MaxGekk commented on a change in pull request #27366: URL: https://github.com/apache/spark/pull/27366#discussion_r452435706
########## File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JsonFilters.scala ########## @@ -0,0 +1,157 @@ +/* + * 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.catalyst.json + +import org.apache.spark.sql.catalyst.{InternalRow, StructFilters} +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.sources +import org.apache.spark.sql.types.StructType + +/** + * The class provides API for applying pushed down source filters to rows with + * a struct schema parsed from JSON records. The class should be used in this way: + * 1. Before processing of the next row, `JacksonParser` (parser for short) resets the internal + * state of JsonFilter` by calling the `reset()` method. + * 2. The parser reads JSON fields one-by-one in streaming fashion. It converts an incoming + * field value to the desired type from the schema. After that, it sets the value to an instance + * of `InternalRow` at the position according to the schema. Order of parsed JSON fields can + * be different from the order in the schema. + * 3. Per every JSON field of the top-level JSON object, the parser calls `skipRow` by passing + * an `InternalRow` in which some of fields can be already set, and the position of the JSON + * field according to the schema. + * 3.1 `skipRow` finds a group of predicates that refers to this JSON field. + * 3.2 Per each predicate from the group, `skipRow` decrements its reference counter. + * 3.2.1 If predicate reference counter becomes 0, it means that all predicate attributes have + * been already set in the internal row, and the predicate can be applied to it. `skipRow` + * invokes the predicate for the row. + * 3.3 `skipRow` applies predicates until one of them returns `false`. In that case, the method + * returns `true` to the parser. + * 3.4 If all predicates with zero reference counter return `true`, the final result of + * the method is `false` which tells the parser to not skip the row. + * 4. If the parser gets `true` from `JsonFilters.skipRow`, it must not call the method anymore + * for this internal row, and should go the step 1. + * + * JsonFilters assumes that: + * - reset() is called before any skipRow() calls for new row. + * - skipRow() can be called for any valid index of the struct fields, + * and in any order. + * - After skipRow() returns `true`, the internal state of `JsonFilter` can be inconsistent, + * so, skipRow() must not be called for the current row anymore without reset(). + * + * @param pushedFilters The pushed down source filters. The filters should refer to + * the fields of the provided schema. + * @param schema The required schema of records from datasource files. + */ +class JsonFilters(pushedFilters: Seq[sources.Filter], schema: StructType) + extends StructFilters(pushedFilters, schema) { + + /** + * Stateful JSON predicate that keeps track of its dependent references in the + * current row via `refCount`. + * + * @param predicate The predicate compiled from pushed down source filters. + * @param totalRefs The total amount of all filters references which the predicate + * compiled from. + * @param refCount The current number of predicate references in the row that have + * been not set yet. When `refCount` reaches zero, the predicate + * has all dependencies are set, and can be applied to the row. + */ + case class JsonPredicate(predicate: BasePredicate, totalRefs: Int, var refCount: Int) { + def reset(): Unit = { + refCount = totalRefs + } + } + + // Predicates compiled from the pushed down filters. The predicates are grouped by their + // attributes. The i-th group contains predicates that refer to the i-th field of the given + // schema. A predicates can be places to many groups if it has many attributes. For example: + // schema: i INTEGER, s STRING + // filters: IsNotNull("i"), AlwaysTrue, And(EqualTo("i", 0), StringStartsWith("s", "abc")) + // predicates: + // 0: Array(IsNotNull("i"), AlwaysTrue, And(EqualTo("i", 0), StringStartsWith("s", "abc"))) + // 1: Array(AlwaysTrue, And(EqualTo("i", 0), StringStartsWith("s", "abc"))) + private val predicates: Array[Array[JsonPredicate]] = { + val groupedPredicates = Array.fill(schema.length)(Array.empty[JsonPredicate]) + if (SQLConf.get.jsonFilterPushDown) { + val groupedByRefSet = filters + // Group filters that have the same set of references. For example: + // IsNotNull("i") -> Set("i"), AlwaysTrue -> Set(), + // And(EqualTo("i", 0), StringStartsWith("s", "abc")) -> Set("i", "s") + // By grouping filters we could avoid tracking their state of references in the + // current row separately. + .groupBy(_.references.toSet) + // Combine all filters from the same group by `And` because all filters should + // return `true` to do not skip a row. The result is compiled to a predicate. + .map { case (refSet, refsFilters) => + (refSet, JsonPredicate(toPredicate(refsFilters), refSet.size, 0)) Review comment: It doesn't matter, actually because `reset()` must be invoke before any `skipRow()` calls. See my comments `JsonFilters assumes that:` ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org
