Hi:
When we use sql for aggregation operation, for example, the following sql select count( distinct name) cnt, type from table group by type Source data can be regarded as bin log data. If the type value of a record is updated in the database, the values before and after the update will be divided into different partitions and handed over to different operators for calculation. Can Retraction happen correctly? |
Logically, when retracting , the old value can be sent to the original partition , as long as the group by operator supports that the two values before and after the update can be sent to different partitions. But don't know whether flink sql does this. 刘大龙 <[hidden email]> 于2020年7月1日周三 上午10:23写道: I think the old value will not retract, because the type value update, it will be calculate in new value, the old value will not be updated |
oh, sorry, if source operator can retract old value, I think it can.
-----原始邮件----- |
In reply to this post by lec ssmi
Hi lec ssmi, > If the type value of a record is updated in the database, the values before and after the update will be divided into different partitions and handed over to different operators for calculation. I think your understanding is correct. > Can Retraction happen correctly? I didn't get your point, can you elaborate your question a little bit? lec ssmi <[hidden email]> 于2020年7月1日周三 上午9:54写道:
Best, Benchao Li |
The old value is already counted in a partition, and when the above update occurs, will the count value of the old partition be subtracted by 1, and then added to the new partition? Benchao Li <[hidden email]> 于2020年7月1日周三 下午1:11写道:
|
If you are using 1.11 new changelog format, I think it will retract old value from old partition correctly. If not, (I assume you are using append only changelog) I think it won't retract old value. lec ssmi <[hidden email]> 于2020年7月1日周三 下午2:39写道:
Best, Benchao Li |
Free forum by Nabble | Edit this page |