Additionally, does the lookup table with CACHE[1][2] meet your needs? If so, you might need to use or implement a dimension table connector with cache.
[1] https://issues.apache.org/jira/browse/FLINK-28415 [2] https://nightlies.apache.org/flink/flink-docs-release-1.17/docs/connectors/table/jdbc/#lookup-cache -- Best！ Xuyang 在 2024-11-04 17:54:36，"Xuyang" <xyzhong...@163.com> 写道： Hi, The BROADCAST[1] join hint currently applies only to batch mode. [1] https://nightlies.apache.org/flink/flink-docs-master/docs/dev/table/sql/queries/hints/#broadcast[1] -- Best！ Xuyang At 2024-11-04 17:06:59, "Guillermo Ortiz Fernández" <guillermo.ortiz.f...@gmail.com> wrote: Hi, I'm running a simple query that joins two tables, where one table is much larger than the other, with the second table being very small. I believe it would be optimal to use a broadcast on the second table for the join. All my tests are being done locally, with very little data in either table. When I apply the hint to perform the broadcast and check the execution plan, I see that, whether the hint is present or not, the join is done using a hash shuffle. Does the hint not enforce the broadcast? Could it be because I’m running it locally or because the tables contain very few records? I'm executing all test from FlinkSQL and sql-client. EXPLAIN PLAN FOR SELECT /*+ BROADCAST(smalltable) */ bigtable.eventTimestamp, bigtable.field1, .... smalltable.technology FROM bigtable JOIN smalltable FOR SYSTEM_TIME AS OF EventTimestampLtz ON bigtable.cgi = smalltable.cgi;
