在 PostgreSQL 中按 UUID 排序是如何工作的？

Question

在 postgresql 上使用 uuid_generate_v1() 是否有任何顺序保证？

如果是，则保证是每台机器，还是在哪台机器上生成 UUID 无关紧要？由于 V1 UUID 是按时间戳 + MAC 地址，Postgres 内部是否先按时间戳部分然后按 MAC 排序？

我可以通过 UUID 类型列“订购”并期望它始终有效（似乎有效）吗？

我想在多台机器上生成 UUID（使用 postgresql uuid_generate_v1()），将它们复制到一个 Postgres 实例，然后按 UUID 列排序。它必须保证机器的顺序，而不是所有机器的 UUID 的顺序。

Answer 1

虽然绝不是一个明确的答案 - 即。“该行为是在所有 PostgreSQL 安装中定义的吗？”此 SQL（对于 SQL Server）用于检查 GUID 中每个字节的顺序。可能需要对 PostgreSQL 进行一些调整。

生成这样的映射应该允许人们查看 PostgreSQL 中特定的 UUID 结构（明确定义的类型之一或其他类型）是否“以特定方式排序”。

With UIDs As (--                           0 1 2 3  4 5  6 7  8 9  A B C D E F
            Select ID = 'F', UID = cast ('00000000-0000-0000-0000-000000000011' as uniqueidentifier)
    Union   Select ID = 'E', UID = cast ('00000000-0000-0000-0000-000000001100' as uniqueidentifier)
    Union   Select ID = 'D', UID = cast ('00000000-0000-0000-0000-000000110000' as uniqueidentifier)
    Union   Select ID = 'C', UID = cast ('00000000-0000-0000-0000-000011000000' as uniqueidentifier)
    Union   Select ID = 'B', UID = cast ('00000000-0000-0000-0000-001100000000' as uniqueidentifier)
    Union   Select ID = 'A', UID = cast ('00000000-0000-0000-0000-110000000000' as uniqueidentifier)
    Union   Select ID = '9', UID = cast ('00000000-0000-0000-0011-000000000000' as uniqueidentifier)
    Union   Select ID = '8', UID = cast ('00000000-0000-0000-1100-000000000000' as uniqueidentifier)
    Union   Select ID = '7', UID = cast ('00000000-0000-0011-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '6', UID = cast ('00000000-0000-1100-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '5', UID = cast ('00000000-0011-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '4', UID = cast ('00000000-1100-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '3', UID = cast ('00000011-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '2', UID = cast ('00001100-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '1', UID = cast ('00110000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '0', UID = cast ('11000000-0000-0000-0000-000000000000' as uniqueidentifier)
)
Select * From UIDs Order By UID desc

在 SQL Server（2014，并与 SQL Server 2005 中的匹配）中，降序排列为：

With UIDs As (--                           0 1 2 3  4 5  6 7  8 9  A B C D E F
            Select ID = 'F', UID = cast ('00000000-0000-0000-0000-000000000011' as uniqueidentifier)
    Union   Select ID = 'E', UID = cast ('00000000-0000-0000-0000-000000001100' as uniqueidentifier)
    Union   Select ID = 'D', UID = cast ('00000000-0000-0000-0000-000000110000' as uniqueidentifier)
    Union   Select ID = 'C', UID = cast ('00000000-0000-0000-0000-000011000000' as uniqueidentifier)
    Union   Select ID = 'B', UID = cast ('00000000-0000-0000-0000-001100000000' as uniqueidentifier)
    Union   Select ID = 'A', UID = cast ('00000000-0000-0000-0000-110000000000' as uniqueidentifier)
    Union   Select ID = '9', UID = cast ('00000000-0000-0000-0011-000000000000' as uniqueidentifier)
    Union   Select ID = '8', UID = cast ('00000000-0000-0000-1100-000000000000' as uniqueidentifier)
    Union   Select ID = '7', UID = cast ('00000000-0000-0011-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '6', UID = cast ('00000000-0000-1100-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '5', UID = cast ('00000000-0011-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '4', UID = cast ('00000000-1100-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '3', UID = cast ('00000011-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '2', UID = cast ('00001100-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '1', UID = cast ('00110000-0000-0000-0000-000000000000' as uniqueidentifier)
    Union   Select ID = '0', UID = cast ('11000000-0000-0000-0000-000000000000' as uniqueidentifier)
)
Select * From UIDs Order By UID desc

由于 SQL Servernewsequentialid使用此顺序来生成索引友好的 GUID，因此该行为可能永远不会改变。SQL Server 还必须在所有系统上维护此行为以支持复制。因此，如果问题是关于 SQL Server 的，我肯定会说“SQL Server中GUID 的顺序是一致的”，这在SQL Server 中绝对可以信赖。

然而，这种排序与 .NET 的 GUID 排序不同，如果 PostgreSQL 中的排序不同，我也不会感到惊讶。SQL Server 中的“翻转”差异是因为它遵循COM GUID的“变体 2”（又名小端）排序；即使对于“变体 1”UUID 也是如此。（然而，为什么这些组本身是从右到左排序的，这似乎更随意：更多的微软历史？）

有趣的问题仍然存在：在PostgreSQL中在哪里/如何指定排序？如果没有明确规定，实现仍然可以被视为行为公理吗？

另请参阅此问题以获取有关 SQL Server UUID 的更多详细信息；以及“为什么”存在这些差异的有趣细节。