gguptp commented on code in PR #193:
URL: 
https://github.com/apache/flink-connector-aws/pull/193#discussion_r2014045507


##########
flink-connector-aws/flink-connector-dynamodb/src/main/java/org/apache/flink/connector/dynamodb/source/reader/DynamoDbStreamsSourceReader.java:
##########
@@ -80,8 +115,54 @@ protected DynamoDbStreamsShardSplit toSplitType(
 
     @Override
     public void addSplits(List<DynamoDbStreamsShardSplit> splits) {
-        splits.forEach(this::registerShardMetricGroup);
-        super.addSplits(splits);
+        List<DynamoDbStreamsShardSplit> dynamoDbStreamsShardSplits = new 
ArrayList<>();
+        for (DynamoDbStreamsShardSplit split : splits) {
+            if (split.isFinished()) {
+                // Restore finished splits state
+                splitFinishedEvents
+                        .computeIfAbsent(split.getFinishedAfterCheckpoint(), k 
-> new HashSet<>())
+                        .add(split);
+                // Replay the finished split event
+                context.sendSourceEventToCoordinator(
+                        new 
SplitsFinishedEvent(Collections.singleton(split.splitId())));
+            } else {
+                dynamoDbStreamsShardSplits.add(split);
+            }
+        }
+        dynamoDbStreamsShardSplits.forEach(this::registerShardMetricGroup);
+        super.addSplits(dynamoDbStreamsShardSplits);
+    }
+
+    /**
+     * At snapshot, we also store the pending finished split ids in the 
current checkpoint so that
+     * in case we have to restore the reader from state, we also send the 
finished split ids
+     * otherwise we run a risk of data loss during restarts of the source 
because of the
+     * SplitsFinishedEvent going missing.
+     *
+     * @param checkpointId
+     * @return
+     */
+    @Override
+    public List<DynamoDbStreamsShardSplit> snapshotState(long checkpointId) {
+        this.currentCheckpointId = checkpointId;
+        List<DynamoDbStreamsShardSplit> splits = 
super.snapshotState(checkpointId);
+
+        if (!splitFinishedEvents.isEmpty()) {
+            // Add all finished splits to the snapshot
+            splitFinishedEvents.values().forEach(splits::addAll);
+        }
+        return splits;
+    }
+
+    /**
+     * During notifyCheckpointComplete, we should clean up the state of 
finished splits that are
+     * older than the checkpoint id.
+     *
+     * @param checkpointId
+     */
+    @Override
+    public void notifyCheckpointComplete(long checkpointId) {
+        splitFinishedEvents.headMap(checkpointId, true).clear();

Review Comment:
   good point, any splits which are <= this checkpoint will be cleared. I guess 
should we remove this true parameter and let the finished splits lay there in 
the checkpoint for one more checkpoint?



-- 
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]

Reply via email to