/*
Test: make sure when we run out of disk space or hit
random IOExceptions in any of the addIndexes(*) calls
that 1) index is not corrupt (searcher can open/search
it) and 2) transactional semantics are followed:
either all or none of the incoming documents were in
fact added.
*/
public void testAddIndexOnDiskFull() throws IOException {
int START_COUNT = 57;
int NUM_DIR = 50;
int END_COUNT = START_COUNT + NUM_DIR * 25;
// 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.close();
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 addIndexesNoOptimize into a copy of this:
MockDirectoryWrapper startDir = newDirectory();
IndexWriter writer =
new IndexWriter(
startDir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random)));
for (int j = 0; j < START_COUNT; j++) {
addDocWithIndex(writer, j);
}
writer.close();
// Make sure starting index seems to be working properly:
Term searchTerm = new Term("content", "aaa");
IndexReader reader = IndexReader.open(startDir, true);
assertEquals("first docFreq", 57, reader.docFreq(searchTerm));
IndexSearcher searcher = newSearcher(reader);
ScoreDoc[] hits = searcher.search(new TermQuery(searchTerm), null, 1000).scoreDocs;
assertEquals("first number of hits", 57, hits.length);
searcher.close();
reader.close();
// 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.sizeInBytes();
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) System.out.println("TEST: iter=" + iter);
// Start with 100 bytes more than we are currently using:
long diskFree = diskUsage + _TestUtil.nextInt(random, 50, 200);
int method = iter;
boolean success = false;
boolean done = false;
String methodName;
if (0 == method) {
methodName = "addIndexes(Directory[]) + optimize()";
} else if (1 == method) {
methodName = "addIndexes(IndexReader[])";
} else {
methodName = "addIndexes(Directory[])";
}
while (!done) {
if (VERBOSE) {
System.out.println("TEST: cycle...");
}
// Make a new dir that will enforce disk usage:
MockDirectoryWrapper dir = new MockDirectoryWrapper(random, new RAMDirectory(startDir));
writer =
new IndexWriter(
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random))
.setOpenMode(OpenMode.APPEND)
.setMergePolicy(newLogMergePolicy()));
IOException err = null;
writer.setInfoStream(VERBOSE ? System.out : null);
MergeScheduler ms = writer.getConfig().getMergeScheduler();
for (int x = 0; x < 2; x++) {
if (ms instanceof ConcurrentMergeScheduler)
// This test intentionally produces exceptions
// in the threads that CMS launches; we don't
// want to pollute test output with these.
if (0 == x) ((ConcurrentMergeScheduler) ms).setSuppressExceptions();
else ((ConcurrentMergeScheduler) 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) {
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 {
thisDiskFree = 0;
rate = 0.0;
if (VERBOSE) testName = "disk full test " + methodName + " with unlimited disk space";
}
if (VERBOSE) System.out.println("\ncycle: " + testName);
dir.setTrackDiskUsage(true);
dir.setMaxSizeInBytes(thisDiskFree);
dir.setRandomIOExceptionRate(rate);
try {
if (0 == method) {
writer.addIndexes(dirs);
writer.optimize();
} else if (1 == method) {
IndexReader readers[] = new IndexReader[dirs.length];
for (int i = 0; i < dirs.length; i++) {
readers[i] = IndexReader.open(dirs[i], true);
}
try {
writer.addIndexes(readers);
} finally {
for (int i = 0; i < dirs.length; i++) {
readers[i].close();
}
}
} else {
writer.addIndexes(dirs);
}
success = true;
if (VERBOSE) {
System.out.println(" success!");
}
if (0 == x) {
done = true;
}
} catch (IOException e) {
success = false;
err = e;
if (VERBOSE) {
System.out.println(" hit IOException: " + e);
e.printStackTrace(System.out);
}
if (1 == x) {
e.printStackTrace(System.out);
fail(methodName + " hit IOException after disk space was freed up");
}
}
// Make sure all threads from
// ConcurrentMergeScheduler are done
_TestUtil.syncConcurrentMerges(writer);
if (VERBOSE) {
System.out.println(" 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):
try {
reader = IndexReader.open(dir, true);
} catch (IOException e) {
e.printStackTrace(System.out);
fail(testName + ": exception when creating IndexReader: " + e);
}
int result = reader.docFreq(searchTerm);
if (success) {
if (result != START_COUNT) {
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) {
err.printStackTrace(System.out);
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) {
e.printStackTrace(System.out);
fail(testName + ": exception when searching: " + e);
}
int result2 = hits.length;
if (success) {
if (result2 != result) {
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) {
err.printStackTrace(System.out);
fail(
testName
+ ": method did throw exception but hits.length for search on term 'aaa' is "
+ result2
+ " instead of expected "
+ result);
}
}
searcher.close();
reader.close();
if (VERBOSE) {
System.out.println(" count is " + result);
}
if (done || result == END_COUNT) {
break;
}
}
if (VERBOSE) {
System.out.println(
" start disk = "
+ startDiskUsage
+ "; input disk = "
+ inputDiskUsage
+ "; max used = "
+ dir.getMaxUsedSizeInBytes());
}
if (done) {
// Javadocs state that temp free Directory space
// required is at most 2X total input size of
// indices so let's make sure:
assertTrue(
"max free Directory space required exceeded 1X the total input index sizes during "
+ methodName
+ ": max temp usage = "
+ (dir.getMaxUsedSizeInBytes() - startDiskUsage)
+ " bytes vs limit="
+ (2 * (startDiskUsage + inputDiskUsage))
+ "; starting disk usage = "
+ startDiskUsage
+ " bytes; "
+ "input index disk usage = "
+ inputDiskUsage
+ " bytes",
(dir.getMaxUsedSizeInBytes() - startDiskUsage)
< 2 * (startDiskUsage + inputDiskUsage));
}
// Make sure we don't hit disk full during close below:
dir.setMaxSizeInBytes(0);
dir.setRandomIOExceptionRate(0.0);
writer.close();
// Wait for all BG threads to finish else
// dir.close() will throw IOException because
// there are still open files
_TestUtil.syncConcurrentMerges(ms);
dir.close();
// Try again with more free space:
diskFree +=
TEST_NIGHTLY
? _TestUtil.nextInt(random, 4000, 8000)
: _TestUtil.nextInt(random, 40000, 80000);
}
}
startDir.close();
for (Directory dir : dirs) dir.close();
}
/*
* Verify that a writer with "commit on close" indeed
* cleans up the temp segments created after opening
* that are not referenced by the starting segments
* file. We check this by using MockDirectoryWrapper to
* measure max temp disk space used.
*/
public void testCommitOnCloseDiskUsage() throws IOException {
// 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):
final String idFormat = _TestUtil.getPostingsFormat("id");
final String contentFormat = _TestUtil.getPostingsFormat("content");
assumeFalse(
"This test cannot run with Memory codec",
idFormat.equals("Memory") || contentFormat.equals("Memory"));
MockDirectoryWrapper dir = newDirectory();
Analyzer analyzer;
if (random().nextBoolean()) {
// no payloads
analyzer =
new Analyzer() {
@Override
public TokenStreamComponents createComponents(String fieldName, Reader reader) {
return new TokenStreamComponents(
new MockTokenizer(reader, MockTokenizer.WHITESPACE, true));
}
};
} else {
// fixed length payloads
final int length = random().nextInt(200);
analyzer =
new Analyzer() {
@Override
public TokenStreamComponents createComponents(String fieldName, Reader reader) {
Tokenizer tokenizer = new MockTokenizer(reader, MockTokenizer.WHITESPACE, true);
return new TokenStreamComponents(
tokenizer, new MockFixedLengthPayloadFilter(random(), tokenizer, length));
}
};
}
IndexWriter writer =
new IndexWriter(
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, analyzer)
.setMaxBufferedDocs(10)
.setReaderPooling(false)
.setMergePolicy(newLogMergePolicy(10)));
for (int j = 0; j < 30; j++) {
TestIndexWriter.addDocWithIndex(writer, j);
}
writer.close();
dir.resetMaxUsedSizeInBytes();
dir.setTrackDiskUsage(true);
long startDiskUsage = dir.getMaxUsedSizeInBytes();
writer =
new IndexWriter(
dir,
newIndexWriterConfig(TEST_VERSION_CURRENT, analyzer)
.setOpenMode(OpenMode.APPEND)
.setMaxBufferedDocs(10)
.setMergeScheduler(new SerialMergeScheduler())
.setReaderPooling(false)
.setMergePolicy(newLogMergePolicy(10)));
for (int j = 0; j < 1470; j++) {
TestIndexWriter.addDocWithIndex(writer, j);
}
long midDiskUsage = dir.getMaxUsedSizeInBytes();
dir.resetMaxUsedSizeInBytes();
writer.forceMerge(1);
writer.close();
DirectoryReader.open(dir).close();
long endDiskUsage = dir.getMaxUsedSizeInBytes();
// Ending index is 50X as large as starting index; due
// to 3X disk usage normally we allow 150X max
// transient usage. If something is wrong w/ deleter
// and it doesn't delete intermediate segments then it
// will exceed this 150X:
// System.out.println("start " + startDiskUsage + "; mid " + midDiskUsage + ";end " +
// endDiskUsage);
assertTrue(
"writer used too much space while adding documents: mid="
+ midDiskUsage
+ " start="
+ startDiskUsage
+ " end="
+ endDiskUsage
+ " max="
+ (startDiskUsage * 150),
midDiskUsage < 150 * startDiskUsage);
assertTrue(
"writer used too much space after close: endDiskUsage="
+ endDiskUsage
+ " startDiskUsage="
+ startDiskUsage
+ " max="
+ (startDiskUsage * 150),
endDiskUsage < 150 * startDiskUsage);
dir.close();
}