聊聊flink Table的OrderBy及Limit

go4it發表於2019-01-31

序

本文主要研究一下flink Table的OrderBy及Limit

例項

Table in = tableEnv.fromDataSet(ds, "a, b, c");
Table result = in.orderBy("a.asc");

Table in = tableEnv.fromDataSet(ds, "a, b, c");

// returns the first 5 records from the sorted result
Table result1 = in.orderBy("a.asc").fetch(5); 

// skips the first 3 records and returns all following records from the sorted result
Table result2 = in.orderBy("a.asc").offset(3);

// skips the first 10 records and returns the next 5 records from the sorted result
Table result3 = in.orderBy("a.asc").offset(10).fetch(5);
複製程式碼

orderBy方法類似sql的order by；limit則由offset及fetch兩個方法構成，類似sql的offset及fetch

Table

flink-table_2.11-1.7.0-sources.jar!/org/apache/flink/table/api/table.scala

class Table(
    private[flink] val tableEnv: TableEnvironment,
    private[flink] val logicalPlan: LogicalNode) {

  //......

  def orderBy(fields: String): Table = {
    val parsedFields = ExpressionParser.parseExpressionList(fields)
    orderBy(parsedFields: _*)
  }

  def orderBy(fields: Expression*): Table = {
    val order: Seq[Ordering] = fields.map {
      case o: Ordering => o
      case e => Asc(e)
    }
    new Table(tableEnv, Sort(order, logicalPlan).validate(tableEnv))
  }

  def offset(offset: Int): Table = {
    new Table(tableEnv, Limit(offset, -1, logicalPlan).validate(tableEnv))
  }

  def fetch(fetch: Int): Table = {
    if (fetch < 0) {
      throw new ValidationException("FETCH count must be equal or larger than 0.")
    }
    this.logicalPlan match {
      case Limit(o, -1, c) =>
        // replace LIMIT without FETCH by LIMIT with FETCH
        new Table(tableEnv, Limit(o, fetch, c).validate(tableEnv))
      case Limit(_, _, _) =>
        throw new ValidationException("FETCH is already defined.")
      case _ =>
        new Table(tableEnv, Limit(0, fetch, logicalPlan).validate(tableEnv))
    }
  }

  //......
}
複製程式碼

Table的orderBy方法，支援String或Expression型別的引數，其中String型別最終是轉為Expression型別；orderBy方法最後使用Sort重新建立了Table；offset及fetch方法，使用Limit重新建立了Table(offset方法建立的Limit其fetch為-1；fetch方法如果之前沒有指定offset則建立的Limit的offset為0)

Sort

flink-table_2.11-1.7.0-sources.jar!/org/apache/flink/table/plan/logical/operators.scala

case class Sort(order: Seq[Ordering], child: LogicalNode) extends UnaryNode {
  override def output: Seq[Attribute] = child.output

  override protected[logical] def construct(relBuilder: RelBuilder): RelBuilder = {
    child.construct(relBuilder)
    relBuilder.sort(order.map(_.toRexNode(relBuilder)).asJava)
  }

  override def validate(tableEnv: TableEnvironment): LogicalNode = {
    if (tableEnv.isInstanceOf[StreamTableEnvironment]) {
      failValidation(s"Sort on stream tables is currently not supported.")
    }
    super.validate(tableEnv)
  }
}
複製程式碼

Sort繼承了UnaryNode，它的構造器接收Set型別的Ordering，其construct方法使用relBuilder.sort來構建sort條件

Ordering

flink-table_2.11-1.7.0-sources.jar!/org/apache/flink/table/expressions/ordering.scala

abstract class Ordering extends UnaryExpression {
  override private[flink] def validateInput(): ValidationResult = {
    if (!child.isInstanceOf[NamedExpression]) {
      ValidationFailure(s"Sort should only based on field reference")
    } else {
      ValidationSuccess
    }
  }
}

case class Asc(child: Expression) extends Ordering {
  override def toString: String = s"($child).asc"

  override private[flink] def toRexNode(implicit relBuilder: RelBuilder): RexNode = {
    child.toRexNode
  }

  override private[flink] def resultType: TypeInformation[_] = child.resultType
}

case class Desc(child: Expression) extends Ordering {
  override def toString: String = s"($child).desc"

  override private[flink] def toRexNode(implicit relBuilder: RelBuilder): RexNode = {
    relBuilder.desc(child.toRexNode)
  }

  override private[flink] def resultType: TypeInformation[_] = child.resultType
}
複製程式碼

Ordering是一個抽象類，它有Asc及Desc兩個子類

Limit