KAFKA-15607: Fix NPE in MirrorCheckpointTask::syncGroupOffset (#14587)

Reviewers: Chris Egerton <chrise@aiven.io>

connect/mirror/src/main/java/org/apache/kafka/connect/mirror/MirrorCheckpointTask.java

@@ -330,12 +330,21 @@ public class MirrorCheckpointTask extends SourceTask {
 
                 // if translated offset from upstream is smaller than the current consumer offset
                 // in the target, skip updating the offset for that partition
-                long latestDownstreamOffset = targetConsumerOffset.get(topicPartition).offset();
+                OffsetAndMetadata targetOffsetAndMetadata = targetConsumerOffset.get(topicPartition);
+                if (targetOffsetAndMetadata != null) {
+                    long latestDownstreamOffset = targetOffsetAndMetadata.offset();
                     if (latestDownstreamOffset >= convertedOffset.offset()) {
                         log.trace("latestDownstreamOffset {} is larger than or equal to convertedUpstreamOffset {} for "
                                 + "TopicPartition {}", latestDownstreamOffset, convertedOffset.offset(), topicPartition);
                         continue;
                     }
+                } else {
+                    // It is possible that when resetting offsets are performed in the java kafka client, the reset to -1 will be intercepted.
+                    // However, there are some other types of clients such as sarama, which can magically reset the group offset to -1, which will cause
+                    // `targetOffsetAndMetadata` here is null. For this case, just sync the offset to target.
+                    log.warn("Group {} offset for partition {} may has been reset to a negative offset, just sync the offset to target.",
+                            consumerGroupId, topicPartition);
+                }
                 offsetToSync.put(topicPartition, convertedOffset);
             }
 

connect/mirror/src/test/java/org/apache/kafka/connect/mirror/MirrorCheckpointTaskTest.java

@@ -169,6 +169,33 @@ public class MirrorCheckpointTaskTest {
                 "Consumer 2 " + topic2 + " failed");
     }
 
+    @Test
+    public void testSyncOffsetForTargetGroupWithNullOffsetAndMetadata() {
+        Map<String, Map<TopicPartition, OffsetAndMetadata>> idleConsumerGroupsOffset = new HashMap<>();
+        Map<String, Map<TopicPartition, Checkpoint>> checkpointsPerConsumerGroup = new HashMap<>();
+
+        String consumer = "consumer";
+        String topic = "topic";
+        Map<TopicPartition, OffsetAndMetadata> ct = new HashMap<>();
+        TopicPartition tp = new TopicPartition(topic, 0);
+        // Simulate other clients such as Sarama, which may reset group offsets to -1. This can cause
+        // the obtained `OffsetAndMetadata` of the target cluster to be null.
+        ct.put(tp, null);
+        idleConsumerGroupsOffset.put(consumer, ct);
+
+        Checkpoint cp = new Checkpoint(consumer, new TopicPartition(topic, 0), 200, 101, "metadata");
+        Map<TopicPartition, Checkpoint> checkpointMap = new HashMap<>();
+        checkpointMap.put(cp.topicPartition(), cp);
+        checkpointsPerConsumerGroup.put(consumer, checkpointMap);
+
+        MirrorCheckpointTask mirrorCheckpointTask = new MirrorCheckpointTask("source", "target",
+                new DefaultReplicationPolicy(), null, idleConsumerGroupsOffset, checkpointsPerConsumerGroup);
+
+        Map<String, Map<TopicPartition, OffsetAndMetadata>> output = mirrorCheckpointTask.syncGroupOffset();
+
+        assertEquals(101, output.get(consumer).get(tp).offset(), "Consumer " + topic + " failed");
+    }
+
     @Test
     public void testNoCheckpointForTopicWithoutOffsetSyncs() {
         OffsetSyncStoreTest.FakeOffsetSyncStore offsetSyncStore = new OffsetSyncStoreTest.FakeOffsetSyncStore();