Skip to content
Permalink
Browse files
[HUDI-4200] Fixing sorting of keys fetched from metadata table (#5773)
- Key fetched from metadata table especially from base file reader is not sorted. and hence may result in throwing NPE (key prefix search) or unnecessary seeks to starting of Hfile (full key look ups). Fixing the same in this patch. This is not an issue with log blocks, since sorting is taking care within HoodieHfileDataBlock.
- Commit where the sorting was mistakenly reverted [HUDI-3760] Adding capability to fetch Metadata Records by prefix  #5208
  • Loading branch information
nsivabalan committed Jun 7, 2022
1 parent 4f5cad8 commit f85cd9b16d2bd0c49fb6a05a4be627dffb4c2065
Showing 4 changed files with 27 additions and 16 deletions.
@@ -50,6 +50,7 @@
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
@@ -316,15 +317,20 @@ public void testReaderGetRecordIteratorByKeyPrefixes() throws Exception {
assertEquals(expectedKey50and0s, recordsByPrefix);

// filter for "key1" and "key0" : entries from 'key10 to key19' and 'key00 to key09' should be matched.
List<GenericRecord> expectedKey1sand0s = expectedKey1s;
expectedKey1sand0s.addAll(allRecords.stream()
.filter(entry -> (entry.get("_row_key").toString()).contains("key0"))
.collect(Collectors.toList()));
List<GenericRecord> expectedKey1sand0s = allRecords.stream()
.filter(entry -> (entry.get("_row_key").toString()).contains("key1") || (entry.get("_row_key").toString()).contains("key0"))
.collect(Collectors.toList());
iterator =
hfileReader.getRecordsByKeyPrefixIterator(Arrays.asList("key1", "key0"), avroSchema);
recordsByPrefix =
StreamSupport.stream(Spliterators.spliteratorUnknownSize(iterator, Spliterator.ORDERED), false)
.collect(Collectors.toList());
Collections.sort(recordsByPrefix, new Comparator<GenericRecord>() {
@Override
public int compare(GenericRecord o1, GenericRecord o2) {
return o1.get("_row_key").toString().compareTo(o2.get("_row_key").toString());
}
});
assertEquals(expectedKey1sand0s, recordsByPrefix);
}

@@ -259,11 +259,9 @@ private static Iterator<GenericRecord> getRecordByKeyPrefixIteratorInternal(HFil
return Collections.emptyIterator();
}
} else if (val == -1) {
// If scanner is aleady on the top of hfile. avoid trigger seekTo again.
Option<Cell> headerCell = Option.fromJavaOptional(scanner.getReader().getFirstKey());
if (headerCell.isPresent() && !headerCell.get().equals(scanner.getCell())) {
scanner.seekTo();
}
// Whenever val == -1 HFile reader will place the pointer right before the first record. We have to advance it to the first record
// of the file to validate whether it matches our search criteria
scanner.seekTo();
}

class KeyPrefixIterator implements Iterator<GenericRecord> {
@@ -144,6 +144,10 @@ protected Option<HoodieRecord<HoodieMetadataPayload>> getRecordByKey(String key,
@Override
public HoodieData<HoodieRecord<HoodieMetadataPayload>> getRecordsByKeyPrefixes(List<String> keyPrefixes,
String partitionName) {
// Sort the columns so that keys are looked up in order
List<String> sortedkeyPrefixes = new ArrayList<>(keyPrefixes);
Collections.sort(sortedkeyPrefixes);

// NOTE: Since we partition records to a particular file-group by full key, we will have
// to scan all file-groups for all key-prefixes as each of these might contain some
// records matching the key-prefix
@@ -171,17 +175,17 @@ public HoodieData<HoodieRecord<HoodieMetadataPayload>> getRecordsByKeyPrefixes(L
boolean fullKeys = false;

Map<String, Option<HoodieRecord<HoodieMetadataPayload>>> logRecords =
readLogRecords(logRecordScanner, keyPrefixes, fullKeys, timings);
readLogRecords(logRecordScanner, sortedkeyPrefixes, fullKeys, timings);

List<Pair<String, Option<HoodieRecord<HoodieMetadataPayload>>>> mergedRecords =
readFromBaseAndMergeWithLogRecords(baseFileReader, keyPrefixes, fullKeys, logRecords, timings, partitionName);
readFromBaseAndMergeWithLogRecords(baseFileReader, sortedkeyPrefixes, fullKeys, logRecords, timings, partitionName);

LOG.debug(String.format("Metadata read for %s keys took [baseFileRead, logMerge] %s ms",
keyPrefixes.size(), timings));
sortedkeyPrefixes.size(), timings));

return mergedRecords.iterator();
} catch (IOException ioe) {
throw new HoodieIOException("Error merging records from metadata table for " + keyPrefixes.size() + " key : ", ioe);
throw new HoodieIOException("Error merging records from metadata table for " + sortedkeyPrefixes.size() + " key : ", ioe);
} finally {
closeReader(readers);
}
@@ -194,7 +198,10 @@ public HoodieData<HoodieRecord<HoodieMetadataPayload>> getRecordsByKeyPrefixes(L
@Override
public List<Pair<String, Option<HoodieRecord<HoodieMetadataPayload>>>> getRecordsByKeys(List<String> keys,
String partitionName) {
Map<Pair<String, FileSlice>, List<String>> partitionFileSliceToKeysMap = getPartitionFileSliceToKeysMapping(partitionName, keys);
// Sort the columns so that keys are looked up in order
List<String> sortedKeys = new ArrayList<>(keys);
Collections.sort(sortedKeys);
Map<Pair<String, FileSlice>, List<String>> partitionFileSliceToKeysMap = getPartitionFileSliceToKeysMapping(partitionName, sortedKeys);
List<Pair<String, Option<HoodieRecord<HoodieMetadataPayload>>>> result = new ArrayList<>();
AtomicInteger fileSlicesKeysCount = new AtomicInteger();
partitionFileSliceToKeysMap.forEach((partitionFileSlicePair, fileSliceKeys) -> {
@@ -219,7 +226,7 @@ public List<Pair<String, Option<HoodieRecord<HoodieMetadataPayload>>>> getRecord
fileSliceKeys.size(), timings));
fileSlicesKeysCount.addAndGet(fileSliceKeys.size());
} catch (IOException ioe) {
throw new HoodieIOException("Error merging records from metadata table for " + keys.size() + " key : ", ioe);
throw new HoodieIOException("Error merging records from metadata table for " + sortedKeys.size() + " key : ", ioe);
} finally {
if (!reuse) {
close(Pair.of(partitionFileSlicePair.getLeft(), partitionFileSlicePair.getRight().getFileId()));
@@ -250,7 +250,7 @@ class TestColumnStatsIndex extends HoodieClientTestBase with ColumnStatsIndexSup

{
// We have to include "c1", since we sort the expected outputs by this column
val requestedColumns = Seq("c1", "c4")
val requestedColumns = Seq("c4", "c1")

val partialColStatsDF = readColumnStatsIndex(spark, basePath, metadataConfig, requestedColumns)
val partialTransposedColStatsDF = transposeColumnStatsIndex(spark, partialColStatsDF, requestedColumns, sourceTableSchema)

0 comments on commit f85cd9b

Please sign in to comment.