期望分布式表返回每个分片的结果，而不是聚合值

如何解决期望分布式表返回每个分片的结果，而不是聚合值

有一个用户标签表table_tag，对应的分布式表是table_tag_all，集群中有6个shard，sharding_key是intHash64(user_id).

通过设置参数distributed_product_mode='local'和distributed_group_by_no_merge=1，返回的结果是6个独立分片的值，而不是一个聚合值。 >

以下是两个测试。测试 1 得到了正确的结果（count_1 的数量的 6 条记录），但测试 2 是聚合的（只有 2 条记录）。如何让测试 2 返回 6 个分片的结果？


    create table database.table_tag on cluster c1
    (
    user_id  String,tag1     String,dt
    )
    order by user_id
    partition by dt
    setting index_granularity=128;
    
    /* test 1: */
    select
    sum(case when tag1='1' then 1 else 0 end ) as count_1
    from database.table_tag_all
    where dt in ('202102','202101')
    SETTINGS distributed_group_by_no_merge=1;
    
    
    /* test 2: */
    select count_1,count(1) 
    from (
    select
    user_id,sum(case when tag1='1' then 1 else 0 end ) as count_1
    from database.table_tag_all
    where dt in ('202102','202101')
    group by user_id
    SETTINGS distributed_group_by_no_merge=1;
    ) t1
    group by count_1
    SETTINGS distributed_group_by_no_merge=1;
    
    /*
    count_1  count()
    0        1222122
    1        33821
    */

解决方法

试试这个查询：

SELECT
    shard_number,count_1,count()
FROM
(
    SELECT
        any(_shard_num) AS shard_number,countIf(tag1 = '1') AS count_1
    FROM database.table_tag_all
    WHERE dt in ('202102','202101')
    GROUP BY user_id
)
GROUP BY
    shard_number,count_1
SETTINGS distributed_group_by_no_merge = 1