Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/1255#discussion_r43005310
--- Diff:
flink-java/src/main/java/org/apache/flink/api/java/functions/AssignRangeIndex.java
---
@@ -0,0 +1,88 @@
+/*
+ * 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.flink.api.java.functions;
+
+import org.apache.flink.api.common.distributions.DataDistribution;
+import org.apache.flink.api.common.functions.RichMapPartitionFunction;
+import org.apache.flink.api.java.tuple.Tuple2;
+import org.apache.flink.configuration.Configuration;
+import org.apache.flink.util.Collector;
+
+import java.util.ArrayList;
+import java.util.List;
+
+/**
+ * This mapPartition function require a DataSet with DataDistribution as
broadcast input, it read
+ * target parallelism from parameter, build partition boundaries with
input DataDistribution, then
+ * compute the range index for each record.
+ *
+ * @param <IN> The original data type.
+ * @param <K> The key type.
+ */
+public class AssignRangeIndex<IN, K extends Comparable<K>>
+ extends RichMapPartitionFunction<Tuple2<K, IN>, Tuple2<Integer, IN>> {
+
+ private List<K> partitionBoundaries;
+ private int numberChannels;
+
+ @Override
+ public void open(Configuration parameters) throws Exception {
+ this.numberChannels =
parameters.getInteger("TargetParallelism", 1);
+ }
+
+ @Override
+ public void mapPartition(Iterable<Tuple2<K, IN>> values,
Collector<Tuple2<Integer, IN>> out) throws Exception {
+
+ List<Object> broadcastVariable =
getRuntimeContext().getBroadcastVariable("DataDistribution");
+ if (broadcastVariable == null || broadcastVariable.size() != 1)
{
+ throw new RuntimeException("AssignRangePartition
require a single DataDistribution as broadcast input.");
+ }
+ DataDistribution<K> dataDistribution = (DataDistribution<K>)
broadcastVariable.get(0);
+
+ partitionBoundaries = new ArrayList<>(numberChannels);
+ for (int i=0; i<numberChannels - 1; i++) {
--- End diff --
Each Assigner will independently initialize the boundaries and have its own
copy of it. It would be better if you could do that only once in a single
operator (I believe you had an AllReduce, i.e., a reduce without groupBy()
before) and broadcast the result of this operator. The benefit is that the
boundaries are only built once and all tasks on a TaskManager share the same
broadcast variable, i.e., there are not multiple copies of the boundaries.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
