danny0405 commented on a change in pull request #4880:
URL: https://github.com/apache/hudi/pull/4880#discussion_r830432504
##########
File path:
hudi-common/src/main/java/org/apache/hudi/common/table/log/HoodieMergedLogRecordScanner.java
##########
@@ -153,9 +154,29 @@ protected void processNextRecord(HoodieRecord<? extends
HoodieRecordPayload> hoo
}
@Override
- protected void processNextDeletedKey(HoodieKey hoodieKey) {
- records.put(hoodieKey.getRecordKey(),
SpillableMapUtils.generateEmptyPayload(hoodieKey.getRecordKey(),
- hoodieKey.getPartitionPath(), getPayloadClassFQN()));
+ protected void processNextDeletedRecord(DeleteRecord deleteRecord) {
+ String key = deleteRecord.getRecordKey();
+ if (records.containsKey(key)) {
+ // Merge and store the merged record. The ordering val is taken to
decide whether the same key record
+ // should be deleted or be kept. The old record is kept only if the
DELETE record has smaller ordering val.
+ // For same ordering values, uses the natural order(arrival time
semantics).
+
+ HoodieRecord<? extends HoodieRecordPayload> oldRecord = records.get(key);
+ Comparable curOrderingVal = oldRecord.getData().getCombineValue();
+ Comparable deleteOrderingVal = deleteRecord.getOrderingVal();
+ // Checks the ordering value does not equal to 0
+ // because we use 0 as the default value which means natural order
+ boolean choosePrev = !deleteOrderingVal.equals(0)
+ && curOrderingVal.getClass() == deleteOrderingVal.getClass()
+ && curOrderingVal.compareTo(deleteOrderingVal) > 0;
Review comment:
@vinothchandar Maybe we should really deprecate the delete block
forever, imagine there are operations in one traction for a database:
INSERT DELETE INSERT (for a key, they all only have process time semantics,
the linearizabiity)
The record is kept for a RDBMS but would lost if we do these operations in
one commit in Hudi even if after this patch.
--
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]
