ResolveWindowFrame Logical Evaluation Rule

ResolveWindowFrame is a logical evaluation rule that Spark SQL’s logical query plan analyzer uses to validate and resolve WindowExpression Catalyst logical expressions.

ResolveWindowFrame is a part of Resolution fixed-point batch of rules.

ResolveWindowFrame takes a logical plan and does the following:

  1. Makes sure that the window frame of a WindowFunction is unspecified or matches the SpecifiedWindowFrame of the WindowSpecDefinition expression.

    Reports a AnalysisException when the frames do not match:

    Window Frame [f] must match the required frame [frame]
  2. Copies the frame specification of WindowFunction to WindowSpecDefinition

  3. Creates a new SpecifiedWindowFrame for WindowExpression with the resolved Catalyst expression and UnspecifiedFrame

ResolveWindowFrame is a Scala object inside Analyzer class.
import import org.apache.spark.sql.expressions.Window
// cume_dist requires ordered windows
val q = spark.
  withColumn("cume_dist", cume_dist() over Window.orderBy("id"))
import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
val planBefore: LogicalPlan = q.queryExecution.logical

// Before ResolveWindowFrame
scala> println(planBefore.numberedTreeString)
00 'Project [*, cume_dist() windowspecdefinition('id ASC NULLS FIRST, UnspecifiedFrame) AS cume_dist#39]
01 +- Range (0, 5, step=1, splits=Some(8))

import spark.sessionState.analyzer.ResolveWindowFrame
val planAfter = ResolveWindowFrame.apply(plan)

// After ResolveWindowFrame
scala> println(planAfter.numberedTreeString)
00 'Project [*, cume_dist() windowspecdefinition('id ASC NULLS FIRST, RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS cume_dist#31]
01 +- Range (0, 5, step=1, splits=Some(8))

results matching ""

    No results matching ""