SPARKC-312: Implementing FilterOptimizer #1019
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,106 @@ | ||
| package org.apache.spark.sql.cassandra | ||
|
|
||
| import org.apache.spark.sql.sources._ | ||
|
|
||
|
|
||
| /** | ||
| * Optimizer will try to transform pushdown filter into `sum of products`. | ||
| * So that the filter like | ||
| * '(field1 < 3 OR field1 > 7) AND (field2 = 'val1' OR field2 = 'val2')' | ||
| * will become equivalent | ||
| * 'field1 < 3 AND field2 = "val1" OR field1 < 3 AND field2 = "val2" OR | ||
| * field1 > 7 AND field2 = "val1" OR field1 > 7 AND field2 = "val2"' | ||
| * | ||
| * @param filters Array of logical statements [[org.apache.spark.sql.sources.Filter]] | ||
| * that forms `where`-clause with `AND` operator, for example: | ||
| * val Array(f1, f2, ... fn) = ... // such that `where f1 AND f2 AND ... AND fn` | ||
| * | ||
| */ | ||
| class FiltersOptimizer(filters: Array[Filter]) { | ||
|
There was a problem hiding this comment. I'm a little confused why there is a separate class here, do we ever use this without calling .build() immediately after? |
||
|
|
||
| private val fullFilterAst = | ||
| if (filters.nonEmpty) Some(filters.reduce((left, right) => And(left, right))) else None | ||
|
|
||
| import FiltersOptimizer._ | ||
|
|
||
| def build(): List[Array[Filter]] = fullFilterAst match { | ||
| case Some(ast) => (toNNF andThen toDNF andThen traverse andThen groupByAnd).apply(ast) | ||
| case None => List.empty | ||
| } | ||
|
|
||
| } | ||
|
|
||
| object FiltersOptimizer{ | ||
|
|
||
| def apply(filters: Array[Filter]): FiltersOptimizer = new FiltersOptimizer(filters) | ||
|
|
||
| private[cassandra] def dist(predL: Filter, predR: Filter): Filter = (predL, predR) match { | ||
| case (Or(l, r), p) => Or(dist(l, p), dist(r, p)) | ||
| case (p, Or(l, r)) => Or(dist(p, l), dist(p, r)) | ||
| case (l, r) => And(l, r) | ||
| } | ||
|
|
||
| /** The 'toNNF' function converts expressions to negation normal form. This | ||
| * function is total: it's defined for all expressions, not just those which | ||
| * only use negation, conjunction and disjunction, although all expressions in | ||
| * negation normal form do in fact only use those connectives. | ||
| * | ||
| * Then de Morgan's laws are applied to convert negated | ||
| * conjunctions and disjunctions into the conjunction or disjunction of the | ||
| * negation of their conjuncts: ¬(φ ∧ ψ) is converted to (¬φ ∨ ¬ψ) | ||
| * while ¬(φ ∨ ψ) becomes (¬φ ∧ ¬ψ). | ||
| */ | ||
| private[cassandra] val toNNF: Filter => Filter = { | ||
| case a@(EqualTo(_, _) | EqualNullSafe(_, _) | GreaterThan(_, _) | | ||
| GreaterThanOrEqual(_, _) | LessThan(_, _) | LessThanOrEqual(_, _) | | ||
| In(_, _) | IsNull(_) | IsNotNull(_) | | ||
| StringStartsWith(_, _) | StringEndsWith(_, _) | StringContains(_, _)) => a | ||
| case a@Not(EqualTo(_, _) | EqualNullSafe(_, _) | In(_, _) | | ||
| StringStartsWith(_, _) | StringEndsWith(_, _) | StringContains(_, _)) => a | ||
| case Not(GreaterThan(a, v)) => LessThanOrEqual(a, v) | ||
| case Not(LessThanOrEqual(a, v)) => GreaterThan(a, v) | ||
| case Not(LessThan(a, v)) => GreaterThanOrEqual(a, v) | ||
| case Not(GreaterThanOrEqual(a, v)) => LessThan(a, v) | ||
| case Not(IsNull(a)) => IsNotNull(a) | ||
| case Not(IsNotNull(a)) => IsNull(a) | ||
| case Not(Not(p)) => p | ||
| case And(l, r) => And(toNNF(l), toNNF(r)) | ||
| case Not(And(l, r)) => toNNF(Or(Not(l), Not(r))) | ||
| case Or(l, r) => Or(toNNF(l), toNNF(r)) | ||
| case Not(Or(l, r)) => toNNF(And(Not(l), Not(r))) | ||
| case p => p | ||
| } | ||
|
|
||
| /** The 'toDNF' function converts expressions to disjunctive normal form: a | ||
| * disjunction of clauses, where a clause is a conjunction of literals | ||
| * (variables and negated variables). | ||
| * | ||
| * The conversion is carried out by first converting the expression into | ||
| * negation normal form, and then applying the distributive law. | ||
| */ | ||
| private[cassandra] val toDNF: Filter => Filter = { | ||
| case And(l, r) => dist(toDNF(l), toDNF(r)) | ||
| case Or(l, r) => Or(toDNF(l), toDNF(r)) | ||
| case p => p | ||
| } | ||
|
|
||
| /** | ||
| * Traverse over disjunctive clauses of AST | ||
| */ | ||
| private[cassandra] val traverse: Filter => List[Filter] = { | ||
| case Or(l, r) => traverse(l) ++ traverse(r) | ||
| case a => a :: Nil | ||
| } | ||
|
|
||
| /** | ||
| * Group all conjunctive clauses into Array[Filter] | ||
| * f1 && f2 && ... && fn => Array(f1, f2, ... fn) | ||
| */ | ||
| private[cassandra] val andToArray: Filter => Array[Filter] = { | ||
| case And(l, r) => andToArray(l) ++ andToArray(r) | ||
| case a => Array(a) | ||
| } | ||
|
|
||
| private[cassandra] val groupByAnd: List[Filter] => List[Array[Filter]] = _.map(andToArray) | ||
|
|
||
| } | ||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,63 @@ | ||
| package org.apache.spark.sql.cassandra | ||
|
|
||
| import org.apache.spark.sql.sources._ | ||
|
|
||
| import org.scalacheck._ | ||
| import org.scalacheck.Prop.forAll | ||
| import org.scalatest.prop.PropertyChecks | ||
| import org.scalatest.{FlatSpec, ShouldMatchers} | ||
|
|
||
| class FiltersOptimizerCheck extends FlatSpec with PropertyChecks with ShouldMatchers { | ||
|
|
||
| // For testing purpose | ||
| case object True extends Filter | ||
| case object False extends Filter | ||
|
|
||
| val genFullTree = for { | ||
| size <- Gen.choose(0, 500) | ||
| tree <- genTree(size) | ||
| } yield tree | ||
|
|
||
| def genTree(maxDepth: Int): Gen[Filter] = | ||
| if (maxDepth == 0) leaf else Gen.oneOf(leaf, genAnd(maxDepth), genOr(maxDepth), genNot(maxDepth)) | ||
|
|
||
| def genAnd(maxDepth: Int): Gen[Filter] = for { | ||
| depthL <- Gen.choose(0, maxDepth - 1) | ||
| depthR <- Gen.choose(0, maxDepth - 1) | ||
| left <- genTree(depthL) | ||
| right <- genTree(depthR) | ||
| } yield And(left, right) | ||
|
|
||
| def genOr(maxDepth: Int): Gen[Filter] = for { | ||
| depthL <- Gen.choose(0, maxDepth - 1) | ||
| depthR <- Gen.choose(0, maxDepth - 1) | ||
| left <- genTree(depthL) | ||
| right <- genTree(depthR) | ||
| } yield Or(left, right) | ||
|
|
||
| def genNot(maxDepth: Int): Gen[Filter] = for { | ||
| depth <- Gen.choose(0, maxDepth - 1) | ||
| expr <- genTree(depth) | ||
| } yield Not(expr) | ||
|
|
||
| def leaf: Gen[Filter] = Gen.oneOf(True, False) | ||
|
|
||
| /** | ||
| * Evaluate logical ADT | ||
| **/ | ||
| private def eval(clause: Filter): Boolean = clause match { | ||
| case And(left, right) => eval(left) && eval(right) | ||
| case Or(left, right) => eval(left) || eval(right) | ||
| case Not(predicate) => !eval(predicate) | ||
| case True => true | ||
| case False => false | ||
| } | ||
|
|
||
| "FiltersOptimizer" should "generate equivalent disjunction normal form for arbitrary logical statement" in { | ||
| forAll(genFullTree){ expr => | ||
| val dnf = (FiltersOptimizer.toNNF andThen FiltersOptimizer.toDNF).apply(expr) | ||
| assert(eval(dnf) == eval(expr)) | ||
| } | ||
| } | ||
|
|
||
| } |
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Replace the string here with the parameter defined in the conf file
EnableWhereClauseOptimizationParam.namejust in case we change things later :)