Merge pull request #421 from metamx/use-druid-hyperloglog

Use druid implementation of HyperLogLog

indexing-hadoop/pom.xml

@@ -97,11 +97,6 @@
             <artifactId>junit</artifactId>
             <scope>test</scope>
         </dependency>
-      <dependency>
-        <groupId>com.clearspring.analytics</groupId>
-        <artifactId>stream</artifactId>
-        <version>2.5.2</version>
-      </dependency>
     </dependencies>
 
     <build>

indexing-hadoop/src/main/java/io/druid/indexer/DetermineHashedPartitionsJob.java

@@ -19,8 +19,6 @@
 
 package io.druid.indexer;
 
-import com.clearspring.analytics.stream.cardinality.CardinalityMergeException;
-import com.clearspring.analytics.stream.cardinality.HyperLogLog;
 import com.fasterxml.jackson.core.type.TypeReference;
 import com.google.common.base.Optional;
 import com.google.common.base.Throwables;
@@ -36,6 +34,7 @@ import io.druid.data.input.InputRow;
 import io.druid.data.input.Rows;
 import io.druid.granularity.QueryGranularity;
 import io.druid.indexer.granularity.UniformGranularitySpec;
+import io.druid.query.aggregation.hyperloglog.HyperLogLogCollector;
 import io.druid.timeline.partition.HashBasedNumberedShardSpec;
 import io.druid.timeline.partition.NoneShardSpec;
 import org.apache.hadoop.conf.Configuration;
@@ -56,6 +55,7 @@ import org.joda.time.Interval;
 
 import java.io.IOException;
 import java.io.OutputStream;
+import java.nio.ByteBuffer;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
@@ -67,7 +67,6 @@ public class DetermineHashedPartitionsJob implements Jobby
 {
   private static final int MAX_SHARDS = 128;
   private static final Logger log = new Logger(DetermineHashedPartitionsJob.class);
-  private static final int HYPER_LOG_LOG_BIT_SIZE = 20;
   private final HadoopDruidIndexerConfig config;
 
   public DetermineHashedPartitionsJob(
@@ -99,7 +98,7 @@ public class DetermineHashedPartitionsJob implements Jobby
       groupByJob.setOutputKeyClass(NullWritable.class);
       groupByJob.setOutputValueClass(NullWritable.class);
       groupByJob.setOutputFormatClass(SequenceFileOutputFormat.class);
-      if(!config.getSegmentGranularIntervals().isPresent()){
+      if (!config.getSegmentGranularIntervals().isPresent()) {
         groupByJob.setNumReduceTasks(1);
       }
       JobHelper.setupClasspath(config, groupByJob);
@@ -194,7 +193,7 @@ public class DetermineHashedPartitionsJob implements Jobby
   {
     private static HashFunction hashFunction = Hashing.murmur3_128();
     private QueryGranularity rollupGranularity = null;
-    private Map<Interval, HyperLogLog> hyperLogLogs;
+    private Map<Interval, HyperLogLogCollector> hyperLogLogs;
     private HadoopDruidIndexerConfig config;
     private boolean determineIntervals;
 
@@ -208,9 +207,9 @@ public class DetermineHashedPartitionsJob implements Jobby
       Optional<Set<Interval>> intervals = config.getSegmentGranularIntervals();
       if (intervals.isPresent()) {
         determineIntervals = false;
-        final ImmutableMap.Builder<Interval, HyperLogLog> builder = ImmutableMap.builder();
+        final ImmutableMap.Builder<Interval, HyperLogLogCollector> builder = ImmutableMap.builder();
         for (final Interval bucketInterval : intervals.get()) {
-          builder.put(bucketInterval, new HyperLogLog(HYPER_LOG_LOG_BIT_SIZE));
+          builder.put(bucketInterval, HyperLogLogCollector.makeLatestCollector());
         }
         hyperLogLogs = builder.build();
       } else {
@@ -236,7 +235,7 @@ public class DetermineHashedPartitionsJob implements Jobby
         interval = config.getGranularitySpec().getGranularity().bucket(new DateTime(inputRow.getTimestampFromEpoch()));
 
         if (!hyperLogLogs.containsKey(interval)) {
-          hyperLogLogs.put(interval, new HyperLogLog(HYPER_LOG_LOG_BIT_SIZE));
+          hyperLogLogs.put(interval, HyperLogLogCollector.makeLatestCollector());
         }
       } else {
         final Optional<Interval> maybeInterval = config.getGranularitySpec()
@@ -248,9 +247,9 @@ public class DetermineHashedPartitionsJob implements Jobby
         interval = maybeInterval.get();
       }
       hyperLogLogs.get(interval)
-                  .offerHashed(
+                  .add(
                       hashFunction.hashBytes(HadoopDruidIndexerConfig.jsonMapper.writeValueAsBytes(groupKey))
-                                  .asLong()
+                                  .asBytes()
                   );
     }
 
@@ -263,10 +262,10 @@ public class DetermineHashedPartitionsJob implements Jobby
         map(context.getCurrentKey(), context.getCurrentValue(), context);
       }
 
-      for (Map.Entry<Interval, HyperLogLog> entry : hyperLogLogs.entrySet()) {
+      for (Map.Entry<Interval, HyperLogLogCollector> entry : hyperLogLogs.entrySet()) {
         context.write(
             new LongWritable(entry.getKey().getStartMillis()),
-            new BytesWritable(entry.getValue().getBytes())
+            new BytesWritable(entry.getValue().toByteArray())
         );
       }
       cleanup(context);
@@ -294,15 +293,9 @@ public class DetermineHashedPartitionsJob implements Jobby
         Context context
     ) throws IOException, InterruptedException
     {
-      HyperLogLog aggregate = new HyperLogLog(HYPER_LOG_LOG_BIT_SIZE);
+      HyperLogLogCollector aggregate = HyperLogLogCollector.makeLatestCollector();
       for (BytesWritable value : values) {
-        HyperLogLog logValue = HyperLogLog.Builder.build(getDataBytes(value));
+        aggregate.fold(ByteBuffer.wrap(value.getBytes(), 0, value.getLength()));
-        try {
-          aggregate.addAll(logValue);
-        }
-        catch (CardinalityMergeException e) {
-          e.printStackTrace(); // TODO: check for better handling
-        }
       }
       Interval interval = config.getGranularitySpec().getGranularity().bucket(new DateTime(key.get()));
       intervals.add(interval);
@@ -318,7 +311,7 @@ public class DetermineHashedPartitionsJob implements Jobby
             }
         ).writeValue(
             out,
-            aggregate.cardinality()
+            new Double(aggregate.estimateCardinality()).longValue()
         );
       }
       finally {