bschell commented on a change in pull request #1091: [HUDI-389] Fixing Index
look up to return right partitions for a given key along with fileId with
Global Bloom
URL: https://github.com/apache/incubator-hudi/pull/1091#discussion_r357363234
##########
File path:
hudi-client/src/main/java/org/apache/hudi/index/bloom/HoodieGlobalBloomIndex.java
##########
@@ -98,26 +100,34 @@ public HoodieGlobalBloomIndex(HoodieWriteConfig config) {
String recordKey = partitionRecordKeyPair._2();
String partitionPath = partitionRecordKeyPair._1();
- return indexFileFilter.getMatchingFiles(partitionPath,
recordKey).stream()
- .map(file -> new Tuple2<>(file, new HoodieKey(recordKey,
indexToPartitionMap.get(file))))
+ return indexFileFilter.getMatchingFilesAndPartition(partitionPath,
recordKey).stream()
+ .map(partitionFileIdPair -> new
Tuple2<>(partitionFileIdPair.getRight(),
+ new HoodieKey(recordKey, partitionFileIdPair.getLeft())))
.collect(Collectors.toList());
}).flatMap(List::iterator);
}
-
/**
* Tagging for global index should only consider the record key.
*/
@Override
protected JavaRDD<HoodieRecord<T>> tagLocationBacktoRecords(
JavaPairRDD<HoodieKey, HoodieRecordLocation> keyFilenamePairRDD,
JavaRDD<HoodieRecord<T>> recordRDD) {
- JavaPairRDD<String, HoodieRecord<T>> rowKeyRecordPairRDD =
- recordRDD.mapToPair(record -> new Tuple2<>(record.getRecordKey(),
record));
- // Here as the recordRDD might have more data than rowKeyRDD (some
rowKeys' fileId is null),
- // so we do left outer join.
- return rowKeyRecordPairRDD.leftOuterJoin(keyFilenamePairRDD.mapToPair(p ->
new Tuple2<>(p._1.getRecordKey(), p._2)))
- .values().map(value -> getTaggedRecord(value._1,
Option.ofNullable(value._2.orNull())));
+ Map<String, Pair<HoodieKey, HoodieRecordLocation>> keyMap =
keyFilenamePairRDD
+ .mapToPair(e -> new Tuple2<>(e._1.getRecordKey(), Pair.of(e._1,
e._2))).collectAsMap();
+
+ JavaRDD<HoodieRecord<T>> toReturn = recordRDD.map(rec -> {
+ if (keyMap.containsKey(rec.getRecordKey())) {
+ Pair<HoodieKey, HoodieRecordLocation> value =
keyMap.get(rec.getRecordKey());
+ return getTaggedRecord(new HoodieRecord(value.getLeft(),
rec.getData()), Option.of(value
Review comment:
That is what is happening here right? This logic only occurs when the
record_key is already matching so we are essentially replacing just the
partitionPath when we replace the HoodieKey.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
