Hi Mans, `LookupTableSource` used to look up rows from external storage system by given keys, it's very suitable for Key/Value storage system (e.g Redis, HBase), or storage system with key concept (e.g, Mysql, Hive). ` ScanTableSource` is used to scan all rows from an external storage system. Some connectors implement both `LookupTableSource` and `ScanTableSource` because they both have two abilities. (e.g Hive, HBase, JDBC).
> 1. Are there other examples/documentation on how to create a query that uses it vs ScanTableSource ? If SQL use Lookup join[1] syntax, optimizer would expect underlying table source which related to the right side has implemented `LookupTableSource`. please note only right side is `LookupTableSource`, left side is still `ScanTableSource`. If SQL uses non-Lookup join syntax (e.g Regular join, Interval join, temporal join, window join)[1], the optimizer would expect the underlying table source to have implemented `ScanTableSource`. > 2. Are there any best practices for using this interface ? What do you expect to find in `a best practice for using this interface`? For a TableApi/SQL user, maybe there are not many difficult problems to use a `LookupTableSource` because they would not deal with `LookupTableSource` explicitly, they only use Lookup join[1] syntax. > 3. How does the planner decide to use LookupTableSource vs ScanTableSource ? Please see the first response above > 4. Are there some hints/etc that can be given to the planner to force it to use LookupTableSource ? No, there is no need to do this. useLookup join[1] syntax if need use `LookupTableSource`, > 5. Is this used only for joins or for regular queries ? Please see the first response above > 6. Can it be used for regular joins ? If so, is there any documentation/example ? If the underlying tablesource only implements `ScanTableSource`, then the user could not use it in lookup join. If the underlying tablesource only implements `LookupTableSource`, then the user could not use it in non-Lookup join syntax (e.g Regular join, Interval join, temporal join, window join). Otherwise an exception would be thrown out in the compile phase. [1] https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/dev/table/sql/queries/joins/ Best regards, JING ZHANG M Singh <mans2si...@yahoo.com> 于2021年7月7日周三 上午8:23写道： > Hey Folks: > > I am trying to understand how LookupTableSource works and have a few > questions: > > 1. Are there other examples/documentation on how create a query that uses > it vs ScanTableSource ? > 2. Are there any best practices for using this interface ? > 3. How does the planner decide to use LookupTableSource vs ScanTableSource > ? > 4. Are there some hints/etc that can be given to the planner to force it > to use LookupTableSource ? > 5. Is this used only for joins or for regular queries ? > 6. Can it be used for regular joins ? If so, is there any > documentation/example ? > > Thanks for your help. > > Mans >
