Skip to content

[HUDI-3918] Improve flink bulk_insert perform for partitioned table #5362

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wxplovecc wants to merge 2 commits into from
Closed

[HUDI-3918] Improve flink bulk_insert perform for partitioned table #5362

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fabdbf1
improve bulk_insert perform
wxplovecc Apr 19, 2022
4426590
format code
Apr 20, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions hudi-flink-datasource/hudi-flink/src/main/java/org/apache/hudi/sink/utils/Pipelines.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,9 +102,15 @@ public static DataStreamSink<Object> bulkInsert(Configuration conf, RowType rowT
InternalTypeInfo<RowData> typeInfo = InternalTypeInfo.of(rowTypeWithFileId);

Map<String, String> bucketIdToFileId = new HashMap<>();
dataStream = dataStream.partitionCustom(partitioner, keyGen::getRecordKey)
.map(record -> BucketBulkInsertWriterHelper.rowWithFileId(bucketIdToFileId, keyGen, record, indexKeys, numBuckets), typeInfo)
.setParallelism(conf.getInteger(FlinkOptions.WRITE_TASKS)); // same parallelism as write task to avoid shuffle
if (OptionsResolver.isPartitionedTable(conf)) {
dataStream = dataStream.keyBy(keyGen::getPartitionPath)
.map(record -> BucketBulkInsertWriterHelper.rowWithFileId(bucketIdToFileId, keyGen, record, indexKeys, numBuckets), typeInfo)
.setParallelism(conf.getInteger(FlinkOptions.WRITE_TASKS)); // same parallelism as write task to avoid shuffle
} else {
dataStream = dataStream.partitionCustom(partitioner, keyGen::getRecordKey)
.map(record -> BucketBulkInsertWriterHelper.rowWithFileId(bucketIdToFileId, keyGen, record, indexKeys, numBuckets), typeInfo)
Copy link
Contributor

@danny0405 danny0405 Apr 20, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why shuffle the input with partition path first ?

Copy link
Contributor Author

@wxplovecc wxplovecc Apr 20, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why shuffle the input with partition path first ?

  • Shuffle with custom partitioner was not friendly to partition table with less bucket (eg. partition by day and 1 bucket),It only use 1 write task
  • Make sure the same partition processed in same task avoid duplicated fileId

.setParallelism(conf.getInteger(FlinkOptions.WRITE_TASKS)); // same parallelism as write task to avoid shuffle
}
if (conf.getBoolean(FlinkOptions.WRITE_BULK_INSERT_SORT_INPUT)) {
SortOperatorGen sortOperatorGen = BucketBulkInsertWriterHelper.getFileIdSorterGen(rowTypeWithFileId);
dataStream = dataStream.transform("file_sorter", typeInfo, sortOperatorGen.createSortOperator())
Expand Down