Github user zsxwing commented on a diff in the pull request: https://github.com/apache/spark/pull/14030#discussion_r69823441 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/ForeachSink.scala --- @@ -30,7 +32,42 @@ import org.apache.spark.sql.{DataFrame, Encoder, ForeachWriter} class ForeachSink[T : Encoder](writer: ForeachWriter[T]) extends Sink with Serializable { override def addBatch(batchId: Long, data: DataFrame): Unit = { - data.as[T].foreachPartition { iter => + // TODO: Refine this method when SPARK-16264 is resolved; see comments below. + + // This logic should've been as simple as: + // ``` + // data.as[T].foreachPartition { iter => ... } + // ``` + // + // Unfortunately, doing that would just break the incremental planing. The reason is, + // `Dataset.foreachPartition()` would further call `Dataset.rdd()`, but `Dataset.rdd()` just + // does not support `IncrementalExecution`. + // + // So as a provisional fix, below we've made a special version of `Dataset` with its `rdd()` + // method supporting incremental planning. But in the long run, we should generally make newly + // created Datasets use `IncrementalExecution` where necessary (which is SPARK-16264 tries to + // resolve). + + val dataAsT = data.as[T] + val datasetWithIncrementalExecution = + new Dataset(data.sparkSession, dataAsT.logicalPlan, dataAsT.encoder) { --- End diff -- `dataAsT` can be removed. You can use `implicitly[Encoder[T]]` to get the encoder. Please also revert the change to Dataset.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org