LadyForest commented on code in PR #138:
URL: https://github.com/apache/flink-table-store/pull/138#discussion_r884394501
##########
flink-table-store-connector/src/main/java/org/apache/flink/table/store/connector/TableStoreManagedFactory.java:
##########
@@ -183,6 +204,84 @@ public void onDropTable(Context context, boolean
ignoreIfNotExists) {
@Override
public Map<String, String> onCompactTable(
Context context, CatalogPartitionSpec catalogPartitionSpec) {
- throw new UnsupportedOperationException("Not implement yet");
+ Map<String, String> newOptions = new
HashMap<>(context.getCatalogTable().getOptions());
+ FileStore fileStore = buildTableStore(context).buildFileStore();
+ FileStoreScan.Plan plan =
+ fileStore
+ .newScan()
+ .withPartitionFilter(
+ PredicateConverter.CONVERTER.fromMap(
+
catalogPartitionSpec.getPartitionSpec(),
+ fileStore.partitionType()))
+ .plan();
+
+ Preconditions.checkState(
+ plan.snapshotId() != null && !plan.files().isEmpty(),
+ "The specified %s to compact does not exist any snapshot",
+ catalogPartitionSpec.getPartitionSpec().isEmpty()
+ ? "table"
+ : String.format("partition %s",
catalogPartitionSpec.getPartitionSpec()));
+ Map<BinaryRowData, Map<Integer, List<DataFileMeta>>> groupBy =
plan.groupByPartFiles();
+ if
(!Boolean.parseBoolean(newOptions.get(COMPACTION_RESCALE_BUCKET.key()))) {
+ groupBy =
+ pickManifest(
+ groupBy,
+ new
FileStoreOptions(Configuration.fromMap(newOptions))
+ .mergeTreeOptions(),
+ new
KeyComparatorSupplier(fileStore.partitionType()).get());
+ }
+ try {
+ newOptions.put(
+ COMPACTION_SCANNED_MANIFEST.key(),
+ Base64.getEncoder()
+ .encodeToString(
+ InstantiationUtil.serializeObject(
+ new PartitionedManifestMeta(
+ plan.snapshotId(),
groupBy))));
+ } catch (IOException e) {
+ throw new RuntimeException(e);
+ }
+ return newOptions;
+ }
+
+ @VisibleForTesting
+ Map<BinaryRowData, Map<Integer, List<DataFileMeta>>> pickManifest(
+ Map<BinaryRowData, Map<Integer, List<DataFileMeta>>> groupBy,
+ MergeTreeOptions options,
+ Comparator<RowData> keyComparator) {
+ Map<BinaryRowData, Map<Integer, List<DataFileMeta>>> filtered = new
HashMap<>();
+
+ for (Map.Entry<BinaryRowData, Map<Integer, List<DataFileMeta>>>
partEntry :
+ groupBy.entrySet()) {
+ Map<Integer, List<DataFileMeta>> manifests = new HashMap<>();
+ for (Map.Entry<Integer, List<DataFileMeta>> bucketEntry :
+ partEntry.getValue().entrySet()) {
+ List<DataFileMeta> smallFiles =
Review Comment:
> For example:
inputs: File1(0-10) File2(10-90) File3(90-100)
merged: File4(0-100) File2(10-90)
This can lead to results with overlapping.
I agree with you that "we cannot pick small files at random". But the
example you provide cannot prove this. These three files are all overlapped.
The small file threshold will pick File1(0-10) and File3(90-100), and the
interval partition will pick all of them. So after deduplication, they all get
compacted.
--
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]
