mjsax commented on a change in pull request #11252:
URL: https://github.com/apache/kafka/pull/11252#discussion_r708758100



##########
File path: 
streams/src/main/java/org/apache/kafka/streams/kstream/internals/KStreamKStreamJoin.java
##########
@@ -209,37 +210,39 @@ private void emitNonJoinedOuterRecords(final 
WindowStore<KeyAndJoinSide<K>, Left
             // reset to MAX_VALUE in case the store is empty
             sharedTimeTracker.minTime = Long.MAX_VALUE;
 
-            try (final KeyValueIterator<Windowed<KeyAndJoinSide<K>>, 
LeftOrRightValue> it = store.all()) {
+            try (final KeyValueIterator<TimestampedKeyAndJoinSide<K>, 
LeftOrRightValue<V1, V2>> it = store.all()) {
                 while (it.hasNext()) {
-                    final KeyValue<Windowed<KeyAndJoinSide<K>>, 
LeftOrRightValue> record = it.next();
+                    final KeyValue<TimestampedKeyAndJoinSide<K>, 
LeftOrRightValue<V1, V2>> record = it.next();
 
-                    final Windowed<KeyAndJoinSide<K>> windowedKey = record.key;
-                    final LeftOrRightValue value = record.value;
-                    sharedTimeTracker.minTime = windowedKey.window().start();
+                    final TimestampedKeyAndJoinSide<K> 
timestampedKeyAndJoinSide = record.key;
+                    final LeftOrRightValue<V1, V2> value = record.value;
+                    final K key = timestampedKeyAndJoinSide.getKey();
+                    final long timestamp = 
timestampedKeyAndJoinSide.getTimestamp();
+                    sharedTimeTracker.minTime = timestamp;
 
                     // Skip next records if window has not closed
-                    if (windowedKey.window().start() + joinAfterMs + 
joinGraceMs >= sharedTimeTracker.streamTime) {
+                    if (timestamp + joinAfterMs + joinGraceMs >= 
sharedTimeTracker.streamTime) {
                         break;
                     }
 
-                    final K key = windowedKey.key().getKey();
-                    final long time = windowedKey.window().start();
-
                     final R nullJoinedValue;
                     if (isLeftSide) {
                         nullJoinedValue = joiner.apply(key,
-                            (V1) value.getLeftValue(),
-                            (V2) value.getRightValue());
+                                value.getLeftValue(),
+                                value.getRightValue());
                     } else {
                         nullJoinedValue = joiner.apply(key,
-                            (V1) value.getRightValue(),
-                            (V2) value.getLeftValue());
+                                (V1) value.getRightValue(),
+                                (V2) value.getLeftValue());
                     }
 
-                    context().forward(key, nullJoinedValue, 
To.all().withTimestamp(time));
+                    context().forward(key, nullJoinedValue, 
To.all().withTimestamp(timestamp));
 
-                    // Delete the key from the outer window store now it is 
emitted
-                    store.put(record.key.key(), null, 
record.key.window().start());
+                    // blind-delete the key from the outer window store now it 
is emitted;
+                    // we may delete some values of the same key which has not 
been iterated yet,
+                    // but since the iterator would still return that key this 
is fine.
+                    // we do not use the delete() call since that would incur 
an extra get
+                    store.put(timestampedKeyAndJoinSide, null);

Review comment:
       If we do the blind delete of the full list, is there not potential data 
loss scenario? Assume you have a list if 2+ left-join candidates in the list. 
We process the first one, and do the delete, and we crash before processing the 
others in the list. After restart, we might never emit the left-join result for 
those record?
   
   Thus, would we need to do the delete only after we have exhausted the list, 
ie, move from key-timestamp to a different key or timestamp ?




-- 
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: jira-unsubscr...@kafka.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to