yihua commented on code in PR #13313:
URL: https://github.com/apache/hudi/pull/13313#discussion_r2101306247
##########
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/index/HoodieIndexUtils.java:
##########
@@ -251,15 +257,52 @@ public static HoodieIndex
createUserDefinedIndex(HoodieWriteConfig config) {
*/
private static <R> HoodieData<HoodieRecord<R>> getExistingRecords(
HoodieData<Pair<String, String>> partitionLocations, HoodieWriteConfig
config, HoodieTable hoodieTable) {
- final Option<String> instantTime = hoodieTable
- .getMetaClient()
+ HoodieTableMetaClient metaClient = hoodieTable.getMetaClient();
+ final Option<String> instantTime = metaClient
.getActiveTimeline() // we need to include all actions and completed
.filterCompletedInstants()
.lastInstant()
.map(HoodieInstant::requestedTime);
- return partitionLocations.flatMap(p
- -> new HoodieMergedReadHandle(config, instantTime, hoodieTable,
Pair.of(p.getKey(), p.getValue()))
- .getMergedRecords().iterator());
+ ReaderContextFactory<R> readerContextFactory =
hoodieTable.getContext().getReaderContextFactory(metaClient);
+ if (instantTime.isEmpty()) {
+ return hoodieTable.getContext().emptyHoodieData();
+ }
+ return partitionLocations.flatMap(p -> {
+ Option<FileSlice> fileSliceOption = Option.fromJavaOptional(hoodieTable
+ .getHoodieView()
+ .getLatestMergedFileSlicesBeforeOrOn(p.getLeft(), instantTime.get())
+ .filter(fileSlice -> fileSlice.getFileId().equals(p.getRight()))
+ .findFirst());
+ if (fileSliceOption.isEmpty()) {
+ return Collections.emptyIterator();
+ }
+ Schema dataSchema = HoodieAvroUtils.addMetadataFields(new
Schema.Parser().parse(config.getWriteSchema()),
config.allowOperationMetadataField());
+ Option<InternalSchema> internalSchemaOption =
SerDeHelper.fromJson(config.getInternalSchema());
+ FileSlice fileSlice = fileSliceOption.get();
+ HoodieReaderContext<R> readerContext = readerContextFactory.getContext();
+ HoodieFileGroupReader<R> fileGroupReader =
HoodieFileGroupReader.<R>newBuilder()
+ .withReaderContext(readerContext)
+ .withHoodieTableMetaClient(metaClient)
+ .withLatestCommitTime(instantTime.get())
+ .withFileSlice(fileSlice)
+ .withDataSchema(dataSchema)
+ .withRequestedSchema(dataSchema)
+ .withInternalSchema(internalSchemaOption)
+ .withShouldUseRecordPosition(false)
+ .withProps(metaClient.getTableConfig().getProps())
+ .build();
+ try {
+ final HoodieRecordLocation currentLocation = new
HoodieRecordLocation(fileSlice.getBaseInstantTime(), fileSlice.getFileId());
+ return new
CloseableMappingIterator<>(fileGroupReader.getClosableHoodieRecordIterator(),
hoodieRecord -> {
+ hoodieRecord.unseal();
+ hoodieRecord.setCurrentLocation(currentLocation);
+ hoodieRecord.seal();
Review Comment:
Is it possible for the file group reader to return an iterator of
`HoodieRecord`s that are already attached with the current location, since the
file group reader knows the file slice, i.e, the base instant time and file ID?
Then the logic here can be simpler.
##########
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/index/HoodieIndexUtils.java:
##########
@@ -251,15 +257,52 @@ public static HoodieIndex
createUserDefinedIndex(HoodieWriteConfig config) {
*/
private static <R> HoodieData<HoodieRecord<R>> getExistingRecords(
HoodieData<Pair<String, String>> partitionLocations, HoodieWriteConfig
config, HoodieTable hoodieTable) {
- final Option<String> instantTime = hoodieTable
- .getMetaClient()
+ HoodieTableMetaClient metaClient = hoodieTable.getMetaClient();
+ final Option<String> instantTime = metaClient
.getActiveTimeline() // we need to include all actions and completed
.filterCompletedInstants()
.lastInstant()
.map(HoodieInstant::requestedTime);
- return partitionLocations.flatMap(p
- -> new HoodieMergedReadHandle(config, instantTime, hoodieTable,
Pair.of(p.getKey(), p.getValue()))
- .getMergedRecords().iterator());
+ ReaderContextFactory<R> readerContextFactory =
hoodieTable.getContext().getReaderContextFactory(metaClient);
+ if (instantTime.isEmpty()) {
+ return hoodieTable.getContext().emptyHoodieData();
+ }
+ return partitionLocations.flatMap(p -> {
+ Option<FileSlice> fileSliceOption = Option.fromJavaOptional(hoodieTable
+ .getHoodieView()
+ .getLatestMergedFileSlicesBeforeOrOn(p.getLeft(), instantTime.get())
+ .filter(fileSlice -> fileSlice.getFileId().equals(p.getRight()))
Review Comment:
`partitionLocations` contains partition path and file ID pairs so
`getLatestMergedFileSlicesBeforeOrOn` can be called on the same partition path
multiple times (which is how it works on master). Wondering if there is a
better way to batch the requests.
--
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]
