Hi Thanks for your reply. The way I want is not only for hash-lookup-join, there are manay operators need a hash-operation to solve the skew-problem. Lookup-join is a special scene. So I hope there is a operator could make a shuffle. Maybe it's a way to solve the problems ? https://docs.google.com/document/d/1D7AX-_wttMNY53TxLQxiDaRyDVCeEZYCE8AwYflDXZM/edit?usp=sharing
lpengdr...@163.com 发件人: Jark Wu 发送时间: 2022-05-09 12:27 收件人: dev 主题: Re: 【Could we support distribute by For FlinkSql】 Hi, If you are looking for the hash lookup join, there is an in-progress FLIP-204[1] working for it. Btw, I still can't see your picture. You can upload your picture to some image service and share a link here. Best, Jark [1]: https://cwiki.apache.org/confluence/display/FLINK/FLIP-204%3A+Introduce+Hash+Lookup+Join On Mon, 9 May 2022 at 11:22, lpengdr...@163.com <lpengdr...@163.com> wrote: > Sorry! > The destroied picture is the attachment ; > > ------------------------------ > lpengdr...@163.com > > > *发件人:* lpengdr...@163.com > *发送时间:* 2022-05-09 11:16 > *收件人:* user-zh <user...@flink.apache.org>; dev <dev@flink.apache.org> > *主题:* 【Could we support distribute by For FlinkSql】 > Hello: > Now we cann't add a shuffle-operation in a sql-job. > Sometimes , for example, I have a kafka-source(three partitions) with > parallelism three. And then I have a lookup-join function, I want process > the data distribute by id so that the data can split into thre parallelism > evenly (The source maybe slant seriously). > In DataStream API i can do it with keyby(), but it's so sad that i can do > nothing when i use a sql; > Maybe we can do it like 'select id, f1,f2 from sourceTable distribute by > id' like we do it in SparkSql. > > Sot that we can make change on the picture in sql-mode; > > > > ------------------------------ > lpengdr...@163.com > >