You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hudi.apache.org by "nsivabalan (via GitHub)" <gi...@apache.org> on 2023/04/30 17:18:04 UTC
[GitHub] [hudi] nsivabalan commented on a diff in pull request #8493: [HUDI-6098] Use bulk insert prepped for the initial write into MDT.
nsivabalan commented on code in PR #8493:
URL: https://github.com/apache/hudi/pull/8493#discussion_r1181264755
##########
hudi-common/src/main/java/org/apache/hudi/metadata/HoodieTableMetadataUtil.java:
##########
@@ -1004,6 +1004,57 @@ public static int mapRecordKeyToFileGroupIndex(String recordKey, int numFileGrou
return Math.abs(Math.abs(h) % numFileGroups);
}
+ /**
+ * Return the complete fileID for a file group within a MDT partition.
+ *
+ * MDT fileGroups have the format <fileIDPrefix>-<index>. The fileIDPrefix is hardcoded for each MDT partition and index is an integer.
+ *
+ * @param partitionType The type of the MDT partition
+ * @param index Index of the file group within the partition
+ * @return The fileID
+ */
+ public static String getFileIDForFileGroup(MetadataPartitionType metadataPartition, int index) {
Review Comment:
do we have UTs for these.
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/metadata/SparkHoodieMetadataBulkInsertPartitioner.java:
##########
@@ -0,0 +1,106 @@
+/*
+ * 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.metadata;
+
+import java.io.Serializable;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.util.ValidationUtils;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+import org.apache.spark.api.java.JavaRDD;
+
+import scala.Tuple2;
+
+/**
+ * A {@code BulkInsertPartitioner} implementation for Metadata Table to improve performance of initialization of metadata
+ * table partition when a very large number of records are inserted.
+ *
+ * This partitioner requires the records to be already tagged with location.
+ */
+public class SparkHoodieMetadataBulkInsertPartitioner implements BulkInsertPartitioner<JavaRDD<HoodieRecord>> {
+ private class FileGroupPartitioner extends Partitioner {
+
+ @Override
+ public int getPartition(Object key) {
+ return ((Tuple2<Integer, String>)key)._1;
+ }
+
+ @Override
+ public int numPartitions() {
+ // Max number of file groups supported per partition in MDT. Refer to HoodieTableMetadataUtil.getFileIDForFileGroup()
+ return 10000;
+ }
+ }
+
+ // FileIDs for the various partitions
+ private List<String> fileIDPfxs;
+
+ /**
+ * Partition the records by their location. The number of partitions is determined by the number of MDT fileGroups being udpated rather than the
+ * specific value of outputSparkPartitions.
+ */
+ @Override
+ public JavaRDD<HoodieRecord> repartitionRecords(JavaRDD<HoodieRecord> records, int outputSparkPartitions) {
+ Comparator<Tuple2<Integer, String>> keyComparator = (Comparator<Tuple2<Integer, String>> & Serializable)(t1, t2) -> {
+ return t1._2.compareTo(t2._2);
+ };
+
+ // Partition the records by their file group
+ JavaRDD<HoodieRecord> partitionedRDD = records
+ // key by <file group index, record key>. The file group index is used to partition and the record key is used to sort within the partition.
Review Comment:
from what I glean, partitioning is based on fileGroupIndex disregarding the MDT partition. So, tell me something.
if we have 2 file groups in col stats and 2 file groups for RLI, does 1st file group for both col stats and RLI belong to same partition in this repartition call ?
should the partitioning be based on fileId itself and sorting within that can be based on the record keys within each partition. or am I missing something ?
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/metadata/SparkHoodieMetadataBulkInsertPartitioner.java:
##########
@@ -0,0 +1,106 @@
+/*
+ * 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.metadata;
+
+import java.io.Serializable;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.util.ValidationUtils;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+import org.apache.spark.api.java.JavaRDD;
+
+import scala.Tuple2;
+
+/**
+ * A {@code BulkInsertPartitioner} implementation for Metadata Table to improve performance of initialization of metadata
+ * table partition when a very large number of records are inserted.
+ *
+ * This partitioner requires the records to be already tagged with location.
+ */
+public class SparkHoodieMetadataBulkInsertPartitioner implements BulkInsertPartitioner<JavaRDD<HoodieRecord>> {
+ private class FileGroupPartitioner extends Partitioner {
+
+ @Override
+ public int getPartition(Object key) {
+ return ((Tuple2<Integer, String>)key)._1;
+ }
+
+ @Override
+ public int numPartitions() {
+ // Max number of file groups supported per partition in MDT. Refer to HoodieTableMetadataUtil.getFileIDForFileGroup()
+ return 10000;
+ }
+ }
+
+ // FileIDs for the various partitions
+ private List<String> fileIDPfxs;
+
+ /**
+ * Partition the records by their location. The number of partitions is determined by the number of MDT fileGroups being udpated rather than the
+ * specific value of outputSparkPartitions.
+ */
+ @Override
+ public JavaRDD<HoodieRecord> repartitionRecords(JavaRDD<HoodieRecord> records, int outputSparkPartitions) {
+ Comparator<Tuple2<Integer, String>> keyComparator = (Comparator<Tuple2<Integer, String>> & Serializable)(t1, t2) -> {
+ return t1._2.compareTo(t2._2);
+ };
+
+ // Partition the records by their file group
+ JavaRDD<HoodieRecord> partitionedRDD = records
+ // key by <file group index, record key>. The file group index is used to partition and the record key is used to sort within the partition.
+ .keyBy(r -> {
+ int fileGroupIndex = HoodieTableMetadataUtil.getFileGroupIndexFromFileId(r.getCurrentLocation().getFileId());
+ return new Tuple2<Integer, String>(fileGroupIndex, r.getRecordKey());
+ })
+ .repartitionAndSortWithinPartitions(new FileGroupPartitioner(), keyComparator)
Review Comment:
we know the total partitions is going to be equal to total file groups. can we override the "numPartitions" for FileGroupPartitioner?
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/metadata/SparkHoodieMetadataBulkInsertPartitioner.java:
##########
@@ -0,0 +1,106 @@
+/*
+ * 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.metadata;
+
+import java.io.Serializable;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.util.ValidationUtils;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+import org.apache.spark.api.java.JavaRDD;
+
+import scala.Tuple2;
+
+/**
+ * A {@code BulkInsertPartitioner} implementation for Metadata Table to improve performance of initialization of metadata
+ * table partition when a very large number of records are inserted.
+ *
+ * This partitioner requires the records to be already tagged with location.
+ */
+public class SparkHoodieMetadataBulkInsertPartitioner implements BulkInsertPartitioner<JavaRDD<HoodieRecord>> {
Review Comment:
do we have UT for this ?
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/metadata/SparkHoodieMetadataBulkInsertPartitioner.java:
##########
@@ -0,0 +1,106 @@
+/*
+ * 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.metadata;
+
+import java.io.Serializable;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.util.ValidationUtils;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+import org.apache.spark.api.java.JavaRDD;
+
+import scala.Tuple2;
+
+/**
+ * A {@code BulkInsertPartitioner} implementation for Metadata Table to improve performance of initialization of metadata
+ * table partition when a very large number of records are inserted.
+ *
+ * This partitioner requires the records to be already tagged with location.
+ */
+public class SparkHoodieMetadataBulkInsertPartitioner implements BulkInsertPartitioner<JavaRDD<HoodieRecord>> {
+ private class FileGroupPartitioner extends Partitioner {
+
+ @Override
+ public int getPartition(Object key) {
+ return ((Tuple2<Integer, String>)key)._1;
+ }
+
+ @Override
+ public int numPartitions() {
+ // Max number of file groups supported per partition in MDT. Refer to HoodieTableMetadataUtil.getFileIDForFileGroup()
+ return 10000;
+ }
+ }
+
+ // FileIDs for the various partitions
+ private List<String> fileIDPfxs;
+
+ /**
+ * Partition the records by their location. The number of partitions is determined by the number of MDT fileGroups being udpated rather than the
+ * specific value of outputSparkPartitions.
+ */
+ @Override
+ public JavaRDD<HoodieRecord> repartitionRecords(JavaRDD<HoodieRecord> records, int outputSparkPartitions) {
+ Comparator<Tuple2<Integer, String>> keyComparator = (Comparator<Tuple2<Integer, String>> & Serializable)(t1, t2) -> {
+ return t1._2.compareTo(t2._2);
+ };
+
+ // Partition the records by their file group
+ JavaRDD<HoodieRecord> partitionedRDD = records
+ // key by <file group index, record key>. The file group index is used to partition and the record key is used to sort within the partition.
Review Comment:
reason being, the argument "JavaRDD<HoodieRecord> records " to this method could contain records for N no of partitions in MDT. not sure if we are making any assumptions on that.
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/metadata/SparkHoodieMetadataBulkInsertPartitioner.java:
##########
@@ -0,0 +1,106 @@
+/*
+ * 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.metadata;
+
+import java.io.Serializable;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.hudi.common.model.HoodieRecord;
+import org.apache.hudi.common.util.ValidationUtils;
+import org.apache.hudi.table.BulkInsertPartitioner;
+import org.apache.spark.Partitioner;
+import org.apache.spark.api.java.JavaRDD;
+
+import scala.Tuple2;
+
+/**
+ * A {@code BulkInsertPartitioner} implementation for Metadata Table to improve performance of initialization of metadata
+ * table partition when a very large number of records are inserted.
+ *
+ * This partitioner requires the records to be already tagged with location.
+ */
+public class SparkHoodieMetadataBulkInsertPartitioner implements BulkInsertPartitioner<JavaRDD<HoodieRecord>> {
+ private class FileGroupPartitioner extends Partitioner {
+
+ @Override
+ public int getPartition(Object key) {
+ return ((Tuple2<Integer, String>)key)._1;
+ }
+
+ @Override
+ public int numPartitions() {
+ // Max number of file groups supported per partition in MDT. Refer to HoodieTableMetadataUtil.getFileIDForFileGroup()
+ return 10000;
+ }
+ }
+
+ // FileIDs for the various partitions
+ private List<String> fileIDPfxs;
+
+ /**
+ * Partition the records by their location. The number of partitions is determined by the number of MDT fileGroups being udpated rather than the
+ * specific value of outputSparkPartitions.
+ */
+ @Override
+ public JavaRDD<HoodieRecord> repartitionRecords(JavaRDD<HoodieRecord> records, int outputSparkPartitions) {
+ Comparator<Tuple2<Integer, String>> keyComparator = (Comparator<Tuple2<Integer, String>> & Serializable)(t1, t2) -> {
+ return t1._2.compareTo(t2._2);
+ };
+
+ // Partition the records by their file group
+ JavaRDD<HoodieRecord> partitionedRDD = records
+ // key by <file group index, record key>. The file group index is used to partition and the record key is used to sort within the partition.
+ .keyBy(r -> {
+ int fileGroupIndex = HoodieTableMetadataUtil.getFileGroupIndexFromFileId(r.getCurrentLocation().getFileId());
+ return new Tuple2<Integer, String>(fileGroupIndex, r.getRecordKey());
+ })
+ .repartitionAndSortWithinPartitions(new FileGroupPartitioner(), keyComparator)
+ .map(t -> t._2);
+
+ fileIDPfxs = partitionedRDD.mapPartitions(recordItr -> {
+ // Due to partitioning, all record in the partition should have same fileID. So we only can get the fileID prefix from the first record.
+ List<String> fileIds = new ArrayList<>(1);
+ if (recordItr.hasNext()) {
+ HoodieRecord record = recordItr.next();
+ final String fileID = HoodieTableMetadataUtil.getFileGroupPrefix(record.getCurrentLocation().getFileId());
+ fileIds.add(fileID);
+ } else {
+ // Empty partition
+ fileIds.add("");
Review Comment:
can you help me understand when we might hit this ?
--
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: commits-unsubscribe@hudi.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org