Kafka source offset-based deduplication. (#33596)

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaCheckpointMark.java

@@ -17,6 +17,8 @@
  */
 package org.apache.beam.sdk.io.kafka;
 
+import static org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.base.Preconditions.checkState;
+
 import java.io.Serializable;
 import java.util.List;
 import java.util.Optional;
@@ -42,6 +44,8 @@ public class KafkaCheckpointMark implements UnboundedSource.CheckpointMark {
   @SuppressWarnings("initialization") // Avro will set the fields by breaking abstraction
   private KafkaCheckpointMark() {} // for Avro
 
+  private static final long OFFSET_DEDUP_PARTITIONS_PER_SPLIT = 1;
+
   public KafkaCheckpointMark(
       List<PartitionMark> partitions, Optional<KafkaUnboundedReader<?, ?>> reader) {
     this.partitions = partitions;
@@ -66,6 +70,23 @@ public String toString() {
     return "KafkaCheckpointMark{partitions=" + Joiner.on(",").join(partitions) + '}';
   }
 
+  @Override
+  public byte[] getOffsetLimit() {
+    if (!reader.isPresent()) {
+      throw new RuntimeException(
+          "KafkaCheckpointMark reader is not present while calling getOffsetLimit().");
+    }
+    if (!reader.get().offsetBasedDeduplicationSupported()) {
+      throw new RuntimeException(
+          "Unexpected getOffsetLimit() called while KafkaUnboundedReader not configured for offset deduplication.");
+    }
+
+    // KafkaUnboundedSource.split() must produce a 1:1 partition to split ratio.
+    checkState(partitions.size() == OFFSET_DEDUP_PARTITIONS_PER_SPLIT);
+    PartitionMark partition = partitions.get(/* index= */ 0);
+    return KafkaIOUtils.OffsetBasedDeduplication.encodeOffset(partition.getNextOffset());
+  }
+
   /**
    * A tuple to hold topic, partition, and offset that comprise the checkpoint for a single
    * partition.

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaIO.java

@@ -717,6 +717,9 @@ public abstract static class Read<K, V>
     @Pure
     public abstract int getRedistributeNumKeys();
 
+    @Pure
+    public abstract @Nullable Boolean getOffsetDeduplication();
+
     @Pure
     public abstract @Nullable Duration getWatchTopicPartitionDuration();
 
@@ -782,6 +785,8 @@ abstract Builder<K, V> setConsumerFactoryFn(
 
       abstract Builder<K, V> setRedistributeNumKeys(int redistributeNumKeys);
 
+      abstract Builder<K, V> setOffsetDeduplication(Boolean offsetDeduplication);
+
       abstract Builder<K, V> setTimestampPolicyFactory(
           TimestampPolicyFactory<K, V> timestampPolicyFactory);
 
@@ -886,11 +891,16 @@ static <K, V> void setupExternalBuilder(
           if (config.allowDuplicates != null) {
             builder.setAllowDuplicates(config.allowDuplicates);
           }
-
+          if (config.redistribute
+              && (config.allowDuplicates == null || !config.allowDuplicates)
+              && config.offsetDeduplication != null) {
+            builder.setOffsetDeduplication(config.offsetDeduplication);
+          }
         } else {
           builder.setRedistributed(false);
           builder.setRedistributeNumKeys(0);
           builder.setAllowDuplicates(false);
+          builder.setOffsetDeduplication(false);
         }
       }
 
@@ -959,6 +969,7 @@ public static class Configuration {
         private Integer redistributeNumKeys;
         private Boolean redistribute;
         private Boolean allowDuplicates;
+        private Boolean offsetDeduplication;
 
         public void setConsumerConfig(Map<String, String> consumerConfig) {
           this.consumerConfig = consumerConfig;
@@ -1015,6 +1026,10 @@ public void setRedistribute(Boolean redistribute) {
         public void setAllowDuplicates(Boolean allowDuplicates) {
           this.allowDuplicates = allowDuplicates;
         }
+
+        public void setOffsetDeduplication(Boolean offsetDeduplication) {
+          this.offsetDeduplication = offsetDeduplication;
+        }
       }
     }
 
@@ -1066,26 +1081,21 @@ public Read<K, V> withTopicPartitions(List<TopicPartition> topicPartitions) {
      * Sets redistribute transform that hints to the runner to try to redistribute the work evenly.
      */
     public Read<K, V> withRedistribute() {
-      if (getRedistributeNumKeys() == 0 && isRedistributed()) {
-        LOG.warn("This will create a key per record, which is sub-optimal for most use cases.");
-      }
       return toBuilder().setRedistributed(true).build();
     }
 
     public Read<K, V> withAllowDuplicates(Boolean allowDuplicates) {
-      if (!isAllowDuplicates()) {
-        LOG.warn("Setting this value without setting withRedistribute() will have no effect.");
-      }
       return toBuilder().setAllowDuplicates(allowDuplicates).build();
     }
 
     public Read<K, V> withRedistributeNumKeys(int redistributeNumKeys) {
-      checkState(
-          isRedistributed(),
-          "withRedistributeNumKeys is ignored if withRedistribute() is not enabled on the transform.");
       return toBuilder().setRedistributeNumKeys(redistributeNumKeys).build();
     }
 
+    public Read<K, V> withOffsetDeduplication(Boolean offsetDeduplication) {
+      return toBuilder().setOffsetDeduplication(offsetDeduplication).build();
+    }
+
     /**
      * Internally sets a {@link java.util.regex.Pattern} of topics to read from. All the partitions
      * from each of the matching topics are read.
@@ -1541,6 +1551,9 @@ public PCollection<KafkaRecord<K, V>> expand(PBegin input) {
               ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG);
         }
       }
+
+      checkRedistributeConfiguration();
+
       warnAboutUnsafeConfigurations(input);
 
       // Infer key/value coders if not specified explicitly
@@ -1573,6 +1586,26 @@ && runnerPrefersLegacyRead(input.getPipeline().getOptions()))) {
       return input.apply(new ReadFromKafkaViaSDF<>(this, keyCoder, valueCoder));
     }
 
+    private void checkRedistributeConfiguration() {
+      if (getRedistributeNumKeys() == 0 && isRedistributed()) {
+        LOG.warn(
+            "withRedistribute without withRedistributeNumKeys will create a key per record, which is sub-optimal for most use cases.");
+      }
+      if (isAllowDuplicates()) {
+        LOG.warn("Setting this value without setting withRedistribute() will have no effect.");
+      }
+      if (getRedistributeNumKeys() > 0) {
+        checkState(
+            isRedistributed(),
+            "withRedistributeNumKeys is ignored if withRedistribute() is not enabled on the transform.");
+      }
+      if (getOffsetDeduplication() != null && getOffsetDeduplication()) {
+        checkState(
+            isRedistributed() && !isAllowDuplicates(),
+            "withOffsetDeduplication should only be used with withRedistribute and withAllowDuplicates(false).");
+      }
+    }
+
     private void warnAboutUnsafeConfigurations(PBegin input) {
       Long checkpointingInterval =
           input

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaIOReadImplementationCompatibility.java

@@ -137,6 +137,7 @@ Object getDefaultValue() {
         return false;
       }
     },
+    OFFSET_DEDUPLICATION(LEGACY),
     ;
 
     private final @NonNull ImmutableSet<KafkaIOReadImplementation> supportedImplementations;

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaIOUtils.java

@@ -19,11 +19,13 @@
 
 import static org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.base.Preconditions.checkArgument;
 
+import java.nio.charset.StandardCharsets;
 import java.util.HashMap;
 import java.util.Map;
 import java.util.Random;
 import org.apache.beam.sdk.transforms.SerializableFunction;
 import org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.collect.ImmutableMap;
+import org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.primitives.Longs;
 import org.apache.kafka.clients.consumer.Consumer;
 import org.apache.kafka.clients.consumer.ConsumerConfig;
 import org.apache.kafka.clients.consumer.KafkaConsumer;
@@ -142,4 +144,15 @@ void update(double quantity) {
       return avg;
     }
   }
+
+  static final class OffsetBasedDeduplication {
+
+    static byte[] encodeOffset(long offset) {
+      return Longs.toByteArray(offset);
+    }
+
+    static byte[] getUniqueId(String topic, int partition, long offset) {
+      return (topic + "-" + partition + "-" + offset).getBytes(StandardCharsets.UTF_8);
+    }
+  }
 }

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaUnboundedReader.java

@@ -66,6 +66,7 @@
 import org.apache.kafka.common.TopicPartition;
 import org.apache.kafka.common.errors.WakeupException;
 import org.apache.kafka.common.serialization.Deserializer;
+import org.checkerframework.checker.nullness.qual.MonotonicNonNull;
 import org.checkerframework.checker.nullness.qual.Nullable;
 import org.joda.time.Duration;
 import org.joda.time.Instant;
@@ -299,6 +300,30 @@ public Instant getCurrentTimestamp() throws NoSuchElementException {
     return curTimestamp;
   }
 
+  @Override
+  public byte[] getCurrentRecordId() throws NoSuchElementException {
+    if (!offsetBasedDeduplicationSupported()) {
+      // Defer result to super if offset deduplication is not supported.
+      return super.getCurrentRecordId();
+    }
+    if (curRecord == null) {
+      throw new NoSuchElementException("KafkaUnboundedReader's curRecord is null.");
+    }
+    return KafkaIOUtils.OffsetBasedDeduplication.getUniqueId(
+        curRecord.getTopic(), curRecord.getPartition(), curRecord.getOffset());
+  }
+
+  @Override
+  public byte[] getCurrentRecordOffset() throws NoSuchElementException {
+    if (!offsetBasedDeduplicationSupported()) {
+      throw new RuntimeException("UnboundedSource must enable offset-based deduplication.");
+    }
+    if (curRecord == null) {
+      throw new NoSuchElementException("KafkaUnboundedReader's curRecord is null.");
+    }
+    return KafkaIOUtils.OffsetBasedDeduplication.encodeOffset(curRecord.getOffset());
+  }
+
   @Override
   public long getSplitBacklogBytes() {
     long backlogBytes = 0;
@@ -313,6 +338,10 @@ public long getSplitBacklogBytes() {
     return backlogBytes;
   }
 
+  public boolean offsetBasedDeduplicationSupported() {
+    return source.offsetBasedDeduplicationSupported();
+  }
+
   ////////////////////////////////////////////////////////////////////////////////////////////////
 
   private static final Logger LOG = LoggerFactory.getLogger(KafkaUnboundedReader.class);
@@ -331,8 +360,8 @@ public long getSplitBacklogBytes() {
   private final String name;
   private @Nullable Consumer<byte[], byte[]> consumer = null;
   private final List<PartitionState<K, V>> partitionStates;
-  private @Nullable KafkaRecord<K, V> curRecord = null;
-  private @Nullable Instant curTimestamp = null;
+  private @MonotonicNonNull KafkaRecord<K, V> curRecord = null;
+  private @MonotonicNonNull Instant curTimestamp = null;
   private Iterator<PartitionState<K, V>> curBatch = Collections.emptyIterator();
 
   private @Nullable Deserializer<K> keyDeserializerInstance = null;

sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaUnboundedSource.java

@@ -113,10 +113,20 @@ public List<KafkaUnboundedSource<K, V>> split(int desiredNumSplits, PipelineOpti
         partitions.size() > 0,
         "Could not find any partitions. Please check Kafka configuration and topic names");
 
-    int numSplits = Math.min(desiredNumSplits, partitions.size());
-    // XXX make all splits have the same # of partitions
-    while (partitions.size() % numSplits > 0) {
-      ++numSplits;
+    int numSplits;
+    if (offsetBasedDeduplicationSupported()) {
+      // Enforce 1:1 split to partition ratio for offset deduplication.
+      numSplits = partitions.size();
+      LOG.info(
+          "Offset-based deduplication is enabled for KafkaUnboundedSource. "
+              + "Forcing the number of splits to equal the number of total partitions: {}.",
+          numSplits);
+    } else {
+      numSplits = Math.min(desiredNumSplits, partitions.size());
+      // Make all splits have the same # of partitions.
+      while (partitions.size() % numSplits > 0) {
+        ++numSplits;
+      }
     }
     List<List<TopicPartition>> assignments = new ArrayList<>(numSplits);
 
@@ -177,6 +187,11 @@ public boolean requiresDeduping() {
     return false;
   }
 
+  @Override
+  public boolean offsetBasedDeduplicationSupported() {
+    return spec.getOffsetDeduplication() != null && spec.getOffsetDeduplication();
+  }
+
   @Override
   public Coder<KafkaRecord<K, V>> getOutputCoder() {
     Coder<K> keyCoder = Preconditions.checkStateNotNull(spec.getKeyCoder());

sdks/java/io/kafka/src/test/java/org/apache/beam/sdk/io/kafka/KafkaIOExternalTest.java

@@ -145,7 +145,6 @@ public void testConstructKafkaRead() throws Exception {
     expansionService.expand(request, observer);
     ExpansionApi.ExpansionResponse result = observer.result;
     RunnerApi.PTransform transform = result.getTransform();
-    System.out.println("xxx : " + result.toString());
     assertThat(
         transform.getSubtransformsList(),
         Matchers.hasItem(MatchesPattern.matchesPattern(".*KafkaIO-Read.*")));

sdks/java/io/kafka/src/test/java/org/apache/beam/sdk/io/kafka/KafkaIOReadImplementationCompatibilityTest.java

@@ -18,6 +18,7 @@
 package org.apache.beam.sdk.io.kafka;
 
 import static org.apache.beam.sdk.io.kafka.KafkaIOTest.mkKafkaReadTransform;
+import static org.apache.beam.sdk.io.kafka.KafkaIOTest.mkKafkaReadTransformWithOffsetDedup;
 import static org.hamcrest.MatcherAssert.assertThat;
 import static org.hamcrest.Matchers.containsInAnyOrder;
 import static org.hamcrest.Matchers.empty;
@@ -114,7 +115,8 @@ private PipelineResult testReadTransformCreationWithImplementationBoundPropertie
                 new ValueAsTimestampFn(),
                 false, /*redistribute*/
                 false, /*allowDuplicates*/
-                0)));
+                0, /*numKeys*/
+                null /*offsetDeduplication*/)));
     return p.run();
   }
 
@@ -139,6 +141,17 @@ public void testReadTransformCreationWithLegacyImplementationBoundProperty() {
     assertThat(Lineage.query(r.metrics(), Lineage.Type.SOURCE), containsInAnyOrder(expect));
   }
 
+  @Test
+  public void testReadTransformCreationWithOffsetDeduplication() {
+    p.apply(mkKafkaReadTransformWithOffsetDedup(1000, new ValueAsTimestampFn()));
+    PipelineResult r = p.run();
+    String[] expect =
+        KafkaIOTest.mkKafkaTopics.stream()
+            .map(topic -> String.format("kafka:`%s`.%s", KafkaIOTest.mkKafkaServers, topic))
+            .toArray(String[]::new);
+    assertThat(Lineage.query(r.metrics(), Lineage.Type.SOURCE), containsInAnyOrder(expect));
+  }
+
   @Test
   public void testReadTransformCreationWithSdfImplementationBoundProperty() {
     PipelineResult r =