kangkaisen commented on issue #245: Colocate Join URL: https://github.com/apache/incubator-doris/issues/245#issuecomment-430946769 > I think there is no need to add "colocate_with" property when create parent table. There is some case that user have already have a table, then they need a colocate table. > **The colocate parent table decide the data distribution for all colocate tables in a group, and the default data distribution couldn't ensure the data locall**y, so we must recreate the colocate parent table. If users want migrate normal tables to colocate tables, They could do this work by recreating the table with different table name and setting "colocate_with" property, then loading the historical data to doris, finally renaming the colocate table and update the "colocate_with" property. > All partition have same Bucket Seqs, this will lead to all partition data must save to same BE. If partition have three buckets, however there are 100 BE in the cluster, then table's data can be located in this 3 BEs. Can this locate method be improved? > Yes, **this is a colocate join internal drawback**. If the cluster node is hundreds of and the table data is huge, the user could set the large bucket number to ease this problem. > And different partition can have different buckets, If user change number of buckets, the locate map need changed too. Yes, **Colocate tables could not change the bucket number in order to keep data locally.**
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
