Guid x, Guid y)
{
    byte byte1, byte2;

    //Swap to the correct order to be compared
    for (int i = 0; i < NUM_BYTES_IN_GUID; i++)
    {
        byte1 = x.ToByteArray()[m_byteOrder[i]];
        byte2 = y.ToByteArray()[m_byteOrder[i]];
        if (byte1 != byte2)
            return (byte1 < byte2) ? (int)EComparison.LT : (int)EComparison.GT;
    } // Next i 

    return (int)EComparison.EQ;
}

完整代码：

namespace BlueMine.Data
{


    public class SqlGuid
        : System.IComparable
        , System.IComparable<SqlGuid>
        , System.Collections.Generic.IComparer<SqlGuid>
        , System.IEquatable<SqlGuid>
    {
        private const int NUM_BYTES_IN_GUID = 16;

        // Comparison orders.
        private static readonly int[] m_byteOrder = new int[16] // 16 Bytes = 128 Bit 
        {10, 11, 12, 13, 14, 15, 8, 9, 6, 7, 4, 5, 0, 1, 2, 3};

        private byte[] m_bytes; // the SqlGuid is null if m_value is null


        public SqlGuid(byte[] guidBytes)
        {
            if (guidBytes == null || guidBytes.Length != NUM_BYTES_IN_GUID)
                throw new System.ArgumentException("Invalid array size");

            m_bytes = new byte[NUM_BYTES_IN_GUID];
            guidBytes.CopyTo(m_bytes, 0);
        }


        public SqlGuid(System.Guid g)
        {
            m_bytes = g.ToByteArray();
        }


        public byte[] ToByteArray()
        {
            byte[] ret = new byte[NUM_BYTES_IN_GUID];
            m_bytes.CopyTo(ret, 0);
            return ret;
        }

        int CompareTo(object obj)
        {
            if (obj == null)
                return 1; // https://msdn.microsoft.com/en-us/library/system.icomparable.compareto(v=vs.110).aspx

            System.Type t = obj.GetType();

            if (object.ReferenceEquals(t, typeof(System.DBNull)))
                return 1;

            if (object.ReferenceEquals(t, typeof(SqlGuid)))
            {
                SqlGuid ui = (SqlGuid)obj;
                return this.Compare(this, ui);
            } // End if (object.ReferenceEquals(t, typeof(UInt128)))

            return 1;
        } // End Function CompareTo(object obj)


        int System.IComparable.CompareTo(object obj)
        {
            return this.CompareTo(obj);
        }


        int CompareTo(SqlGuid other)
        {
            return this.Compare(this, other);
        }


        int System.IComparable<SqlGuid>.CompareTo(SqlGuid other)
        {
            return this.Compare(this, other);
        }


        enum EComparison : int
        {
            LT = -1, // itemA precedes itemB in the sort order.
            EQ = 0, // itemA occurs in the same position as itemB in the sort order.
            GT = 1 // itemA follows itemB in the sort order.
        }


        public int Compare(SqlGuid x, SqlGuid y)
        {
            byte byte1, byte2;

            //Swap to the correct order to be compared
            for (int i = 0; i < NUM_BYTES_IN_GUID; i++)
            {
                byte1 = x.m_bytes[m_byteOrder[i]];
                byte2 = y.m_bytes[m_byteOrder[i]];
                if (byte1 != byte2)
                    return (byte1 < byte2) ? (int)EComparison.LT : (int)EComparison.GT;
            } // Next i 

            return (int)EComparison.EQ;
        }


        int System.Collections.Generic.IComparer<SqlGuid>.Compare(SqlGuid x, SqlGuid y)
        {
            return this.Compare(x, y);
        }


        public bool Equals(SqlGuid other)
        {
            return Compare(this, other) == 0;
        }


        bool System.IEquatable<SqlGuid>.Equals(SqlGuid other)
        {
            return this.Equals(other);
        }


    }


}

Answer 3

这是一种不同的方法。 GUID 被简单地打乱，准备进行正常的字符串比较，就像在 SQL Server 中发生的那样。这是 Javascript，但它很容易转换为任何语言。

function guidForComparison(guid) {
  /*
  character positions:  
            11111111112222222222333333
  012345678901234567890123456789012345

  00000000-0000-0000-0000-000000000000

  byte positions:  
                          111111111111
  00112233 4455 6677 8899 001122334455
  */
  return guid.substr(24, 12) + 
         guid.substr(19, 4) + 
         guid.substr(16, 2) + 
         guid.substr(14, 2) + 
         guid.substr(11, 2) + 
         guid.substr(9, 2) + 
         guid.substr(6, 2) +
         guid.substr(4, 2) +
         guid.substr(2, 2) +
         guid.substr(0, 2);
};