public virtual void TestForceMergeTempSpaceUsage()
{
MockDirectoryWrapper dir = NewMockDirectory();
IndexWriter writer = new IndexWriter(dir, NewIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random)).SetMaxBufferedDocs(10).SetMergePolicy(NewLogMergePolicy()));
if (VERBOSE)
{
Console.WriteLine("TEST: config1=" + writer.Config);
}
for (int j = 0; j < 500; j++)
{
AddDocWithIndex(writer, j);
}
int termIndexInterval = writer.Config.TermIndexInterval;
// force one extra segment w/ different doc store so
// we see the doc stores get merged
writer.Commit();
AddDocWithIndex(writer, 500);
writer.Dispose();
if (VERBOSE)
{
Console.WriteLine("TEST: start disk usage");
}
long startDiskUsage = 0;
string[] files = dir.ListAll();
for (int i = 0; i < files.Length; i++)
{
startDiskUsage += dir.FileLength(files[i]);
if (VERBOSE)
{
Console.WriteLine(files[i] + ": " + dir.FileLength(files[i]));
}
}
dir.ResetMaxUsedSizeInBytes();
dir.TrackDiskUsage = true;
// Import to use same term index interval else a
// smaller one here could increase the disk usage and
// cause a false failure:
writer = new IndexWriter(dir, NewIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random)).SetOpenMode(OpenMode.APPEND).SetTermIndexInterval(termIndexInterval).SetMergePolicy(NewLogMergePolicy()));
writer.ForceMerge(1);
writer.Dispose();
long maxDiskUsage = dir.MaxUsedSizeInBytes;
Assert.IsTrue(maxDiskUsage <= 4 * startDiskUsage, "forceMerge used too much temporary space: starting usage was " + startDiskUsage + " bytes; max temp usage was " + maxDiskUsage + " but should have been " + (4 * startDiskUsage) + " (= 4X starting usage)");
dir.Dispose();
}
public virtual void TestAddIndexOnDiskFull()
{
// MemoryCodec, since it uses FST, is not necessarily
// "additive", ie if you add up N small FSTs, then merge
// them, the merged result can easily be larger than the
// sum because the merged FST may use array encoding for
// some arcs (which uses more space):
string idFormat = TestUtil.GetPostingsFormat("id");
string contentFormat = TestUtil.GetPostingsFormat("content");
AssumeFalse("this test cannot run with Memory codec", idFormat.Equals("Memory") || contentFormat.Equals("Memory"));
int START_COUNT = 57;
int NUM_DIR = TEST_NIGHTLY ? 50 : 5;
int END_COUNT = START_COUNT + NUM_DIR * (TEST_NIGHTLY ? 25 : 5);
// Build up a bunch of dirs that have indexes which we
// will then merge together by calling addIndexes(*):
Directory[] dirs = new Directory[NUM_DIR];
long inputDiskUsage = 0;
for (int i = 0; i < NUM_DIR; i++)
{
dirs[i] = NewDirectory();
IndexWriter writer = new IndexWriter(dirs[i], NewIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random())));
for (int j = 0; j < 25; j++)
{
AddDocWithIndex(writer, 25 * i + j);
}
writer.Dispose();
string[] files = dirs[i].ListAll();
for (int j = 0; j < files.Length; j++)
{
inputDiskUsage += dirs[i].FileLength(files[j]);
}
}
// Now, build a starting index that has START_COUNT docs. We
// will then try to addIndexes into a copy of this:
MockDirectoryWrapper startDir = NewMockDirectory();
IndexWriter indWriter = new IndexWriter(startDir, NewIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random())));
for (int j = 0; j < START_COUNT; j++)
{
AddDocWithIndex(indWriter, j);
}
indWriter.Dispose();
// Make sure starting index seems to be working properly:
Term searchTerm = new Term("content", "aaa");
IndexReader reader = DirectoryReader.Open(startDir);
Assert.AreEqual(57, reader.DocFreq(searchTerm), "first docFreq");
IndexSearcher searcher = NewSearcher(reader);
ScoreDoc[] hits = searcher.Search(new TermQuery(searchTerm), null, 1000).ScoreDocs;
Assert.AreEqual(57, hits.Length, "first number of hits");
reader.Dispose();
// Iterate with larger and larger amounts of free
// disk space. With little free disk space,
// addIndexes will certainly run out of space &
// fail. Verify that when this happens, index is
// not corrupt and index in fact has added no
// documents. Then, we increase disk space by 2000
// bytes each iteration. At some point there is
// enough free disk space and addIndexes should
// succeed and index should show all documents were
// added.
// String[] files = startDir.ListAll();
long diskUsage = startDir.GetSizeInBytes();
long startDiskUsage = 0;
string[] files_ = startDir.ListAll();
for (int i = 0; i < files_.Length; i++)
{
startDiskUsage += startDir.FileLength(files_[i]);
}
for (int iter = 0; iter < 3; iter++)
{
if (VERBOSE)
{
Console.WriteLine("TEST: iter=" + iter);
}
// Start with 100 bytes more than we are currently using:
long diskFree = diskUsage + TestUtil.NextInt(Random(), 50, 200);
int method = iter;
bool success = false;
bool done = false;
string methodName;
if (0 == method)
{
methodName = "addIndexes(Directory[]) + forceMerge(1)";
}
else if (1 == method)
{
methodName = "addIndexes(IndexReader[])";
}
else
{
methodName = "addIndexes(Directory[])";
}
while (!done)
{
if (VERBOSE)
{
Console.WriteLine("TEST: cycle...");
}
// Make a new dir that will enforce disk usage:
MockDirectoryWrapper dir = new MockDirectoryWrapper(Random(), new RAMDirectory(startDir, NewIOContext(Random())));
indWriter = new IndexWriter(dir, NewIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random())).SetOpenMode(OpenMode.APPEND).SetMergePolicy(NewLogMergePolicy(false)));
IOException err = null;
IMergeScheduler ms = indWriter.Config.MergeScheduler;
for (int x = 0; x < 2; x++)
{
if (ms is IConcurrentMergeScheduler)
// this test intentionally produces exceptions
// in the threads that CMS launches; we don't
// want to pollute test output with these.
{
if (0 == x)
{
((IConcurrentMergeScheduler)ms).SetSuppressExceptions();
}
else
{
((IConcurrentMergeScheduler)ms).ClearSuppressExceptions();
}
}
// Two loops: first time, limit disk space &
// throw random IOExceptions; second time, no
// disk space limit:
double rate = 0.05;
double diskRatio = ((double)diskFree) / diskUsage;
long thisDiskFree;
string testName = null;
if (0 == x)
{
dir.RandomIOExceptionRateOnOpen = Random().NextDouble() * 0.01;
thisDiskFree = diskFree;
if (diskRatio >= 2.0)
{
rate /= 2;
}
if (diskRatio >= 4.0)
{
rate /= 2;
}
if (diskRatio >= 6.0)
{
rate = 0.0;
}
if (VERBOSE)
{
testName = "disk full test " + methodName + " with disk full at " + diskFree + " bytes";
}
}
else
{
dir.RandomIOExceptionRateOnOpen = 0.0;
thisDiskFree = 0;
rate = 0.0;
if (VERBOSE)
{
testName = "disk full test " + methodName + " with unlimited disk space";
}
}
if (VERBOSE)
{
Console.WriteLine("\ncycle: " + testName);
}
dir.TrackDiskUsage = true;
dir.MaxSizeInBytes = thisDiskFree;
dir.RandomIOExceptionRate = rate;
try
{
if (0 == method)
{
if (VERBOSE)
{
Console.WriteLine("TEST: now addIndexes count=" + dirs.Length);
}
indWriter.AddIndexes(dirs);
if (VERBOSE)
{
Console.WriteLine("TEST: now forceMerge");
}
indWriter.ForceMerge(1);
}
else if (1 == method)
{
IndexReader[] readers = new IndexReader[dirs.Length];
for (int i = 0; i < dirs.Length; i++)
{
readers[i] = DirectoryReader.Open(dirs[i]);
}
try
{
indWriter.AddIndexes(readers);
}
finally
{
for (int i = 0; i < dirs.Length; i++)
{
readers[i].Dispose();
}
}
}
else
{
indWriter.AddIndexes(dirs);
}
success = true;
if (VERBOSE)
{
Console.WriteLine(" success!");
}
if (0 == x)
{
done = true;
}
}
catch (IOException e)
{
success = false;
err = e;
if (VERBOSE)
{
Console.WriteLine(" hit IOException: " + e);
Console.WriteLine(e.StackTrace);
}
if (1 == x)
{
Console.WriteLine(e.StackTrace);
Assert.Fail(methodName + " hit IOException after disk space was freed up");
}
}
// Make sure all threads from
// ConcurrentMergeScheduler are done
TestUtil.SyncConcurrentMerges(indWriter);
if (VERBOSE)
{
Console.WriteLine(" now test readers");
}
// Finally, verify index is not corrupt, and, if
// we succeeded, we see all docs added, and if we
// failed, we see either all docs or no docs added
// (transactional semantics):
dir.RandomIOExceptionRateOnOpen = 0.0;
try
{
reader = DirectoryReader.Open(dir);
}
catch (IOException e)
{
Console.WriteLine(e.StackTrace);
Assert.Fail(testName + ": exception when creating IndexReader: " + e);
}
int result = reader.DocFreq(searchTerm);
if (success)
{
if (result != START_COUNT)
{
Assert.Fail(testName + ": method did not throw exception but docFreq('aaa') is " + result + " instead of expected " + START_COUNT);
}
}
else
{
// On hitting exception we still may have added
// all docs:
if (result != START_COUNT && result != END_COUNT)
{
Console.WriteLine(err.StackTrace);
Assert.Fail(testName + ": method did throw exception but docFreq('aaa') is " + result + " instead of expected " + START_COUNT + " or " + END_COUNT);
}
}
searcher = NewSearcher(reader);
try
{
hits = searcher.Search(new TermQuery(searchTerm), null, END_COUNT).ScoreDocs;
}
catch (IOException e)
{
Console.WriteLine(e.StackTrace);
Assert.Fail(testName + ": exception when searching: " + e);
}
int result2 = hits.Length;
if (success)
{
if (result2 != result)
{
Assert.Fail(testName + ": method did not throw exception but hits.Length for search on term 'aaa' is " + result2 + " instead of expected " + result);
}
}
else
{
// On hitting exception we still may have added
// all docs:
if (result2 != result)
{
Console.WriteLine(err.StackTrace);
Assert.Fail(testName + ": method did throw exception but hits.Length for search on term 'aaa' is " + result2 + " instead of expected " + result);
}
}
reader.Dispose();
if (VERBOSE)
{
Console.WriteLine(" count is " + result);
}
if (done || result == END_COUNT)
{
break;
}
}
if (VERBOSE)
{
Console.WriteLine(" start disk = " + startDiskUsage + "; input disk = " + inputDiskUsage + "; max used = " + dir.MaxUsedSizeInBytes);
}
if (done)
{
// Javadocs state that temp free Directory space
// required is at most 2X total input size of
// indices so let's make sure:
Assert.IsTrue((dir.MaxUsedSizeInBytes - startDiskUsage) < 2 * (startDiskUsage + inputDiskUsage), "max free Directory space required exceeded 1X the total input index sizes during " + methodName + ": max temp usage = " + (dir.MaxUsedSizeInBytes - startDiskUsage) + " bytes vs limit=" + (2 * (startDiskUsage + inputDiskUsage)) + "; starting disk usage = " + startDiskUsage + " bytes; " + "input index disk usage = " + inputDiskUsage + " bytes");
}
// Make sure we don't hit disk full during close below:
dir.MaxSizeInBytes = 0;
dir.RandomIOExceptionRate = 0.0;
dir.RandomIOExceptionRateOnOpen = 0.0;
indWriter.Dispose();
// Wait for all BG threads to finish else
// dir.Dispose() will throw IOException because
// there are still open files
TestUtil.SyncConcurrentMerges(ms);
dir.Dispose();
// Try again with more free space:
diskFree += TEST_NIGHTLY ? TestUtil.NextInt(Random(), 4000, 8000) : TestUtil.NextInt(Random(), 40000, 80000);
}
}
startDir.Dispose();
foreach (Directory dir in dirs)
{
dir.Dispose();
}
}
public virtual void Test([ValueSource(typeof(ConcurrentMergeSchedulerFactories), "Values")] Func <IConcurrentMergeScheduler> newScheduler)
{
MockDirectoryWrapper dir = new MockDirectoryWrapper(Random(), new MMapDirectory(CreateTempDir("4GBStoredFields")));
dir.Throttling = MockDirectoryWrapper.Throttling_e.NEVER;
var config = new IndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random()))
.SetMaxBufferedDocs(IndexWriterConfig.DISABLE_AUTO_FLUSH)
.SetRAMBufferSizeMB(256.0)
.SetMergeScheduler(newScheduler())
.SetMergePolicy(NewLogMergePolicy(false, 10))
.SetOpenMode(OpenMode.CREATE);
IndexWriter w = new IndexWriter(dir, config);
MergePolicy mp = w.Config.MergePolicy;
if (mp is LogByteSizeMergePolicy)
{
// 1 petabyte:
((LogByteSizeMergePolicy)mp).MaxMergeMB = 1024 * 1024 * 1024;
}
Document doc = new Document();
FieldType ft = new FieldType();
ft.IsIndexed = false;
ft.IsStored = true;
ft.Freeze();
int valueLength = RandomInts.NextIntBetween(Random(), 1 << 13, 1 << 20);
var value = new byte[valueLength];
for (int i = 0; i < valueLength; ++i)
{
// random so that even compressing codecs can't compress it
value[i] = (byte)Random().Next(256);
}
Field f = new Field("fld", value, ft);
doc.Add(f);
int numDocs = (int)((1L << 32) / valueLength + 100);
for (int i = 0; i < numDocs; ++i)
{
w.AddDocument(doc);
if (VERBOSE && i % (numDocs / 10) == 0)
{
Console.WriteLine(i + " of " + numDocs + "...");
}
}
w.ForceMerge(1);
w.Dispose();
if (VERBOSE)
{
bool found = false;
foreach (string file in dir.ListAll())
{
if (file.EndsWith(".fdt", StringComparison.Ordinal))
{
long fileLength = dir.FileLength(file);
if (fileLength >= 1L << 32)
{
found = true;
}
Console.WriteLine("File length of " + file + " : " + fileLength);
}
}
if (!found)
{
Console.WriteLine("No .fdt file larger than 4GB, test bug?");
}
}
DirectoryReader rd = DirectoryReader.Open(dir);
Document sd = rd.Document(numDocs - 1);
Assert.IsNotNull(sd);
Assert.AreEqual(1, sd.Fields.Count);
BytesRef valueRef = sd.GetBinaryValue("fld");
Assert.IsNotNull(valueRef);
Assert.AreEqual(new BytesRef(value), valueRef);
rd.Dispose();
dir.Dispose();
}
public virtual void Test([ValueSource(typeof(ConcurrentMergeSchedulers), "Values")]IConcurrentMergeScheduler scheduler)
{
MockDirectoryWrapper dir = new MockDirectoryWrapper(Random(), new MMapDirectory(CreateTempDir("4GBStoredFields")));
dir.Throttling = MockDirectoryWrapper.Throttling_e.NEVER;
var config = new IndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(Random()))
.SetMaxBufferedDocs(IndexWriterConfig.DISABLE_AUTO_FLUSH)
.SetRAMBufferSizeMB(256.0)
.SetMergeScheduler(scheduler)
.SetMergePolicy(NewLogMergePolicy(false, 10))
.SetOpenMode(IndexWriterConfig.OpenMode_e.CREATE);
IndexWriter w = new IndexWriter(dir, config);
MergePolicy mp = w.Config.MergePolicy;
if (mp is LogByteSizeMergePolicy)
{
// 1 petabyte:
((LogByteSizeMergePolicy)mp).MaxMergeMB = 1024 * 1024 * 1024;
}
Document doc = new Document();
FieldType ft = new FieldType();
ft.Indexed = false;
ft.Stored = true;
ft.Freeze();
int valueLength = RandomInts.NextIntBetween(Random(), 1 << 13, 1 << 20);
var value = new byte[valueLength];
for (int i = 0; i < valueLength; ++i)
{
// random so that even compressing codecs can't compress it
value[i] = (byte)Random().Next(256);
}
Field f = new Field("fld", value, ft);
doc.Add(f);
int numDocs = (int)((1L << 32) / valueLength + 100);
for (int i = 0; i < numDocs; ++i)
{
w.AddDocument(doc);
if (VERBOSE && i % (numDocs / 10) == 0)
{
Console.WriteLine(i + " of " + numDocs + "...");
}
}
w.ForceMerge(1);
w.Dispose();
if (VERBOSE)
{
bool found = false;
foreach (string file in dir.ListAll())
{
if (file.EndsWith(".fdt"))
{
long fileLength = dir.FileLength(file);
if (fileLength >= 1L << 32)
{
found = true;
}
Console.WriteLine("File length of " + file + " : " + fileLength);
}
}
if (!found)
{
Console.WriteLine("No .fdt file larger than 4GB, test bug?");
}
}
DirectoryReader rd = DirectoryReader.Open(dir);
Document sd = rd.Document(numDocs - 1);
Assert.IsNotNull(sd);
Assert.AreEqual(1, sd.Fields.Count);
BytesRef valueRef = sd.GetBinaryValue("fld");
Assert.IsNotNull(valueRef);
Assert.AreEqual(new BytesRef(value), valueRef);
rd.Dispose();
dir.Dispose();
}
