[GitHub] [iceberg] pvary commented on a diff in pull request #6382: Implement ShuffleOperator to collect data statistics

[GitBox <gi...@apache.org>]

pvary commented on code in PR #6382:
URL: https://github.com/apache/iceberg/pull/6382#discussion_r1043214544


##########
flink/v1.16/flink/src/main/java/org/apache/iceberg/flink/sink/shuffle/ShuffleOperator.java:
##########
@@ -0,0 +1,140 @@
+/*
+ * 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.iceberg.flink.sink.shuffle;
+
+import java.io.Serializable;
+import java.util.Map;
+import org.apache.flink.annotation.Internal;
+import org.apache.flink.api.common.state.ListState;
+import org.apache.flink.api.common.state.ListStateDescriptor;
+import org.apache.flink.api.common.typeinfo.TypeInformation;
+import org.apache.flink.api.java.functions.KeySelector;
+import org.apache.flink.api.java.typeutils.MapTypeInfo;
+import org.apache.flink.runtime.operators.coordination.OperatorEvent;
+import org.apache.flink.runtime.operators.coordination.OperatorEventGateway;
+import org.apache.flink.runtime.operators.coordination.OperatorEventHandler;
+import org.apache.flink.runtime.state.StateInitializationContext;
+import org.apache.flink.runtime.state.StateSnapshotContext;
+import org.apache.flink.streaming.api.operators.AbstractStreamOperator;
+import org.apache.flink.streaming.api.operators.OneInputStreamOperator;
+import org.apache.flink.streaming.runtime.streamrecord.StreamRecord;
+import org.apache.iceberg.relocated.com.google.common.annotations.VisibleForTesting;
+import org.apache.iceberg.relocated.com.google.common.collect.Maps;
+
+/**
+ * Shuffle operator can help to improve data clustering based on the key.
+ *
+ * <p>It collects the data statistics information, sends to coordinator and gets the global data
+ * distribution weight from coordinator. Then it will ingest the weight into data stream(wrap by a
+ * class{@link ShuffleRecordWrapper}) and send to partitioner.
+ */
+@Internal
+public class ShuffleOperator<T, K extends Serializable>
+    extends AbstractStreamOperator<ShuffleRecordWrapper<T, K>>
+    implements OneInputStreamOperator<T, ShuffleRecordWrapper<T, K>>, OperatorEventHandler {
+
+  private static final long serialVersionUID = 1L;
+
+  private final KeySelector<T, K> keySelector;
+  // the type of the key to collect data statistics
+  private final TypeInformation<K> keyType;
+  private final OperatorEventGateway operatorEventGateway;
+  // key is generated by applying KeySelector to record
+  // value is the times key occurs
+  private transient Map<K, Long> localDataStatisticsMap;
+  private transient Map<K, Long> globalDataDistributionWeightMap;
+  private transient ListState<Map<K, Long>> globalDataDistributionWeightState;
+
+  public ShuffleOperator(
+      KeySelector<T, K> keySelector,
+      TypeInformation<K> keyType,
+      OperatorEventGateway operatorEventGateway) {
+    this.keySelector = keySelector;
+    this.keyType = keyType;
+    this.operatorEventGateway = operatorEventGateway;
+  }
+
+  @VisibleForTesting
+  ListStateDescriptor<Map<K, Long>> generateGlobalDataDistributionWeightDescriptor() {
+    return new ListStateDescriptor<>(
+        "globalDataDistributionWeight", new MapTypeInfo<>(keyType, TypeInformation.of(Long.class)));
+  }
+
+  @Override
+  public void initializeState(StateInitializationContext context) throws Exception {
+    localDataStatisticsMap = Maps.newHashMap();
+    globalDataDistributionWeightState =

Review Comment:
   The global data distribution should be the same (or samish) for every instance of the operator.
   Shall we use `getBroadcastState` instead?
   Based on my understanding that would better approximate our requirements for redistributing the state when rescaling. Even if there are some small discrepancies in the values between the different operation instances, we should not be concerned too much about them.
   
   WDYT?



-- 
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: issues-unsubscribe@iceberg.apache.org

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


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@iceberg.apache.org
For additional commands, e-mail: issues-help@iceberg.apache.org