public virtual void TestInMemoryAndOnDiskMerger()
{
JobID jobId = new JobID("a", 0);
TaskAttemptID reduceId1 = new TaskAttemptID(new TaskID(jobId, TaskType.Reduce, 0)
, 0);
TaskAttemptID mapId1 = new TaskAttemptID(new TaskID(jobId, TaskType.Map, 1), 0);
TaskAttemptID mapId2 = new TaskAttemptID(new TaskID(jobId, TaskType.Map, 2), 0);
LocalDirAllocator lda = new LocalDirAllocator(MRConfig.LocalDir);
MergeManagerImpl <Text, Text> mergeManager = new MergeManagerImpl <Text, Text>(reduceId1
, jobConf, fs, lda, Reporter.Null, null, null, null, null, null, null, null, new
Progress(), new MROutputFiles());
// write map outputs
IDictionary <string, string> map1 = new SortedDictionary <string, string>();
map1["apple"] = "disgusting";
map1["carrot"] = "delicious";
IDictionary <string, string> map2 = new SortedDictionary <string, string>();
map1["banana"] = "pretty good";
byte[] mapOutputBytes1 = WriteMapOutput(conf, map1);
byte[] mapOutputBytes2 = WriteMapOutput(conf, map2);
InMemoryMapOutput <Text, Text> mapOutput1 = new InMemoryMapOutput <Text, Text>(conf
, mapId1, mergeManager, mapOutputBytes1.Length, null, true);
InMemoryMapOutput <Text, Text> mapOutput2 = new InMemoryMapOutput <Text, Text>(conf
, mapId2, mergeManager, mapOutputBytes2.Length, null, true);
System.Array.Copy(mapOutputBytes1, 0, mapOutput1.GetMemory(), 0, mapOutputBytes1.
Length);
System.Array.Copy(mapOutputBytes2, 0, mapOutput2.GetMemory(), 0, mapOutputBytes2.
Length);
// create merger and run merge
MergeThread <InMemoryMapOutput <Text, Text>, Text, Text> inMemoryMerger = mergeManager
.CreateInMemoryMerger();
IList <InMemoryMapOutput <Text, Text> > mapOutputs1 = new AList <InMemoryMapOutput <Text
, Text> >();
mapOutputs1.AddItem(mapOutput1);
mapOutputs1.AddItem(mapOutput2);
inMemoryMerger.Merge(mapOutputs1);
NUnit.Framework.Assert.AreEqual(1, mergeManager.onDiskMapOutputs.Count);
TaskAttemptID reduceId2 = new TaskAttemptID(new TaskID(jobId, TaskType.Reduce, 3)
, 0);
TaskAttemptID mapId3 = new TaskAttemptID(new TaskID(jobId, TaskType.Map, 4), 0);
TaskAttemptID mapId4 = new TaskAttemptID(new TaskID(jobId, TaskType.Map, 5), 0);
// write map outputs
IDictionary <string, string> map3 = new SortedDictionary <string, string>();
map3["apple"] = "awesome";
map3["carrot"] = "amazing";
IDictionary <string, string> map4 = new SortedDictionary <string, string>();
map4["banana"] = "bla";
byte[] mapOutputBytes3 = WriteMapOutput(conf, map3);
byte[] mapOutputBytes4 = WriteMapOutput(conf, map4);
InMemoryMapOutput <Text, Text> mapOutput3 = new InMemoryMapOutput <Text, Text>(conf
, mapId3, mergeManager, mapOutputBytes3.Length, null, true);
InMemoryMapOutput <Text, Text> mapOutput4 = new InMemoryMapOutput <Text, Text>(conf
, mapId4, mergeManager, mapOutputBytes4.Length, null, true);
System.Array.Copy(mapOutputBytes3, 0, mapOutput3.GetMemory(), 0, mapOutputBytes3.
Length);
System.Array.Copy(mapOutputBytes4, 0, mapOutput4.GetMemory(), 0, mapOutputBytes4.
Length);
// // create merger and run merge
MergeThread <InMemoryMapOutput <Text, Text>, Text, Text> inMemoryMerger2 = mergeManager
.CreateInMemoryMerger();
IList <InMemoryMapOutput <Text, Text> > mapOutputs2 = new AList <InMemoryMapOutput <Text
, Text> >();
mapOutputs2.AddItem(mapOutput3);
mapOutputs2.AddItem(mapOutput4);
inMemoryMerger2.Merge(mapOutputs2);
NUnit.Framework.Assert.AreEqual(2, mergeManager.onDiskMapOutputs.Count);
IList <MergeManagerImpl.CompressAwarePath> paths = new AList <MergeManagerImpl.CompressAwarePath
>();
IEnumerator <MergeManagerImpl.CompressAwarePath> iterator = mergeManager.onDiskMapOutputs
.GetEnumerator();
IList <string> keys = new AList <string>();
IList <string> values = new AList <string>();
while (iterator.HasNext())
{
MergeManagerImpl.CompressAwarePath next = iterator.Next();
ReadOnDiskMapOutput(conf, fs, next, keys, values);
paths.AddItem(next);
}
NUnit.Framework.Assert.AreEqual(keys, Arrays.AsList("apple", "banana", "carrot",
"apple", "banana", "carrot"));
NUnit.Framework.Assert.AreEqual(values, Arrays.AsList("awesome", "bla", "amazing"
, "disgusting", "pretty good", "delicious"));
mergeManager.Close();
mergeManager = new MergeManagerImpl <Text, Text>(reduceId2, jobConf, fs, lda, Reporter
.Null, null, null, null, null, null, null, null, new Progress(), new MROutputFiles
());
MergeThread <MergeManagerImpl.CompressAwarePath, Text, Text> onDiskMerger = mergeManager
.CreateOnDiskMerger();
onDiskMerger.Merge(paths);
NUnit.Framework.Assert.AreEqual(1, mergeManager.onDiskMapOutputs.Count);
keys = new AList <string>();
values = new AList <string>();
ReadOnDiskMapOutput(conf, fs, mergeManager.onDiskMapOutputs.GetEnumerator().Next(
), keys, values);
NUnit.Framework.Assert.AreEqual(keys, Arrays.AsList("apple", "apple", "banana", "banana"
, "carrot", "carrot"));
NUnit.Framework.Assert.AreEqual(values, Arrays.AsList("awesome", "disgusting", "pretty good"
, "bla", "amazing", "delicious"));
mergeManager.Close();
NUnit.Framework.Assert.AreEqual(0, mergeManager.inMemoryMapOutputs.Count);
NUnit.Framework.Assert.AreEqual(0, mergeManager.inMemoryMergedMapOutputs.Count);
NUnit.Framework.Assert.AreEqual(0, mergeManager.onDiskMapOutputs.Count);
}
public MergeManagerImpl(TaskAttemptID reduceId, JobConf jobConf, FileSystem localFS
, LocalDirAllocator localDirAllocator, Reporter reporter, CompressionCodec codec
, Type combinerClass, Task.CombineOutputCollector <K, V> combineCollector, Counters.Counter
spilledRecordsCounter, Counters.Counter reduceCombineInputCounter, Counters.Counter
mergedMapOutputsCounter, ExceptionReporter exceptionReporter, Progress mergePhase
, MapOutputFile mapOutputFile)
{
/* Maximum percentage of the in-memory limit that a single shuffle can
* consume*/
this.reduceId = reduceId;
this.jobConf = jobConf;
this.localDirAllocator = localDirAllocator;
this.exceptionReporter = exceptionReporter;
this.reporter = reporter;
this.codec = codec;
this.combinerClass = combinerClass;
this.combineCollector = combineCollector;
this.reduceCombineInputCounter = reduceCombineInputCounter;
this.spilledRecordsCounter = spilledRecordsCounter;
this.mergedMapOutputsCounter = mergedMapOutputsCounter;
this.mapOutputFile = mapOutputFile;
this.mapOutputFile.SetConf(jobConf);
this.localFS = localFS;
this.rfs = ((LocalFileSystem)localFS).GetRaw();
float maxInMemCopyUse = jobConf.GetFloat(MRJobConfig.ShuffleInputBufferPercent, MRJobConfig
.DefaultShuffleInputBufferPercent);
if (maxInMemCopyUse > 1.0 || maxInMemCopyUse < 0.0)
{
throw new ArgumentException("Invalid value for " + MRJobConfig.ShuffleInputBufferPercent
+ ": " + maxInMemCopyUse);
}
// Allow unit tests to fix Runtime memory
this.memoryLimit = (long)(jobConf.GetLong(MRJobConfig.ReduceMemoryTotalBytes, Runtime
.GetRuntime().MaxMemory()) * maxInMemCopyUse);
this.ioSortFactor = jobConf.GetInt(MRJobConfig.IoSortFactor, 100);
float singleShuffleMemoryLimitPercent = jobConf.GetFloat(MRJobConfig.ShuffleMemoryLimitPercent
, DefaultShuffleMemoryLimitPercent);
if (singleShuffleMemoryLimitPercent <= 0.0f || singleShuffleMemoryLimitPercent >
1.0f)
{
throw new ArgumentException("Invalid value for " + MRJobConfig.ShuffleMemoryLimitPercent
+ ": " + singleShuffleMemoryLimitPercent);
}
usedMemory = 0L;
commitMemory = 0L;
this.maxSingleShuffleLimit = (long)(memoryLimit * singleShuffleMemoryLimitPercent
);
this.memToMemMergeOutputsThreshold = jobConf.GetInt(MRJobConfig.ReduceMemtomemThreshold
, ioSortFactor);
this.mergeThreshold = (long)(this.memoryLimit * jobConf.GetFloat(MRJobConfig.ShuffleMergePercent
, 0.90f));
Log.Info("MergerManager: memoryLimit=" + memoryLimit + ", " + "maxSingleShuffleLimit="
+ maxSingleShuffleLimit + ", " + "mergeThreshold=" + mergeThreshold + ", " + "ioSortFactor="
+ ioSortFactor + ", " + "memToMemMergeOutputsThreshold=" + memToMemMergeOutputsThreshold
);
if (this.maxSingleShuffleLimit >= this.mergeThreshold)
{
throw new RuntimeException("Invalid configuration: " + "maxSingleShuffleLimit should be less than mergeThreshold "
+ "maxSingleShuffleLimit: " + this.maxSingleShuffleLimit + "mergeThreshold: " +
this.mergeThreshold);
}
bool allowMemToMemMerge = jobConf.GetBoolean(MRJobConfig.ReduceMemtomemEnabled, false
);
if (allowMemToMemMerge)
{
this.memToMemMerger = new MergeManagerImpl.IntermediateMemoryToMemoryMerger(this,
this, memToMemMergeOutputsThreshold);
this.memToMemMerger.Start();
}
else
{
this.memToMemMerger = null;
}
this.inMemoryMerger = CreateInMemoryMerger();
this.inMemoryMerger.Start();
this.onDiskMerger = new MergeManagerImpl.OnDiskMerger(this, this);
this.onDiskMerger.Start();
this.mergePhase = mergePhase;
}
/// <exception cref="System.IO.IOException"/>
/// <exception cref="Sharpen.URISyntaxException"/>
/// <exception cref="System.Exception"/>
public virtual void TestOnDiskMerger()
{
JobConf jobConf = new JobConf();
int SortFactor = 5;
jobConf.SetInt(MRJobConfig.IoSortFactor, SortFactor);
MapOutputFile mapOutputFile = new MROutputFiles();
FileSystem fs = FileSystem.GetLocal(jobConf);
MergeManagerImpl <IntWritable, IntWritable> manager = new MergeManagerImpl <IntWritable
, IntWritable>(null, jobConf, fs, null, null, null, null, null, null, null, null
, null, null, mapOutputFile);
MergeThread <MapOutput <IntWritable, IntWritable>, IntWritable, IntWritable> onDiskMerger
= (MergeThread <MapOutput <IntWritable, IntWritable>, IntWritable, IntWritable>)Whitebox
.GetInternalState(manager, "onDiskMerger");
int mergeFactor = (int)Whitebox.GetInternalState(onDiskMerger, "mergeFactor");
// make sure the io.sort.factor is set properly
NUnit.Framework.Assert.AreEqual(mergeFactor, SortFactor);
// Stop the onDiskMerger thread so that we can intercept the list of files
// waiting to be merged.
onDiskMerger.Suspend();
//Send the list of fake files waiting to be merged
Random rand = new Random();
for (int i = 0; i < 2 * SortFactor; ++i)
{
Path path = new Path("somePath");
MergeManagerImpl.CompressAwarePath cap = new MergeManagerImpl.CompressAwarePath(path
, 1l, rand.Next());
manager.CloseOnDiskFile(cap);
}
//Check that the files pending to be merged are in sorted order.
List <IList <MergeManagerImpl.CompressAwarePath> > pendingToBeMerged = (List <IList <MergeManagerImpl.CompressAwarePath
> >)Whitebox.GetInternalState(onDiskMerger, "pendingToBeMerged");
NUnit.Framework.Assert.IsTrue("No inputs were added to list pending to merge", pendingToBeMerged
.Count > 0);
for (int i_1 = 0; i_1 < pendingToBeMerged.Count; ++i_1)
{
IList <MergeManagerImpl.CompressAwarePath> inputs = pendingToBeMerged[i_1];
for (int j = 1; j < inputs.Count; ++j)
{
NUnit.Framework.Assert.IsTrue("Not enough / too many inputs were going to be merged"
, inputs.Count > 0 && inputs.Count <= SortFactor);
NUnit.Framework.Assert.IsTrue("Inputs to be merged were not sorted according to size: "
, inputs[j].GetCompressedSize() >= inputs[j - 1].GetCompressedSize());
}
}
}