[GSoC 2026] Kafka Streams runner — Redistribute translator + ExecutableStage type-agnostic edge

runners/kafka-streams/src/main/java/org/apache/beam/runners/kafka/streams/translation/ExecutableStageProcessor.java

@@ -59,7 +59,7 @@
  * receivers.
  */
 class ExecutableStageProcessor
-    implements Processor<byte[], KStreamsPayload<byte[]>, byte[], KStreamsPayload<byte[]>> {
+    implements Processor<byte[], KStreamsPayload<?>, byte[], KStreamsPayload<?>> {
 
   private static final Logger LOG = LoggerFactory.getLogger(ExecutableStageProcessor.class);
 
@@ -68,9 +68,13 @@ class ExecutableStageProcessor
 
   // pendingOutputs is enqueued by SDK harness threads (inside the OutputReceiverFactory callback)
   // and drained by the Kafka Streams processing thread on bundle close; needs to be thread-safe.
-  private final Queue<WindowedValue<byte[]>> pendingOutputs = new ConcurrentLinkedQueue<>();
+  // The element type is intentionally wildcarded: the runner does not need to know the runtime
+  // value type — the bundle factory handles all coder application at the Fn-API boundary using
+  // the PCollection coders from the ExecutableStagePayload. Pretending the type was byte[] was
+  // only safe because the Impulse output coder happens to be ByteArrayCoder.
+  private final Queue<WindowedValue<?>> pendingOutputs = new ConcurrentLinkedQueue<>();
 
-  private @Nullable ProcessorContext<byte[], KStreamsPayload<byte[]>> context;
+  private @Nullable ProcessorContext<byte[], KStreamsPayload<?>> context;
   private @Nullable ExecutableStageContext stageContext;
   private @Nullable StageBundleFactory stageBundleFactory;
   private @Nullable RemoteBundle currentBundle;
@@ -81,16 +85,16 @@ class ExecutableStageProcessor
   }
 
   @Override
-  public void init(ProcessorContext<byte[], KStreamsPayload<byte[]>> context) {
+  public void init(ProcessorContext<byte[], KStreamsPayload<?>> context) {
     this.context = context;
     ExecutableStage executableStage = ExecutableStage.fromPayload(stagePayload);
     this.stageContext = KafkaStreamsExecutableStageContextFactory.getInstance().get(jobInfo);
     this.stageBundleFactory = stageContext.getStageBundleFactory(executableStage);
   }
 
   @Override
-  public void process(Record<byte[], KStreamsPayload<byte[]>> record) {
-    KStreamsPayload<byte[]> payload = record.value();
+  public void process(Record<byte[], KStreamsPayload<?>> record) {
+    KStreamsPayload<?> payload = record.value();
     if (payload.isWatermark()) {
       // NOTE: flushing the bundle on every received watermark is provisional. Once the
       // WatermarkManager lands, a stage will receive watermarks from multiple parent instances and
@@ -122,7 +126,7 @@ public <OutputT> FnDataReceiver<OutputT> create(String pCollectionId) {
             // after the bundle closes.
             return receivedElement -> {
               if (receivedElement != null) {
-                pendingOutputs.add((WindowedValue<byte[]>) receivedElement);
+                pendingOutputs.add((WindowedValue<?>) receivedElement);
               }
             };
           }
@@ -143,7 +147,7 @@ private FnDataReceiver<WindowedValue<?>> mainInputReceiver() {
     return receiver;
   }
 
-  private void closeBundleAndFlush(Record<byte[], KStreamsPayload<byte[]>> record) {
+  private void closeBundleAndFlush(Record<byte[], KStreamsPayload<?>> record) {
     RemoteBundle bundle = currentBundle;
     if (bundle == null) {
       return;
@@ -157,26 +161,25 @@ private void closeBundleAndFlush(Record<byte[], KStreamsPayload<byte[]>> record)
     } finally {
       currentBundle = null;
     }
-    ProcessorContext<byte[], KStreamsPayload<byte[]>> ctx = checkInitialized(context);
+    ProcessorContext<byte[], KStreamsPayload<?>> ctx = checkInitialized(context);
     // The harness has finished the bundle (close() returned) so no further enqueues happen.
-    // ConcurrentLinkedQueue's weakly-consistent iterator is therefore safe to drain via forEach.
-    pendingOutputs.forEach(
-        output ->
-            ctx.forward(
-                new Record<byte[], KStreamsPayload<byte[]>>(
-                    record.key(), KStreamsPayload.data(output), record.timestamp())));
-    pendingOutputs.clear();
+    // Drain via poll() so each element is removed as it is forwarded.
+    WindowedValue<?> output;
+    while ((output = pendingOutputs.poll()) != null) {
+      ctx.forward(
+          new Record<byte[], KStreamsPayload<?>>(
+              record.key(), KStreamsPayload.data(output), record.timestamp()));
+    }
   }
 
-  private void forwardWatermark(
-      Record<byte[], KStreamsPayload<byte[]>> record, long watermarkMillis) {
+  private void forwardWatermark(Record<byte[], KStreamsPayload<?>> record, long watermarkMillis) {
     // TODO(#38743 / WatermarkManager): a watermark must reach every parallel instance of every
     // downstream processor, but ctx.forward routes to one downstream partition per Kafka Streams'
     // partitioning. The simplest correct approach is to fan the watermark out to all downstream
     // partitions; that wiring lands with the WatermarkManager sub-issue (per Jan on PR #38764).
-    ProcessorContext<byte[], KStreamsPayload<byte[]>> ctx = checkInitialized(context);
+    ProcessorContext<byte[], KStreamsPayload<?>> ctx = checkInitialized(context);
     ctx.forward(
-        new Record<byte[], KStreamsPayload<byte[]>>(
+        new Record<byte[], KStreamsPayload<?>>(
             record.key(), KStreamsPayload.watermark(watermarkMillis), record.timestamp()));
   }
 

runners/kafka-streams/src/main/java/org/apache/beam/runners/kafka/streams/translation/KafkaStreamsPipelineTranslator.java

@@ -17,16 +17,21 @@
  */
 package org.apache.beam.runners.kafka.streams.translation;
 
+import com.google.auto.service.AutoService;
 import java.util.Map;
+import java.util.Set;
 import org.apache.beam.model.pipeline.v1.RunnerApi;
 import org.apache.beam.runners.fnexecution.provisioning.JobInfo;
 import org.apache.beam.runners.kafka.streams.KafkaStreamsPipelineOptions;
+import org.apache.beam.sdk.util.construction.NativeTransforms;
 import org.apache.beam.sdk.util.construction.PTransformTranslation;
 import org.apache.beam.sdk.util.construction.graph.ExecutableStage;
 import org.apache.beam.sdk.util.construction.graph.GreedyPipelineFuser;
 import org.apache.beam.sdk.util.construction.graph.PipelineNode;
 import org.apache.beam.sdk.util.construction.graph.QueryablePipeline;
+import org.apache.beam.sdk.util.construction.graph.TrivialNativeTransformExpander;
 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.collect.ImmutableSet;
 
 /**
  * Translates a portable Beam pipeline into a Kafka Streams {@link
@@ -45,6 +50,7 @@ public KafkaStreamsPipelineTranslator() {
     this(
         ImmutableMap.<String, PTransformTranslator>builder()
             .put(PTransformTranslation.IMPULSE_TRANSFORM_URN, new ImpulseTranslator())
+            .put(PTransformTranslation.REDISTRIBUTE_ARBITRARILY_URN, new RedistributeTranslator())
             .put(ExecutableStage.URN, new ExecutableStageTranslator())
             .build());
   }
@@ -59,11 +65,28 @@ public KafkaStreamsTranslationContext createTranslationContext(
   }
 
   /**
-   * Fuses the pipeline so that stateless user code is grouped into {@code ExecutableStage} nodes.
+   * Returns the set of URNs this translator handles natively. {@link
+   * TrivialNativeTransformExpander} uses this set to strip the sub-transforms of runner-native
+   * composites (e.g. {@code Redistribute.arbitrarily}) before fusion, so they survive into
+   * translation as leaves instead of being expanded into primitives the runner does not implement
+   * yet (e.g. GroupByKey).
+   */
+  public Set<String> knownUrns() {
+    return urnToTranslator.keySet();
+  }
+
+  /**
+   * Prepares the pipeline for translation:
+   *
+   * <ol>
+   *   <li>Trim sub-transforms of runner-native composites listed in {@link #knownUrns()} so the
+   *       fuser leaves them as primitives.
+   *   <li>Fuse remaining stateless user code into {@code ExecutableStage} nodes via {@link
+   *       GreedyPipelineFuser}.
+   * </ol>
    *
-   * <p>Runner-executed primitives that have their own translator (e.g. Impulse) are left intact;
-   * everything else is fused. If the pipeline already contains {@code ExecutableStage} transforms
-   * it is returned unchanged.
+   * <p>If the pipeline already contains {@code ExecutableStage} transforms it is returned
+   * unchanged.
    */
   public RunnerApi.Pipeline prepareForTranslation(RunnerApi.Pipeline pipeline) {
     boolean alreadyFused =
@@ -72,7 +95,8 @@ public RunnerApi.Pipeline prepareForTranslation(RunnerApi.Pipeline pipeline) {
     if (alreadyFused) {
       return pipeline;
     }
-    return GreedyPipelineFuser.fuse(pipeline).toPipeline();
+    RunnerApi.Pipeline trimmed = TrivialNativeTransformExpander.forKnownUrns(pipeline, knownUrns());
+    return GreedyPipelineFuser.fuse(trimmed).toPipeline();
   }
 
   /**
@@ -99,4 +123,23 @@ public void translate(KafkaStreamsTranslationContext context, RunnerApi.Pipeline
       translator.translate(node.getId(), pipeline, context);
     }
   }
+
+  /**
+   * Tells the SDK that URNs handled directly by the Kafka Streams runner should be treated as
+   * primitives by {@link QueryablePipeline}. Mirrors Flink's {@code IsFlinkNativeTransform}
+   * pattern. Without this, {@link TrivialNativeTransformExpander} strips a composite's
+   * sub-transforms but {@link QueryablePipeline} still does not recognise the composite itself as a
+   * producer of its outputs, and pipeline validation fails with "consumed but never produced".
+   */
+  @AutoService(NativeTransforms.IsNativeTransform.class)
+  public static class IsKafkaStreamsNativeTransform implements NativeTransforms.IsNativeTransform {
+    private static final Set<String> URNS =
+        ImmutableSet.of(PTransformTranslation.REDISTRIBUTE_ARBITRARILY_URN);
+
+    @Override
+    public boolean test(RunnerApi.PTransform pTransform) {
+      String urn = PTransformTranslation.urnForTransformOrNull(pTransform);
+      return urn != null && URNS.contains(urn);
+    }
+  }
 }

runners/kafka-streams/src/main/java/org/apache/beam/runners/kafka/streams/translation/RedistributeTranslator.java

@@ -0,0 +1,54 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.beam.runners.kafka.streams.translation;
+
+import org.apache.beam.model.pipeline.v1.RunnerApi;
+import org.apache.beam.vendor.guava.v32_1_2_jre.com.google.common.collect.Iterables;
+
+/**
+ * Runner-native translator for {@code beam:transform:redistribute_arbitrarily:v1}.
+ *
+ * <p>The default Beam expansion of {@code Redistribute.arbitrarily()} goes through {@code
+ * GroupByKey} for materialization. The Kafka Streams runner does not need to do any actual
+ * redistribution in the single-instance topology — Kafka Streams is already handling per-task
+ * processing — so we provide a passthrough: the output PCollection is mapped to the same processor
+ * that already produces the input PCollection. No topology node is added.
+ *
+ * <p>The translator is registered in {@link KafkaStreamsPipelineTranslator#knownUrns()} so {@link
+ * org.apache.beam.sdk.util.construction.graph.TrivialNativeTransformExpander} strips the
+ * sub-transforms of {@code Redistribute} before the fuser runs. That keeps the Redistribute
+ * boundary intact and lets the fuser split adjacent stages correctly.
+ *
+ * <p>The keyed variant ({@code redistribute_by_key:v1}) is intentionally not handled here: it
+ * implies a rehash/partition step that the runner can implement on top of GroupByKey when that
+ * lands, but cannot reasonably no-op the way the arbitrary variant can.
+ */
+class RedistributeTranslator implements PTransformTranslator {
+
+  @Override
+  public void translate(
+      String transformId, RunnerApi.Pipeline pipeline, KafkaStreamsTranslationContext context) {
+    RunnerApi.PTransform transform = pipeline.getComponents().getTransformsOrThrow(transformId);
+    String inputPCollectionId = Iterables.getOnlyElement(transform.getInputsMap().values());
+    String parentProcessor = context.getProcessorNameForPCollection(inputPCollectionId);
+    String outputPCollectionId = Iterables.getOnlyElement(transform.getOutputsMap().values());
+    // Passthrough: downstream lookups for the output PCollection resolve to the producer of the
+    // input PCollection. No KS Processor / state store / source is added.
+    context.registerPCollectionProducer(outputPCollectionId, parentProcessor);
+  }
+}