tmnd1991 commented on code in PR #9233: URL: https://github.com/apache/iceberg/pull/9233#discussion_r1433893181
########## spark/v3.5/spark-extensions/src/test/java/org/apache/iceberg/spark/extensions/TestSPJWithBucketing.java: ########## @@ -0,0 +1,219 @@ +/* + * 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.iceberg.spark.extensions; + +import java.util.Map; +import java.util.concurrent.TimeoutException; +import java.util.concurrent.atomic.AtomicInteger; +import java.util.function.Consumer; +import org.apache.iceberg.relocated.com.google.common.collect.ImmutableMap; +import org.apache.iceberg.spark.SparkCatalogConfig; +import org.apache.iceberg.spark.SparkSQLProperties; +import org.apache.spark.scheduler.SparkListener; +import org.apache.spark.scheduler.SparkListenerTaskEnd; +import org.apache.spark.sql.Dataset; +import org.apache.spark.sql.Row; +import org.apache.spark.sql.SparkSession; +import org.apache.spark.sql.internal.SQLConf; +import org.assertj.core.api.Assertions; +import org.junit.After; +import org.junit.Before; +import org.junit.Test; +import org.junit.runners.Parameterized; + +public class TestSPJWithBucketing extends SparkExtensionsTestBase { + + @Test + public void testMergeSPJwithCondition() { + testWithCondition( + " AND (" + + "(t.year_month='202306' AND t.day='01' AND testhive.system.bucket(4, t.id) = 0) OR\n" + + "(t.year_month='202306' AND t.day='01' AND testhive.system.bucket(4, t.id) = 1) OR\n" + + "(t.year_month='202306' AND t.day='02' AND testhive.system.bucket(4, t.id) = 0) OR\n" + + "(t.year_month='202307' AND t.day='01' AND testhive.system.bucket(4, t.id) = 3)\n" + + ")"); + } + + @Test + public void testMergeSPJwithoutCondition() { + testWithCondition(""); + } + + private void testWithCondition(String condition) { + createPartitionedTable(spark, targetTableName); + insertRecords(spark, targetTableName); + createPartitionedTable(spark, sourceTableName); + insertRecordsToUpdate(spark, sourceTableName); + int tasks = + executeAndCountTasks( + spark, + (s) -> + withSQLConf( + ENABLED_SPJ_SQL_CONF, + () -> + // id STRING, year_month STRING, day STRING, data STRING + sql( + s, + "MERGE INTO %s t USING (SELECT * FROM %s) s \n" + + "ON t.id = s.id AND t.year_month = s.year_month AND t.day = s.day\n" + + "%s\n" + + "WHEN MATCHED THEN UPDATE SET\n" + + " t.data = s.data\n" + + "WHEN NOT MATCHED THEN INSERT *", + targetTableName, + sourceTableName, + condition))); + long affectedPartitions = + sql(spark, "SELECT DISTINCT(partition) FROM %s.files", sourceTableName).count(); + int shufflePartitions = Integer.parseInt(spark.conf().get("spark.sql.shuffle.partitions")); + Assertions.assertThat(tasks).isEqualTo(affectedPartitions * 2 + shufflePartitions); Review Comment: sure. the target table is created with the following partitions (year_month, day, bucket(4, id)): - **202306/01/0** - **202306/01/1** - 202306/01/2 - 202306/01/3 - **202306/02/0** - 202306/02/1 - **202307/01/3** the source table is created with the following partitions: - 202306/01/0 - 202306/01/1 - 202306/02/0 - 202307/01/3 so the source table partitions are (is) a subset of the target table partitions. <u>Spark **statically knows** that info, because it's part of the metadata that iceberg keeps.</u> So copy-on-write "merge" consists of 2 jobs: 1. left-semi join to understand which files are affected by the merge 2. full-outer join where on the left side discards all the files not found while executing 1 In our particular case (where we know that source table partitions are a subset of target table partitions) if we do that with a Storage Partitioned Join, the most efficient way to do it is to: 1. create 1 task for each partition that will change, read all the files from both tables, join locally, collect the file names 2. create 1 task for each partition that will change, for each task, read all the files from the target table / partition except the ones that will not change (that's the effect of the IN), read all the files from the source table, join and apply merge logic locally, write out new files, add these files to the snapshot and remove the original files from the snapshot Disclaimer: I know very little about internals and I can only imagine how hard can this be to actually done like that, but I'm quite sure that is "logically" doable 😄 so the reasoning of the number of tasks is: - 1 task per partition that is going to change - `spark.sql.shuffle.partitions` to shuffle the file list (which I thought could be broadcasted, but I think it's not important right now) - 1 task per partition that is going to change to actually rewrite it Let me know if there is any fallacy in my reasoning -- 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. To unsubscribe, e-mail: issues-unsubscr...@iceberg.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@iceberg.apache.org For additional commands, e-mail: issues-h...@iceberg.apache.org
