Hi devs,
I’d like to start a discussion about the current and future state of our Flink Sink Connectors. As it stands today, we currently have 3 sink implementations: 1. FlinkSink [1] 2. IcebergSink [2] 3. DynamicSink [3] FlinkSink [1] is the current and default implementation of the Flink Sink Connector. IcebergSink [2] is another implementation of the Flink Sink Connector which was introduced in https://github.com/apache/iceberg/pull/10179. <https://github.com/apache/iceberg/pull/10179> It leverages the latest SinkV2 interfaces in Flink, and it offers the possibility of adding cleanup tasks by the way of implementing the `PostCommitTopology` interface. There is already some work in progress to enable this functionality: https://github.com/apache/iceberg/pull/12979 DynamicSink [3] has been recently contributed in https://github.com/apache/iceberg/pull/13304 and it can be used to write to any number of tables, dynamically creating and updating tables and dynamically updating the schema and partition spec of tables. Currently, `IcebergSink` is marked as `@Experimental` and it already offers feature parity with `FlinkSink` (the missing RANGE distribution was recently merged https://github.com/apache/iceberg/pull/12071). With https://github.com/apache/iceberg/pull/11244, users have the choice of specifying which Sink Implementation ([1] or [2]) they want to use for Flink SQL. With all this said, we’re proposing the following for Iceberg* 1.11*: 1. @Deprecate FlinkSink 2. Promote IcebergSink from @Experimental to @PublicEvolving 3. Make IcebergSink the default implementation in Flink SQL. Then in Iceberg *1.12* we will: 1. Remove the `FlinkSink` implementation. 2. Remove @PublicEvolving from IcebergSink What do you think about this plan? Thanks, Rodrigo
