Thanks Peter and Ryan for the info. As identifier fields need to be "required", how can I alter an optional column to be required in Spark SQL?
Thanks, Manu On Fri, Jan 5, 2024 at 12:50 AM Ryan Blue <b...@tabular.io> wrote: > You can set the primary key fields in Spark using `ALTER TABLE`: > > `ALTER TABLE t SET IDENTIFIER FIELDS id` > > Spark doesn't support any primary key syntax, so you have to do this as a > separate step. > > On Thu, Jan 4, 2024 at 8:46 AM Péter Váry <peter.vary.apa...@gmail.com> > wrote: > >> Hi Manu, >> >> The Iceberg Schema defines `identifierFieldIds` method [1], and Flink >> uses that as the primary key. >> Are you saying there is no way to set it in Spark and Trino? >> >> Thanks, >> Peter >> >> [1] >> https://github.com/apache/iceberg/blob/9a00f7477dedac4501fb2de9e1e6d7aa83dc20b7/api/src/main/java/org/apache/iceberg/Schema.java#L280 >> >> Manu Zhang <owenzhang1...@gmail.com> ezt írta (időpont: 2024. jan. 4., >> Cs, 16:45): >> >>> Hi all, >>> >>> Currently, we support upserting a Flink created table with Flink SQL >>> where primary keys are required as equality fields. They are not required >>> in Java API. >>> >>> However, if the table is created by Spark, where there's no primary key, >>> we cannot upsert with Flink SQL. Hence, I proposed >>> https://github.com/apache/iceberg/pull/8195 to support specifying >>> equality columns with Flink SQL write options. >>> >>> @pvary <https://github.com/pvary> suggested it would be better to >>> support primary keys in Spark, Trino, etc. Since these engines don't have >>> primary keys in their table definitions, a workaround is to put primary key >>> columns in table properties. Maybe there are other options I've missed. >>> >>> Flink SQL sinking to Spark tables for analysis is a typical pipeline in >>> our datalake. I'd like to hear your thoughts on best supporting this case. >>> >>> Happy New Year! >>> Manu >>> >> > > -- > Ryan Blue > Tabular >
