|
| 1 | +/* |
| 2 | + * Licensed to the Apache Software Foundation (ASF) under one or more |
| 3 | + * contributor license agreements. See the NOTICE file distributed with |
| 4 | + * this work for additional information regarding copyright ownership. |
| 5 | + * The ASF licenses this file to You under the Apache License, Version 2.0 |
| 6 | + * (the "License"); you may not use this file except in compliance with |
| 7 | + * the License. You may obtain a copy of the License at |
| 8 | + * |
| 9 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | + * |
| 11 | + * Unless required by applicable law or agreed to in writing, software |
| 12 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | + * See the License for the specific language governing permissions and |
| 15 | + * limitations under the License. |
| 16 | + */ |
| 17 | + |
| 18 | +package org.apache.spark.sql.catalyst.analysis |
| 19 | + |
| 20 | +import org.apache.spark.sql.QueryTest |
| 21 | +import org.apache.spark.sql.catalyst.expressions.NamedExpression |
| 22 | +import org.apache.spark.sql.catalyst.plans.logical.{Filter, Project} |
| 23 | +import org.apache.spark.sql.internal.SQLConf |
| 24 | +import org.apache.spark.sql.test.SharedSparkSession |
| 25 | + |
| 26 | +class AddMetadataColumnsSuite extends QueryTest with SharedSparkSession { |
| 27 | + |
| 28 | + test("Add only necessary metadata columns") { |
| 29 | + // For a query like: |
| 30 | + // |
| 31 | + // {{{ |
| 32 | + // SELECT t1.k |
| 33 | + // FROM VALUES(1,2) AS t1(k,v1) FULL OUTER JOIN VALUES(1,2) AS t2(k,v2) USING (k) |
| 34 | + // }}} |
| 35 | + // |
| 36 | + // the analyzed plan would look like: |
| 37 | + // Project [k#0] |
| 38 | + // +- Project [coalesce(k#0, k#2) AS k#4, v1#1, v2#3, k#0, k#2] |
| 39 | + // +- Join FullOuter, (k#0 = k#2) |
| 40 | + // :- SubqueryAlias nt1 |
| 41 | + // : +- LocalRelation [k#0, v1#1] |
| 42 | + // +- SubqueryAlias nt2 |
| 43 | + // +- LocalRelation [k#2, v2#3] |
| 44 | + // The inner project in this case contains a reference to k#2, which is not needed in the |
| 45 | + // top-most project. With `spark.sql.analyzer.uniqueNecessaryMetadataColumns` set to false, we |
| 46 | + // will add k#2 to the project list because it is a metadata column. Otherwise, we don't need |
| 47 | + // it and can avoid adding it in AddMetadataColumns rule. |
| 48 | + withTable("t1", "t2") { |
| 49 | + sql("CREATE TABLE t1(k INT, v1 INT)") |
| 50 | + sql("CREATE TABLE t2(k INT, v2 INT)") |
| 51 | + val left = sql("select * from t1") |
| 52 | + val right = sql("select * from t2") |
| 53 | + val join = left.join(right, Seq("k"), "full_outer") |
| 54 | + |
| 55 | + val rightKeyExprId = right |
| 56 | + .select(right("k")) |
| 57 | + .queryExecution |
| 58 | + .analyzed |
| 59 | + .asInstanceOf[Project] |
| 60 | + .projectList |
| 61 | + .head |
| 62 | + .exprId |
| 63 | + |
| 64 | + withSQLConf(SQLConf.ONLY_NECESSARY_AND_UNIQUE_METADATA_COLUMNS.key -> "true") { |
| 65 | + // Inner project list shouldn't contain a reference to the right key. |
| 66 | + val analyzed = join.select(left("k")).queryExecution.analyzed |
| 67 | + analyzed match { |
| 68 | + case Project(_, Project(innerProjectList: Seq[NamedExpression], _)) => |
| 69 | + assert(Seq("k", "v1", "v2", "k") == innerProjectList.map(_.name)) |
| 70 | + assert(!innerProjectList.map(_.exprId).contains(rightKeyExprId)) |
| 71 | + } |
| 72 | + } |
| 73 | + |
| 74 | + withSQLConf(SQLConf.ONLY_NECESSARY_AND_UNIQUE_METADATA_COLUMNS.key -> "false") { |
| 75 | + // Inner project list should contain a reference to the right key. |
| 76 | + val analyzed = join.select(left("k")).queryExecution.analyzed |
| 77 | + analyzed match { |
| 78 | + case Project(_, Project(innerProjectList: Seq[NamedExpression], _)) => |
| 79 | + assert(Seq("k", "v1", "v2", "k", "k") == innerProjectList.map(_.name)) |
| 80 | + assert(innerProjectList.map(_.exprId).contains(rightKeyExprId)) |
| 81 | + } |
| 82 | + } |
| 83 | + } |
| 84 | + } |
| 85 | + |
| 86 | + test("Add only unique metadata columns") { |
| 87 | + // For a query like: |
| 88 | + // {{{ |
| 89 | + // SELECT t1.k |
| 90 | + // FROM VALUES(1,2) AS t1(k, v1) FULL OUTER JOIN VALUES(1,2) AS t2(k,v2) USING (k) |
| 91 | + // WHERE t1.k IS NOT NULL |
| 92 | + // }}} |
| 93 | + // |
| 94 | + // the analyzed plan will look like: |
| 95 | + // Project [k#0] |
| 96 | + // +- Project [k#4, v1#1, v2#3, k#0, k#2] |
| 97 | + // +- Filter isnotnull(k#0) |
| 98 | + // +- Project [coalesce(k#0, k#2) AS k#4, v1#1, v2#3, k#0, k#0, k#2] |
| 99 | + // +- Join FullOuter, (k#0 = k#2) |
| 100 | + // :- SubqueryAlias t1 |
| 101 | + // : +- LocalRelation [k#0, v1#1] |
| 102 | + // +- SubqueryAlias t2 |
| 103 | + // +- LocalRelation [k#2, v2#3] |
| 104 | + // |
| 105 | + // In this case, the Project under Filter contains a duplicate #k#0 attribute reference as well |
| 106 | + // as an unnecessary k#2 attribute reference. Additionally, the second top-most Project has an |
| 107 | + // extra k#2 that can also be removed. Duplicate reference comes from the fact that this |
| 108 | + // attribute will first be added by ResolveReferences rule as missing input, but |
| 109 | + // AddMetadataColumns doesn't respect the fact that this attribute already exists in the |
| 110 | + // project list and duplicates it. With `spark.sql.analyzer.uniqueNecessaryMetadataColumns` set |
| 111 | + // to true, we remove this duplication and the unnecessary attribute. |
| 112 | + withTable("t1", "t2") { |
| 113 | + sql("CREATE TABLE t1(k INT, v1 INT)") |
| 114 | + sql("CREATE TABLE t2(k INT, v2 INT)") |
| 115 | + val left = sql("select * from t1") |
| 116 | + val right = sql("select * from t2") |
| 117 | + val join = left.join(right, Seq("k"), "full_outer") |
| 118 | + val filter = join.filter(left("k").isNull) |
| 119 | + |
| 120 | + val leftKeyExprId = left |
| 121 | + .select(left("k")) |
| 122 | + .queryExecution |
| 123 | + .analyzed |
| 124 | + .asInstanceOf[Project] |
| 125 | + .projectList |
| 126 | + .head |
| 127 | + .exprId |
| 128 | + val rightKeyExprId = right |
| 129 | + .select(right("k")) |
| 130 | + .queryExecution |
| 131 | + .analyzed |
| 132 | + .asInstanceOf[Project] |
| 133 | + .projectList |
| 134 | + .head |
| 135 | + .exprId |
| 136 | + |
| 137 | + withSQLConf(SQLConf.ONLY_NECESSARY_AND_UNIQUE_METADATA_COLUMNS.key -> "true") { |
| 138 | + // With conf on, no duplication of left key and no unnecessary right key. |
| 139 | + val analyzed = filter.select(left("k")).queryExecution.analyzed |
| 140 | + analyzed match { |
| 141 | + case Project( |
| 142 | + _, |
| 143 | + Project(_, Filter(_, Project(innerProjectList: Seq[NamedExpression], _))) |
| 144 | + ) => |
| 145 | + assert(Seq("k", "v1", "v2", "k") == innerProjectList.map(_.name)) |
| 146 | + assert(innerProjectList.map(_.exprId).count(_ == rightKeyExprId) == 0) |
| 147 | + assert(innerProjectList.map(_.exprId).count(_ == leftKeyExprId) == 1) |
| 148 | + } |
| 149 | + } |
| 150 | + |
| 151 | + withSQLConf(SQLConf.ONLY_NECESSARY_AND_UNIQUE_METADATA_COLUMNS.key -> "false") { |
| 152 | + // With conf off, duplication of left key and an unnecessary right key. |
| 153 | + val analyzed = filter.select(left("k")).queryExecution.analyzed |
| 154 | + analyzed match { |
| 155 | + case Project( |
| 156 | + _, |
| 157 | + Project(_, Filter(_, Project(innerProjectList: Seq[NamedExpression], _))) |
| 158 | + ) => |
| 159 | + assert(Seq("k", "v1", "v2", "k", "k", "k") == innerProjectList.map(_.name)) |
| 160 | + assert(innerProjectList.map(_.exprId).count(_ == rightKeyExprId) == 1) |
| 161 | + assert(innerProjectList.map(_.exprId).count(_ == leftKeyExprId) == 2) |
| 162 | + } |
| 163 | + } |
| 164 | + } |
| 165 | + } |
| 166 | +} |
0 commit comments