cocolate join在查询中带函数的情况下会失效

[hwwxg]

同一个表包含聚合函数的查询时，也必须要join，但是却不是cocolate join，总是报"colocate: false, reason: Node type not match"，为什么会这样，能优化吗？
比如下列语句：
select count() from (select user_id from t_event where event_name = 'AppOpen' group by user_id having count() > 5) t1 where user_id in (select user_id from t_event where event_name = 'Buy' group by user_id having count() > 3)

[HappenLee]

This feature already support in Doris. If you query do not support colocate join, Please check the query and principle of colocate join.

[hwwxg]

我发现和是不是同一个表没关系，而是只要条件中带了函数，就会报"colocate: false, reason: Node type not match"，比如
select count() from (select user_id from t_event where event_name = 'AppOpen' group by user_id having count() > 3) t1 where user_id in (select user_id from t_tag where tag_1 ='male')
select count() from (select user_id from t_event where event_name = 'AppOpen' ) t1 where user_id in (select distinct(user_id) from t_tag where tag_1 ='male')

如果没带函数，就可以走cocolate join，如
select count() from (select user_id from t_event where event_name = 'AppOpen' ) t1 where user_id in (select user_id from t_tag where tag_1 ='male')

[HappenLee]

Yes, It seems not the colocate join problem. It's the problem of group by should know the data distribution of colocate table and the normal table. In this case, no need two layer aggregation operation.

Doris should condiser to do optimization of this case.

[hwwxg]

not only 'group by', but also 'distinct' and other functions

[EmmyMiao87]

This is indeed a point that can be optimized.
When the partitioning method of the upper plan node is the same as that of the lower layer, Doris can use colocate.
I can be responsible for this function.

[EmmyMiao87]

This pr resolved this problem #5521 @hwwxg