the-other-tim-brown commented on code in PR #17770:
URL: https://github.com/apache/hudi/pull/17770#discussion_r2665083885
##########
hudi-common/src/main/java/org/apache/hudi/common/table/timeline/versioning/v2/ArchivedTimelineLoaderV2.java:
##########
@@ -84,44 +87,43 @@ public void loadInstants(HoodieTableMetaClient metaClient,
// Sort files in reverse chronological order if limit is specified
(newest first for limit queries)
if (hasLimit) {
- filteredFiles.sort(Comparator.comparing((String fileName) -> {
- return LSMTimeline.getMaxInstantTime(fileName);
- }).reversed());
+
filteredFiles.sort(Comparator.comparing(LSMTimeline::getMaxInstantTime).reversed());
}
Schema readSchema = LSMTimeline.getReadSchema(loadMode);
// Use serial stream when limit is involved to guarantee order
- java.util.stream.Stream<String> fileStream = hasLimit
+ Stream<String> fileStream = hasLimit
? filteredFiles.stream()
: filteredFiles.parallelStream();
- fileStream.forEach(fileName -> {
+ return Option.fromJavaOptional(fileStream.map(fileName -> {
if (hasLimit && loadedCount.get() >= limit.get()) {
- return;
+ return null;
}
// Read the archived file
try (HoodieAvroFileReader reader = (HoodieAvroFileReader)
HoodieIOFactory.getIOFactory(metaClient.getStorage())
.getReaderFactory(HoodieRecord.HoodieRecordType.AVRO)
.getFileReader(DEFAULT_HUDI_CONFIG_FOR_READER, new
StoragePath(metaClient.getArchivePath(), fileName))) {
//TODO boundary to revisit in later pr to use HoodieSchema directly
try (ClosableIterator<IndexedRecord> iterator =
reader.getIndexedRecordIterator(HoodieSchema.fromAvroSchema(HoodieLSMTimelineInstant.getClassSchema()),
- HoodieSchema.fromAvroSchema(readSchema))) {
+ HoodieSchema.fromAvroSchema(readSchema))) {
+ int instantTimeFieldPosition =
readSchema.getField(INSTANT_TIME_ARCHIVED_META_FIELD).pos();
while (iterator.hasNext() && (!hasLimit || loadedCount.get() <
limit.get())) {
GenericRecord record = (GenericRecord) iterator.next();
- String instantTime =
record.get(INSTANT_TIME_ARCHIVED_META_FIELD).toString();
- if ((filter == null || filter.isInRange(instantTime))
Review Comment:
The issue is described in the PR description. The update here is to allow
all the instants in this file to be processed so that we don't read the file
many times when building a file system view. In practice, I observed that the
metadata table for example can read the same file 3 times in a single write due
to the filtering logic.
If we can achieve the same performance improvement in a different way,
please suggest something more concrete.
--
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]
