Open ss892714028 opened 3 months ago
It’s not a bug, we have multi insert concurrency and try to insert into db_stage_DEVELOP.destination_table
, last arrival win. And we cannot guarantee the order of arrival of data with the same key in a multi-concurrency scenario, unless you use set pipeline_dop = 1
, to make sure only one concurrency to handle insert.
@luohaha Honestly, I think this is something StarRocks' planner can handle internally. Maybe something to consider.
@ss892714028 So, what you need is similar to merge_condition
in https://docs.starrocks.io/docs/3.3/loading/Load_to_Primary_Key_tables/#conditional-updates, right?
@ss892714028 Have you solved your requirement?
merge_condition 这个属性可否加入到表属性中,这样的话,根据上述例子指定merge_condition='timestamp' 用sql做ETL时,就不会出现乱序的问题了。在官方文档中,我只看到在导入数据中增加merge_condition该属性,而使用SQL做ETL的场景却不知道该如何确定顺序,期待回复,谢谢。
Steps to reproduce the behavior (Required)
You can also reproduce this with unique key table.
Expected behavior (Required)
If you make the source table single tablet, also add a paraphrasis to the SELECT statement inside of the INSERT, the issue goes away..
Real behavior (Required)
different result every single time
StarRocks version (Required)
3.2.8 allinone ubuntu docker