codope commented on code in PR #5328:
URL: https://github.com/apache/hudi/pull/5328#discussion_r927363643
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/execution/bulkinsert/RowCustomColumnsSortPartitioner.java:
##########
@@ -19,42 +19,69 @@
package org.apache.hudi.execution.bulkinsert;
import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.table.HoodieTableConfig;
import org.apache.hudi.config.HoodieWriteConfig;
-import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.sql.Column;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
+import scala.collection.JavaConverters;
import java.util.Arrays;
+import java.util.stream.Collectors;
+
+import static org.apache.hudi.common.util.ValidationUtils.checkState;
+import static
org.apache.hudi.execution.bulkinsert.RDDCustomColumnsSortPartitioner.getOrderByColumnNames;
/**
- * A partitioner that does sorting based on specified column values for each
spark partitions.
+ * A partitioner that does local sorting for each RDD partition based on the
tuple of
+ * values of the columns configured for ordering.
*/
-public class RowCustomColumnsSortPartitioner implements
BulkInsertPartitioner<Dataset<Row>> {
+public class RowCustomColumnsSortPartitioner extends
RepartitioningBulkInsertPartitionerBase<Dataset<Row>> {
+
+ private final String[] orderByColumnNames;
- private final String[] sortColumnNames;
+ public RowCustomColumnsSortPartitioner(HoodieWriteConfig config,
HoodieTableConfig tableConfig) {
+ super(tableConfig);
+ this.orderByColumnNames = getOrderByColumnNames(config);
- public RowCustomColumnsSortPartitioner(HoodieWriteConfig config) {
- this.sortColumnNames = getSortColumnName(config);
+ checkState(orderByColumnNames.length > 0);
}
- public RowCustomColumnsSortPartitioner(String[] columnNames) {
- this.sortColumnNames = columnNames;
+ public RowCustomColumnsSortPartitioner(String[] columnNames,
HoodieTableConfig tableConfig) {
+ super(tableConfig);
+ this.orderByColumnNames = columnNames;
+
+ checkState(orderByColumnNames.length > 0);
}
@Override
- public Dataset<Row> repartitionRecords(Dataset<Row> records, int
outputSparkPartitions) {
- final String[] sortColumns = this.sortColumnNames;
- return records.coalesce(outputSparkPartitions)
- .sortWithinPartitions(HoodieRecord.PARTITION_PATH_METADATA_FIELD,
sortColumns);
+ public Dataset<Row> repartitionRecords(Dataset<Row> dataset, int
outputSparkPartitions) {
+ Dataset<Row> repartitionedDataset;
+
+ // NOTE: In case of partitioned table even "global" ordering (across all
RDD partitions) could
+ // not change table's partitioning and therefore there's no point in
doing global sorting
+ // across "physical" partitions, and instead we can reduce total
amount of data being
+ // shuffled by doing do "local" sorting:
+ // - First, re-partitioning dataset such that "logical"
partitions are aligned w/
+ // "physical" ones
+ // - Sorting locally w/in RDD ("logical") partitions
+ //
+ // Non-partitioned tables will be globally sorted.
+ if (isPartitionedTable) {
+ repartitionedDataset = dataset.repartition(outputSparkPartitions, new
Column(HoodieRecord.PARTITION_PATH_METADATA_FIELD));
Review Comment:
Most tables will be partitioned. Isn't repartitioning just gonna increase
the amount of shuffle?
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/execution/bulkinsert/RepartitioningBulkInsertPartitionerBase.java:
##########
@@ -0,0 +1,62 @@
+/*
+ * 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.hudi.execution.bulkinsert;
+
+import org.apache.hudi.common.function.SerializableFunctionUnchecked;
+import org.apache.hudi.common.table.HoodieTableConfig;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+
+import java.io.Serializable;
+import java.util.Objects;
+
+/**
+ * Base class for any {@link BulkInsertPartitioner} implementation that does
re-partitioning,
+ * to better align "logical" (query-engine's partitioning of the incoming
dataset) w/ the table's
+ * "physical" partitioning
+ */
+public abstract class RepartitioningBulkInsertPartitionerBase<I> implements
BulkInsertPartitioner<I> {
+
+ protected final boolean isPartitionedTable;
+
+ public RepartitioningBulkInsertPartitionerBase(HoodieTableConfig
tableConfig) {
Review Comment:
instead of having tableConfig across executors, should we just define needed
attributes here?
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/execution/bulkinsert/BulkInsertInternalPartitionerFactory.java:
##########
@@ -27,16 +28,18 @@
*/
public abstract class BulkInsertInternalPartitionerFactory {
- public static BulkInsertPartitioner get(BulkInsertSortMode sortMode) {
- switch (sortMode) {
+ public static BulkInsertPartitioner get(BulkInsertSortMode bulkInsertMode,
HoodieTableConfig tableConfig) {
Review Comment:
I would avoid cosmetic changes in this PR. But, if you prefer to change the
name `bulkInsertSortMode` might be better.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
