Github user cloud-fan commented on a diff in the pull request:

    https://github.com/apache/spark/pull/13494#discussion_r68882374
  
    --- Diff: 
sql/core/src/main/scala/org/apache/spark/sql/execution/MetadataOnlyOptimizer.scala
 ---
    @@ -0,0 +1,171 @@
    +/*
    + * Licensed to the Apache Software Foundation (ASF) under one or more
    + * contributor license agreements.  See the NOTICE file distributed with
    + * this work for additional information regarding copyright ownership.
    + * The ASF licenses this file to You under the Apache License, Version 2.0
    + * (the "License"); you may not use this file except in compliance with
    + * the License.  You may obtain a copy of the License at
    + *
    + *    http://www.apache.org/licenses/LICENSE-2.0
    + *
    + * Unless required by applicable law or agreed to in writing, software
    + * distributed under the License is distributed on an "AS IS" BASIS,
    + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    + * See the License for the specific language governing permissions and
    + * limitations under the License.
    + */
    +
    +package org.apache.spark.sql.execution
    +
    +import org.apache.spark.sql.{AnalysisException, SparkSession}
    +import org.apache.spark.sql.catalyst.InternalRow
    +import org.apache.spark.sql.catalyst.catalog.{CatalogRelation, 
SessionCatalog}
    +import org.apache.spark.sql.catalyst.expressions._
    +import org.apache.spark.sql.catalyst.expressions.aggregate._
    +import org.apache.spark.sql.catalyst.plans.logical._
    +import org.apache.spark.sql.catalyst.rules.Rule
    +import org.apache.spark.sql.execution.datasources.{HadoopFsRelation, 
LogicalRelation}
    +
    +/**
    + * When scanning only partition columns, get results based on metadata 
without scanning files.
    + * It is used for distinct, distinct aggregations or distinct-like 
aggregations(example: Max/Min).
    + * First of all, scanning only partition columns are required, then the 
rule does the following
    + * things here:
    + * 1. aggregate expression is partition columns,
    + *  e.g. SELECT col FROM tbl GROUP BY col or SELECT col FROM tbl GROUP BY 
cube(col).
    + * 2. aggregate function on partition columns with DISTINCT,
    + *  e.g. SELECT count(DISTINCT col) FROM tbl GROUP BY col.
    + * 3. aggregate function on partition columns which have same result with 
DISTINCT keyword.
    + *  e.g. SELECT Max(col2) FROM tbl GROUP BY col1.
    + */
    +case class MetadataOnlyOptimizer(
    +    sparkSession: SparkSession,
    +    catalog: SessionCatalog) extends Rule[LogicalPlan] {
    +
    +  private def canSupportMetadataOnly(a: Aggregate): Boolean = {
    +    val aggregateExpressions = a.aggregateExpressions.flatMap { expr =>
    +      expr.collect {
    +        case agg: AggregateExpression => agg
    +      }
    +    }.distinct
    +    if (aggregateExpressions.isEmpty) {
    +      // Support for aggregate that has no aggregateFunction when 
expressions are partition columns
    +      // example: select partitionCol from table group by partitionCol.
    +      // Moreover, multiple-distinct has been rewritted into it by 
RewriteDistinctAggregates.
    +      true
    +    } else {
    +      aggregateExpressions.forall { agg =>
    +        if (agg.isDistinct) {
    +          true
    +        } else {
    +          // If function can be evaluated on just the distinct values of a 
column, it can be used
    +          // by metadata-only optimizer.
    +          agg.aggregateFunction match {
    +            case max: Max => true
    +            case min: Min => true
    +            case hyperLog: HyperLogLogPlusPlus => true
    +            case _ => false
    +          }
    +        }
    +      }
    +    }
    +  }
    +
    +  private def convertLogicalToMetadataOnly(
    +      project: LogicalPlan,
    +      filter: Option[Expression],
    +      logical: LogicalRelation,
    +      files: HadoopFsRelation): LogicalPlan = {
    +    val attributeMap = logical.output.map(attr => (attr.name, attr)).toMap
    +    val partitionColumns = files.partitionSchema.map { field =>
    +      attributeMap.getOrElse(field.name, throw new AnalysisException(
    +        s"Unable to resolve ${field.name} given 
[${logical.output.map(_.name).mkString(", ")}]"))
    +    }
    +    val projectSet = filter.map(project.references ++ 
_.references).getOrElse(project.references)
    +    if (projectSet.subsetOf(AttributeSet(partitionColumns))) {
    +      val selectedPartitions = 
files.location.listFiles(filter.map(Seq(_)).getOrElse(Seq.empty))
    +      val valuesRdd = 
sparkSession.sparkContext.parallelize(selectedPartitions.map(_.values), 1)
    +      val valuesPlan = LogicalRDD(partitionColumns, 
valuesRdd)(sparkSession)
    +      valuesPlan
    +    } else {
    +      logical
    +    }
    +  }
    +
    +  private def convertCatalogToMetadataOnly(
    +      project: LogicalPlan,
    +      filter: Option[Expression],
    +      relation: CatalogRelation): LogicalPlan = {
    +    val attributeMap = relation.output.map(attr => (attr.name, attr)).toMap
    +    val partitionColumns = relation.catalogTable.partitionColumnNames.map 
{ column =>
    +      attributeMap.getOrElse(column, throw new AnalysisException(
    +        s"Unable to resolve ${column} given 
[${relation.output.map(_.name).mkString(", ")}]"))
    +    }
    +    val projectSet = filter.map(project.references ++ 
_.references).getOrElse(project.references)
    +    if (projectSet.subsetOf(AttributeSet(partitionColumns))) {
    +      val partitionColumnDataTypes = partitionColumns.map(_.dataType)
    +      val partitionValues = 
catalog.listPartitions(relation.catalogTable.identifier)
    +        .map { p =>
    +          InternalRow.fromSeq(
    +            partitionColumns.map(a => 
p.spec(a.name)).zip(partitionColumnDataTypes).map {
    +              case (rawValue, dataType) => Cast(Literal(rawValue), 
dataType).eval(null)
    +            })
    +        }
    +      val valuesRdd = 
sparkSession.sparkContext.parallelize(partitionValues, 1)
    +      val valuesPlan = LogicalRDD(partitionColumns, 
valuesRdd)(sparkSession)
    +      valuesPlan
    +    } else {
    +      relation
    +    }
    +  }
    +
    +  private def convertToMetadataOnly(plan: LogicalPlan): LogicalPlan = plan 
match {
    --- End diff --
    
    We need to think about it more carefully, i.e. how can the partition 
information propagate up from table relation?
    It's obvious that `Filter` can retain all partition information, but for 
others, it's not trivial to explain.
    
    Since this PR definitely need more people to review, how about we only 
handle `Filter` for now and improve it later? Then it's easier for other people 
to review and get this PR in. Thanks!


---
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

Reply via email to