Github user tammymendt commented on a diff in the pull request:
https://github.com/apache/flink/pull/605#discussion_r28776656
--- Diff:
flink-core/src/main/java/org/apache/flink/statistics/OperatorStatistics.java ---
@@ -0,0 +1,154 @@
+/*
+ * 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.statistics;
+
+import
com.clearspring.analytics.stream.cardinality.CardinalityMergeException;
+import com.clearspring.analytics.stream.cardinality.HyperLogLog;
+import com.clearspring.analytics.stream.cardinality.ICardinality;
+import com.clearspring.analytics.stream.cardinality.LinearCounting;
+import org.apache.flink.statistics.heavyhitters.IHeavyHitter;
+import org.apache.flink.statistics.heavyhitters.LossyCounting;
+import org.apache.flink.statistics.heavyhitters.CountMinHeavyHitter;
+import org.apache.flink.statistics.heavyhitters.HeavyHitterMergeException;
+
+import java.io.Serializable;
+import java.util.Map;
+
+/**
+ * Data structure that encapsulates statistical information of data that
has only been processed by one pass
+ * This statistical information is meant to help determine the
distribution of the data that has been processed
+ * in an operator so that we can determine if it is necessary to
repartition the data
+ *
+ * The statistics to be gathered are configurable and represented by a
{@link OperatorStatisticsConfig} object.
+ *
+ * The information encapsulated in this class is min, max, a structure
enabling estimation of count distinct and a
+ * structure holding the heavy hitters along with their frequency.
+ *
+ */
+public class OperatorStatistics implements Serializable {
+
+ OperatorStatisticsConfig config;
+
+ Object min;
+ Object max;
+ ICardinality countDistinct;
+ IHeavyHitter heavyHitter;
+ long cardinality = 0;
+
+ public OperatorStatistics(OperatorStatisticsConfig config) {
+ this.config = config;
+ if
(config.countDistinctAlgorithm.equals(OperatorStatisticsConfig.CountDistinctAlgorithm.LINEAR_COUNTING))
{
+ countDistinct = new
LinearCounting(OperatorStatisticsConfig.COUNTD_BITMAP_SIZE);
+ }
+
if(config.countDistinctAlgorithm.equals(OperatorStatisticsConfig.CountDistinctAlgorithm.HYPERLOGLOG)){
+ countDistinct = new
HyperLogLog(OperatorStatisticsConfig.COUNTD_LOG2M);
+ }
+ if
(config.heavyHitterAlgorithm.equals(OperatorStatisticsConfig.HeavyHitterAlgorithm.LOSSY_COUNTING)){
+ heavyHitter =
+ new
LossyCounting(OperatorStatisticsConfig.HEAVY_HITTER_FRACTION,
OperatorStatisticsConfig.HEAVY_HITTER_ERROR);
+ }
+ if
(config.heavyHitterAlgorithm.equals(OperatorStatisticsConfig.HeavyHitterAlgorithm.COUNT_MIN_SKETCH)){
+ heavyHitter =
+ new
CountMinHeavyHitter(OperatorStatisticsConfig.HEAVY_HITTER_FRACTION,
+
OperatorStatisticsConfig.HEAVY_HITTER_ERROR,
+
OperatorStatisticsConfig.HEAVY_HITTER_CONFIDENCE,
+
OperatorStatisticsConfig.HEAVY_HITTER_SEED);
+ }
+ }
+
+ public void process(Object tupleObject){
--- End diff --
The problem with processing only every n-th element is that precision would
be affected, and the algorithms are already estimating as it is. I am planning
to measure performance overhead but haven't done it so far.
I have removed unnecessary checks inside the "process" function. The point
of them was to allow the user to configure which stats wished to be tracked, by
means of the OperatorStatisticsConfig class. This way for example an
OperatorStatisticsAccumulator could be defined, but only used to track count
distinct. Do you think this makes sense, or is it preferable to create multiple
accumulators, one for each type of statistic? I was avoiding this because the
common case would be to want all 4 statistics (max, min, heavy hitters and
count distinct), and I didn't want the user to have to create 4 separate
accumulators for this purpose.
---
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.
---
