如何使用多列中的值分配 table?

How do I distributed a table using values from multiple columns?

根据 the Citus documentation,使用单列分布 table 很容易:

SELECT master_create_distributed_table('github_events', 'created_at', 'append');

有没有办法使用多列分布 table?例如,类似于:

SELECT master_create_distributed_table('github_events', 'user_id,site_id', 'append');

Citus 不支持按多列分布。但是,您可以创建复合类型和 partition your data by that composite type.

-- link 的内容在下面内联,以防 link 失效 --

复合类型的散列分区步骤

  1. 在主节点和所有工作节点上创建类型:

    CREATE TYPE new_composite_type as (project_key text, date text);
    
  2. 创建一个检查相等性的函数,并将其与新类型的相等运算符相关联

    CREATE FUNCTION equal_test_composite_type_function(new_composite_type, new_composite_type) RETURNS boolean
    AS 'select .project_key = .project_key AND .date = .date;'
    LANGUAGE SQL
    IMMUTABLE
    RETURNS NULL ON NULL INPUT;
    
    -- ... use that function to create a custom equality operator...
    CREATE OPERATOR = (
        LEFTARG = new_composite_type,
        RIGHTARG = new_composite_type,
        PROCEDURE = equal_test_composite_type_function,
        HASHES
    );
    
  3. 创建一个新的哈希函数。

    注意:这只是一个简单的示例,可能无法提供良好的均匀散列分布。有几个很好的散列函数示例,可以在单独的 C 函数中实现,而不是 SQL.

    CREATE FUNCTION new_composite_type_hash(new_composite_type) RETURNS int
    AS 'SELECT hashtext( (.project_key || .date)::text);'   
    LANGUAGE SQL
    IMMUTABLE
    RETURNS NULL ON NULL INPUT;
    
  4. 为 BTREE 和 HASH 访问方法定义运算符 类:

    CREATE OPERATOR CLASS new_op_fam_btree_class
    DEFAULT FOR TYPE new_composite_type USING BTREE AS
    OPERATOR 3 = (new_composite_type, new_composite_type);
    
    CREATE OPERATOR CLASS new_op_fam_hash_class
    DEFAULT FOR TYPE new_composite_type USING HASH AS
    OPERATOR 1 = (new_composite_type, new_composite_type),
    FUNCTION 1 new_composite_type_hash(new_composite_type);
    
  5. 使用新类型创建 table 并分发它。

    CREATE TABLE composite_type_partitioned_table
    (
        id integer,
        composite_column new_composite_type
    );
    
    SELECT master_create_distributed_table('composite_type_partitioned_table','composite_column', 'hash');
    
    SELECT master_create_worker_shards('composite_type_partitioned_table', 4, 1);
    
  6. 运行 插入和 SELECT。请注意,正确的修剪将需要在这些查询中显示的引用。

    INSERT INTO composite_type_partitioned_table VALUES  (1, '("key1","20160101")'::new_composite_type);
    INSERT INTO composite_type_partitioned_table VALUES  (2, '("key1","20160102")'::new_composite_type);
    INSERT INTO composite_type_partitioned_table VALUES  (3, '("key2","20160101")'::new_composite_type);
    INSERT INTO composite_type_partitioned_table VALUES  (4, '("key2","20160102")'::new_composite_type);
    
    SELECT * FROM composite_type_partitioned_table WHERE composite_column =  '("key1", "20160101")'::new_composite_type;
    
    UPDATE composite_type_partitioned_table SET id = 6 WHERE composite_column =  '("key2", "20160101")'::new_composite_type;
    
    SELECT * FROM composite_type_partitioned_table WHERE composite_column =  '("key2", "20160101")'::new_composite_type;
    

其他说明:

有两点需要注意:

  1. 输入文件必须正确分隔才能使 copy_to_distributed_table 正常工作。为此,使用 COPY (SELECT ()::composite_type_field, .... ); 从普通 table 到一个文件,然后加载。

  2. 为了使用 select 查询进行修剪,复合类型字段应该用引号引起来。