cloud-fan commented on code in PR #36295: URL: https://github.com/apache/spark/pull/36295#discussion_r904487223
########## sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/V2ScanRelationPushDown.scala: ########## @@ -407,7 +407,67 @@ object V2ScanRelationPushDown extends Rule[LogicalPlan] with PredicateHelper wit case other => (other, false) } - def pushDownLimits(plan: LogicalPlan): LogicalPlan = plan.transform { + private def pushDownOffset( + plan: LogicalPlan, + offset: Int): Boolean = plan match { + case sHolder: ScanBuilderHolder => + val isPushed = PushDownUtils.pushOffset(sHolder.builder, offset) + if (isPushed) { + sHolder.pushedOffset = Some(offset) + } + isPushed + case Project(projectList, child) if projectList.forall(_.deterministic) => + pushDownOffset(child, offset) + case _ => false + } + + def pushDownLimitAndOffset(plan: LogicalPlan): LogicalPlan = plan.transform { + case offset @ LimitAndOffset(limit, offsetValue, child) => + val (newChild, canRemoveLimit) = pushDownLimit(child, limit) + if (canRemoveLimit) { + // Try to push down OFFSET only if the LIMIT operator has been pushed and can be removed. + // For `df.limit(m).offset(n)`, try to push down `limit(m).offset(n)`. + // For example, `df.limit(5).offset(3)`, we can push down `limit(5).offset(3)`. + val isPushed = pushDownOffset(newChild, offsetValue) + if (isPushed) { + newChild + } else { + // For `df.limit(m).offset(n)`, only push down `limit(m)`. + // Keep the OFFSET operator if we failed to push down OFFSET to the data source. + offset.withNewChildren(Seq(newChild)) + } + } else { + // Keep the OFFSET operator if we can't remove LIMIT operator. + offset + } + case globalLimit @ OffsetAndLimit(offset, limit, child) => + // For `df.offset(n).limit(m)`, we can push down `limit(m + n)` first. + val (newChild, canRemoveLimit) = pushDownLimit(child, limit + offset) + if (canRemoveLimit) { + // Try to push down OFFSET only if the LIMIT operator has been pushed and can be removed. + // For `df.offset(n).limit(m)`, try to push down `limit(m + n).offset(n)`. + // For example, `df.offset(3).limit(5)`, we can push down `limit(8).offset(3)`. + val isPushed = pushDownOffset(newChild, offset) + if (isPushed) { + newChild + } else { + // Still keep the OFFSET operator if we can't push it down. + // For example, `df.offset(3).limit(5)`, `limit(8)` has been pushed + // and can be removed, Spark still do `offset(3)`. Review Comment: ```suggestion ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org