LUCENE-2739: refactor TestIndexWriter, pull out _OnDiskFull and _WithThreads

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1031076 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Robert Muir 2010-11-04 16:52:27 +00:00
parent 154105635d
commit ac4674aa20
3 changed files with 948 additions and 847 deletions

View File

@ -159,377 +159,7 @@ public class TestIndexWriter extends LuceneTestCase {
writer.addDocument(doc);
}
/*
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()));
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()));
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 = new IndexSearcher(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+100;
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) {
// Make a new dir that will enforce disk usage:
MockDirectoryWrapper dir = new MockDirectoryWrapper(new RAMDirectory(startDir));
writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer()).setOpenMode(OpenMode.APPEND));
IOException err = 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, diskFree);
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 = new IndexSearcher(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; " +
"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, 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 5000 more bytes of free space:
diskFree += 5000;
}
}
startDir.close();
for (Directory dir : dirs)
dir.close();
}
/*
* Make sure IndexWriter cleans up on hitting a disk
* full exception in addDocument.
* TODO: how to do this on windows with FSDirectory?
*/
public void testAddDocumentOnDiskFull() throws IOException {
for(int pass=0;pass<2;pass++) {
if (VERBOSE)
System.out.println("TEST: pass=" + pass);
boolean doAbort = pass == 1;
long diskFree = 200;
while(true) {
if (VERBOSE)
System.out.println("TEST: cycle: diskFree=" + diskFree);
MockDirectoryWrapper dir = new MockDirectoryWrapper(new RAMDirectory());
dir.setMaxSizeInBytes(diskFree);
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer()));
MergeScheduler ms = writer.getConfig().getMergeScheduler();
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.
((ConcurrentMergeScheduler) ms).setSuppressExceptions();
boolean hitError = false;
try {
for(int i=0;i<200;i++) {
addDoc(writer);
}
writer.commit();
} catch (IOException e) {
if (VERBOSE) {
System.out.println("TEST: exception on addDoc");
e.printStackTrace(System.out);
}
hitError = true;
}
if (hitError) {
if (doAbort) {
writer.rollback();
} else {
try {
writer.close();
} catch (IOException e) {
if (VERBOSE) {
System.out.println("TEST: exception on close");
e.printStackTrace(System.out);
}
dir.setMaxSizeInBytes(0);
writer.close();
}
}
//_TestUtil.syncConcurrentMerges(ms);
if (_TestUtil.anyFilesExceptWriteLock(dir)) {
assertNoUnreferencedFiles(dir, "after disk full during addDocument");
// Make sure reader can open the index:
IndexReader.open(dir, true).close();
}
dir.close();
// Now try again w/ more space:
diskFree += 500;
} else {
//_TestUtil.syncConcurrentMerges(writer);
writer.close();
dir.close();
break;
}
}
}
}
public static void assertNoUnreferencedFiles(Directory dir, String message) throws IOException {
String[] startFiles = dir.listAll();
@ -2190,421 +1820,6 @@ public class TestIndexWriter extends LuceneTestCase {
directory.close();
}
// Used by test cases below
private class IndexerThread extends Thread {
boolean diskFull;
Throwable error;
AlreadyClosedException ace;
IndexWriter writer;
boolean noErrors;
volatile int addCount;
public IndexerThread(IndexWriter writer, boolean noErrors) {
this.writer = writer;
this.noErrors = noErrors;
}
@Override
public void run() {
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
int idUpto = 0;
int fullCount = 0;
final long stopTime = System.currentTimeMillis() + 200;
do {
try {
writer.updateDocument(new Term("id", ""+(idUpto++)), doc);
addCount++;
} catch (IOException ioe) {
//System.out.println(Thread.currentThread().getName() + ": hit exc");
//ioe.printStackTrace(System.out);
if (ioe.getMessage().startsWith("fake disk full at") ||
ioe.getMessage().equals("now failing on purpose")) {
diskFull = true;
try {
Thread.sleep(1);
} catch (InterruptedException ie) {
throw new ThreadInterruptedException(ie);
}
if (fullCount++ >= 5)
break;
} else {
if (noErrors) {
System.out.println(Thread.currentThread().getName() + ": ERROR: unexpected IOException:");
ioe.printStackTrace(System.out);
error = ioe;
}
break;
}
} catch (Throwable t) {
//t.printStackTrace(System.out);
if (noErrors) {
System.out.println(Thread.currentThread().getName() + ": ERROR: unexpected Throwable:");
t.printStackTrace(System.out);
error = t;
}
break;
}
} while(System.currentTimeMillis() < stopTime);
}
}
// LUCENE-1130: make sure we can close() even while
// threads are trying to add documents. Strictly
// speaking, this isn't valid us of Lucene's APIs, but we
// still want to be robust to this case:
public void testCloseWithThreads() throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<7;iter++) {
Directory dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(10).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect AlreadyClosedException
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, false);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
boolean done = false;
while(!done) {
Thread.sleep(100);
for(int i=0;i<NUM_THREADS;i++)
// only stop when at least one thread has added a doc
if (threads[i].addCount > 0) {
done = true;
break;
}
}
writer.close(false);
// Make sure threads that are adding docs are not hung:
for(int i=0;i<NUM_THREADS;i++) {
// Without fix for LUCENE-1130: one of the
// threads will hang
threads[i].join();
if (threads[i].isAlive())
fail("thread seems to be hung");
}
// Quick test to make sure index is not corrupt:
IndexReader reader = IndexReader.open(dir, true);
DocsEnum tdocs = MultiFields.getTermDocsEnum(reader,
MultiFields.getDeletedDocs(reader),
"field",
new BytesRef("aaa"));
int count = 0;
while(tdocs.nextDoc() != DocsEnum.NO_MORE_DOCS) {
count++;
}
assertTrue(count > 0);
reader.close();
dir.close();
}
}
// LUCENE-1130: make sure immeidate disk full on creating
// an IndexWriter (hit during DW.ThreadState.init()) is
// OK:
public void testImmediateDiskFull() throws IOException {
MockDirectoryWrapper dir = newDirectory();
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler()));
dir.setMaxSizeInBytes(Math.max(1, dir.getRecomputedActualSizeInBytes()));
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
try {
writer.addDocument(doc);
fail("did not hit disk full");
} catch (IOException ioe) {
}
// Without fix for LUCENE-1130: this call will hang:
try {
writer.addDocument(doc);
fail("did not hit disk full");
} catch (IOException ioe) {
}
try {
writer.close(false);
fail("did not hit disk full");
} catch (IOException ioe) {
}
// Make sure once disk space is avail again, we can
// cleanly close:
dir.setMaxSizeInBytes(0);
writer.close(false);
dir.close();
}
// LUCENE-1130: make sure immediate disk full on creating
// an IndexWriter (hit during DW.ThreadState.init()), with
// multiple threads, is OK:
public void testImmediateDiskFullWithThreads() throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<10;iter++) {
MockDirectoryWrapper dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect disk full exceptions in the merge threads
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
dir.setMaxSizeInBytes(4*1024+20*iter);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, true);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
for(int i=0;i<NUM_THREADS;i++) {
// Without fix for LUCENE-1130: one of the
// threads will hang
threads[i].join();
assertTrue("hit unexpected Throwable", threads[i].error == null);
}
// Make sure once disk space is avail again, we can
// cleanly close:
dir.setMaxSizeInBytes(0);
writer.close(false);
dir.close();
}
}
// Throws IOException during FieldsWriter.flushDocument and during DocumentsWriter.abort
private static class FailOnlyOnAbortOrFlush extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyOnAbortOrFlush(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("abort".equals(trace[i].getMethodName()) ||
"flushDocument".equals(trace[i].getMethodName())) {
if (onlyOnce)
doFail = false;
//System.out.println(Thread.currentThread().getName() + ": now fail");
//new Throwable().printStackTrace(System.out);
throw new IOException("now failing on purpose");
}
}
}
}
}
// Runs test, with one thread, using the specific failure
// to trigger an IOException
public void _testSingleThreadFailure(MockDirectoryWrapper.Failure failure) throws IOException {
MockDirectoryWrapper dir = newDirectory();
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler()));
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
for(int i=0;i<6;i++)
writer.addDocument(doc);
dir.failOn(failure);
failure.setDoFail();
try {
writer.addDocument(doc);
writer.addDocument(doc);
writer.commit();
fail("did not hit exception");
} catch (IOException ioe) {
}
failure.clearDoFail();
writer.addDocument(doc);
writer.close(false);
dir.close();
}
// Runs test, with multiple threads, using the specific
// failure to trigger an IOException
public void _testMultipleThreadsFailure(MockDirectoryWrapper.Failure failure) throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<2;iter++) {
MockDirectoryWrapper dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT,
new MockAnalyzer()).setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect disk full exceptions in the merge threads
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, true);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
Thread.sleep(10);
dir.failOn(failure);
failure.setDoFail();
for(int i=0;i<NUM_THREADS;i++) {
threads[i].join();
assertTrue("hit unexpected Throwable", threads[i].error == null);
}
boolean success = false;
try {
writer.close(false);
success = true;
} catch (IOException ioe) {
failure.clearDoFail();
writer.close(false);
}
if (success) {
IndexReader reader = IndexReader.open(dir, true);
final Bits delDocs = MultiFields.getDeletedDocs(reader);
for(int j=0;j<reader.maxDoc();j++) {
if (delDocs == null || !delDocs.get(j)) {
reader.document(j);
reader.getTermFreqVectors(j);
}
}
reader.close();
}
dir.close();
}
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), is OK:
public void testIOExceptionDuringAbort() throws IOException {
_testSingleThreadFailure(new FailOnlyOnAbortOrFlush(false));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), is OK:
public void testIOExceptionDuringAbortOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyOnAbortOrFlush(true));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), with multiple threads, is OK:
public void testIOExceptionDuringAbortWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyOnAbortOrFlush(false));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), with multiple threads, is OK:
public void testIOExceptionDuringAbortWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyOnAbortOrFlush(true));
}
// Throws IOException during DocumentsWriter.closeDocStore
private static class FailOnlyInCloseDocStore extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyInCloseDocStore(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("closeDocStore".equals(trace[i].getMethodName())) {
if (onlyOnce)
doFail = false;
throw new IOException("now failing on purpose");
}
}
}
}
}
// LUCENE-1130: test IOException in closeDocStore
public void testIOExceptionDuringCloseDocStore() throws IOException {
_testSingleThreadFailure(new FailOnlyInCloseDocStore(false));
}
// LUCENE-1130: test IOException in closeDocStore
public void testIOExceptionDuringCloseDocStoreOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyInCloseDocStore(true));
}
// LUCENE-1130: test IOException in closeDocStore, with threads
public void testIOExceptionDuringCloseDocStoreWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInCloseDocStore(false));
}
// LUCENE-1130: test IOException in closeDocStore, with threads
public void testIOExceptionDuringCloseDocStoreWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInCloseDocStore(true));
}
// Throws IOException during DocumentsWriter.writeSegment
private static class FailOnlyInWriteSegment extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyInWriteSegment(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("flush".equals(trace[i].getMethodName()) && "org.apache.lucene.index.DocFieldProcessor".equals(trace[i].getClassName())) {
if (onlyOnce)
doFail = false;
throw new IOException("now failing on purpose");
}
}
}
}
}
// LUCENE-1130: test IOException in writeSegment
public void testIOExceptionDuringWriteSegment() throws IOException {
_testSingleThreadFailure(new FailOnlyInWriteSegment(false));
}
// LUCENE-1130: test IOException in writeSegment
public void testIOExceptionDuringWriteSegmentOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyInWriteSegment(true));
}
// LUCENE-1130: test IOException in writeSegment, with threads
public void testIOExceptionDuringWriteSegmentWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInWriteSegment(false));
}
// LUCENE-1130: test IOException in writeSegment, with threads
public void testIOExceptionDuringWriteSegmentWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInWriteSegment(true));
}
// LUCENE-1084: test unlimited field length
public void testUnlimitedMaxFieldLength() throws IOException {
@ -5250,65 +4465,4 @@ public class TestIndexWriter extends LuceneTestCase {
w.close();
dir.close();
}
private static class FailTwiceDuringMerge extends MockDirectoryWrapper.Failure {
public boolean didFail1;
public boolean didFail2;
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (!doFail) {
return;
}
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("org.apache.lucene.index.SegmentMerger".equals(trace[i].getClassName()) && "mergeTerms".equals(trace[i].getMethodName()) && !didFail1) {
didFail1 = true;
throw new IOException("fake disk full during mergeTerms");
}
if ("org.apache.lucene.util.BitVector".equals(trace[i].getClassName()) && "write".equals(trace[i].getMethodName()) && !didFail2) {
didFail2 = true;
throw new IOException("fake disk full while writing BitVector");
}
}
}
}
// LUCENE-2593
public void testCorruptionAfterDiskFullDuringMerge() throws IOException {
MockDirectoryWrapper dir = newDirectory();
//IndexWriter w = new IndexWriter(dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer()).setReaderPooling(true));
IndexWriter w = new IndexWriter(dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer()).setMergeScheduler(new SerialMergeScheduler()).setReaderPooling(true));
((LogMergePolicy) w.getMergePolicy()).setMergeFactor(2);
Document doc = new Document();
doc.add(newField("f", "doctor who", Field.Store.YES, Field.Index.ANALYZED));
w.addDocument(doc);
w.commit();
w.deleteDocuments(new Term("f", "who"));
w.addDocument(doc);
// disk fills up!
FailTwiceDuringMerge ftdm = new FailTwiceDuringMerge();
ftdm.setDoFail();
dir.failOn(ftdm);
try {
w.commit();
fail("fake disk full IOExceptions not hit");
} catch (IOException ioe) {
// expected
assertTrue(ftdm.didFail1);
}
_TestUtil.checkIndex(dir);
ftdm.clearDoFail();
w.addDocument(doc);
w.close();
_TestUtil.checkIndex(dir);
dir.close();
}
}

View File

@ -0,0 +1,525 @@
package org.apache.lucene.index;
/**
* 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.
*/
import java.io.IOException;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.index.IndexWriterConfig.OpenMode;
import org.apache.lucene.search.IndexSearcher;
import org.apache.lucene.search.ScoreDoc;
import org.apache.lucene.search.TermQuery;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.MockDirectoryWrapper;
import org.apache.lucene.store.RAMDirectory;
import org.apache.lucene.util.LuceneTestCase;
import org.apache.lucene.util._TestUtil;
import static org.apache.lucene.index.TestIndexWriter.assertNoUnreferencedFiles;
/**
* Tests for IndexWriter when the disk runs out of space
*/
public class TestIndexWriterOnDiskFull extends LuceneTestCase {
/*
* Make sure IndexWriter cleans up on hitting a disk
* full exception in addDocument.
* TODO: how to do this on windows with FSDirectory?
*/
public void testAddDocumentOnDiskFull() throws IOException {
for(int pass=0;pass<2;pass++) {
if (VERBOSE)
System.out.println("TEST: pass=" + pass);
boolean doAbort = pass == 1;
long diskFree = 200;
while(true) {
if (VERBOSE)
System.out.println("TEST: cycle: diskFree=" + diskFree);
MockDirectoryWrapper dir = new MockDirectoryWrapper(new RAMDirectory());
dir.setMaxSizeInBytes(diskFree);
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer()));
MergeScheduler ms = writer.getConfig().getMergeScheduler();
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.
((ConcurrentMergeScheduler) ms).setSuppressExceptions();
boolean hitError = false;
try {
for(int i=0;i<200;i++) {
addDoc(writer);
}
writer.commit();
} catch (IOException e) {
if (VERBOSE) {
System.out.println("TEST: exception on addDoc");
e.printStackTrace(System.out);
}
hitError = true;
}
if (hitError) {
if (doAbort) {
writer.rollback();
} else {
try {
writer.close();
} catch (IOException e) {
if (VERBOSE) {
System.out.println("TEST: exception on close");
e.printStackTrace(System.out);
}
dir.setMaxSizeInBytes(0);
writer.close();
}
}
//_TestUtil.syncConcurrentMerges(ms);
if (_TestUtil.anyFilesExceptWriteLock(dir)) {
assertNoUnreferencedFiles(dir, "after disk full during addDocument");
// Make sure reader can open the index:
IndexReader.open(dir, true).close();
}
dir.close();
// Now try again w/ more space:
diskFree += 500;
} else {
//_TestUtil.syncConcurrentMerges(writer);
writer.close();
dir.close();
break;
}
}
}
}
/*
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()));
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()));
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 = new IndexSearcher(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+100;
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) {
// Make a new dir that will enforce disk usage:
MockDirectoryWrapper dir = new MockDirectoryWrapper(new RAMDirectory(startDir));
writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer()).setOpenMode(OpenMode.APPEND));
IOException err = 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, diskFree);
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 = new IndexSearcher(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; " +
"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, 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 5000 more bytes of free space:
diskFree += 5000;
}
}
startDir.close();
for (Directory dir : dirs)
dir.close();
}
private static class FailTwiceDuringMerge extends MockDirectoryWrapper.Failure {
public boolean didFail1;
public boolean didFail2;
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (!doFail) {
return;
}
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("org.apache.lucene.index.SegmentMerger".equals(trace[i].getClassName()) && "mergeTerms".equals(trace[i].getMethodName()) && !didFail1) {
didFail1 = true;
throw new IOException("fake disk full during mergeTerms");
}
if ("org.apache.lucene.util.BitVector".equals(trace[i].getClassName()) && "write".equals(trace[i].getMethodName()) && !didFail2) {
didFail2 = true;
throw new IOException("fake disk full while writing BitVector");
}
}
}
}
// LUCENE-2593
public void testCorruptionAfterDiskFullDuringMerge() throws IOException {
MockDirectoryWrapper dir = newDirectory();
//IndexWriter w = new IndexWriter(dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer()).setReaderPooling(true));
IndexWriter w = new IndexWriter(dir, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer()).setMergeScheduler(new SerialMergeScheduler()).setReaderPooling(true));
((LogMergePolicy) w.getMergePolicy()).setMergeFactor(2);
Document doc = new Document();
doc.add(newField("f", "doctor who", Field.Store.YES, Field.Index.ANALYZED));
w.addDocument(doc);
w.commit();
w.deleteDocuments(new Term("f", "who"));
w.addDocument(doc);
// disk fills up!
FailTwiceDuringMerge ftdm = new FailTwiceDuringMerge();
ftdm.setDoFail();
dir.failOn(ftdm);
try {
w.commit();
fail("fake disk full IOExceptions not hit");
} catch (IOException ioe) {
// expected
assertTrue(ftdm.didFail1);
}
_TestUtil.checkIndex(dir);
ftdm.clearDoFail();
w.addDocument(doc);
w.close();
_TestUtil.checkIndex(dir);
dir.close();
}
// LUCENE-1130: make sure immeidate disk full on creating
// an IndexWriter (hit during DW.ThreadState.init()) is
// OK:
public void testImmediateDiskFull() throws IOException {
MockDirectoryWrapper dir = newDirectory();
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler()));
dir.setMaxSizeInBytes(Math.max(1, dir.getRecomputedActualSizeInBytes()));
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
try {
writer.addDocument(doc);
fail("did not hit disk full");
} catch (IOException ioe) {
}
// Without fix for LUCENE-1130: this call will hang:
try {
writer.addDocument(doc);
fail("did not hit disk full");
} catch (IOException ioe) {
}
try {
writer.close(false);
fail("did not hit disk full");
} catch (IOException ioe) {
}
// Make sure once disk space is avail again, we can
// cleanly close:
dir.setMaxSizeInBytes(0);
writer.close(false);
dir.close();
}
// TODO: these are also in TestIndexWriter... add a simple doc-writing method
// like this to LuceneTestCase?
private void addDoc(IndexWriter writer) throws IOException
{
Document doc = new Document();
doc.add(newField("content", "aaa", Field.Store.NO, Field.Index.ANALYZED));
writer.addDocument(doc);
}
private void addDocWithIndex(IndexWriter writer, int index) throws IOException
{
Document doc = new Document();
doc.add(newField("content", "aaa " + index, Field.Store.YES, Field.Index.ANALYZED));
doc.add(newField("id", "" + index, Field.Store.YES, Field.Index.ANALYZED));
writer.addDocument(doc);
}
}

View File

@ -0,0 +1,422 @@
package org.apache.lucene.index;
/**
* 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.
*/
import java.io.IOException;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.store.AlreadyClosedException;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.MockDirectoryWrapper;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.LuceneTestCase;
import org.apache.lucene.util.ThreadInterruptedException;
/**
* MultiThreaded IndexWriter tests
*/
public class TestIndexWriterWithThreads extends LuceneTestCase {
// Used by test cases below
private class IndexerThread extends Thread {
boolean diskFull;
Throwable error;
AlreadyClosedException ace;
IndexWriter writer;
boolean noErrors;
volatile int addCount;
public IndexerThread(IndexWriter writer, boolean noErrors) {
this.writer = writer;
this.noErrors = noErrors;
}
@Override
public void run() {
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
int idUpto = 0;
int fullCount = 0;
final long stopTime = System.currentTimeMillis() + 200;
do {
try {
writer.updateDocument(new Term("id", ""+(idUpto++)), doc);
addCount++;
} catch (IOException ioe) {
//System.out.println(Thread.currentThread().getName() + ": hit exc");
//ioe.printStackTrace(System.out);
if (ioe.getMessage().startsWith("fake disk full at") ||
ioe.getMessage().equals("now failing on purpose")) {
diskFull = true;
try {
Thread.sleep(1);
} catch (InterruptedException ie) {
throw new ThreadInterruptedException(ie);
}
if (fullCount++ >= 5)
break;
} else {
if (noErrors) {
System.out.println(Thread.currentThread().getName() + ": ERROR: unexpected IOException:");
ioe.printStackTrace(System.out);
error = ioe;
}
break;
}
} catch (Throwable t) {
//t.printStackTrace(System.out);
if (noErrors) {
System.out.println(Thread.currentThread().getName() + ": ERROR: unexpected Throwable:");
t.printStackTrace(System.out);
error = t;
}
break;
}
} while(System.currentTimeMillis() < stopTime);
}
}
// LUCENE-1130: make sure immediate disk full on creating
// an IndexWriter (hit during DW.ThreadState.init()), with
// multiple threads, is OK:
public void testImmediateDiskFullWithThreads() throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<10;iter++) {
MockDirectoryWrapper dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect disk full exceptions in the merge threads
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
dir.setMaxSizeInBytes(4*1024+20*iter);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, true);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
for(int i=0;i<NUM_THREADS;i++) {
// Without fix for LUCENE-1130: one of the
// threads will hang
threads[i].join();
assertTrue("hit unexpected Throwable", threads[i].error == null);
}
// Make sure once disk space is avail again, we can
// cleanly close:
dir.setMaxSizeInBytes(0);
writer.close(false);
dir.close();
}
}
// LUCENE-1130: make sure we can close() even while
// threads are trying to add documents. Strictly
// speaking, this isn't valid us of Lucene's APIs, but we
// still want to be robust to this case:
public void testCloseWithThreads() throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<7;iter++) {
Directory dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(10).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect AlreadyClosedException
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, false);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
boolean done = false;
while(!done) {
Thread.sleep(100);
for(int i=0;i<NUM_THREADS;i++)
// only stop when at least one thread has added a doc
if (threads[i].addCount > 0) {
done = true;
break;
}
}
writer.close(false);
// Make sure threads that are adding docs are not hung:
for(int i=0;i<NUM_THREADS;i++) {
// Without fix for LUCENE-1130: one of the
// threads will hang
threads[i].join();
if (threads[i].isAlive())
fail("thread seems to be hung");
}
// Quick test to make sure index is not corrupt:
IndexReader reader = IndexReader.open(dir, true);
DocsEnum tdocs = MultiFields.getTermDocsEnum(reader,
MultiFields.getDeletedDocs(reader),
"field",
new BytesRef("aaa"));
int count = 0;
while(tdocs.nextDoc() != DocsEnum.NO_MORE_DOCS) {
count++;
}
assertTrue(count > 0);
reader.close();
dir.close();
}
}
// Runs test, with multiple threads, using the specific
// failure to trigger an IOException
public void _testMultipleThreadsFailure(MockDirectoryWrapper.Failure failure) throws Exception {
int NUM_THREADS = 3;
for(int iter=0;iter<2;iter++) {
MockDirectoryWrapper dir = newDirectory();
IndexWriterConfig conf = newIndexWriterConfig( TEST_VERSION_CURRENT,
new MockAnalyzer()).setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler());
// We expect disk full exceptions in the merge threads
((ConcurrentMergeScheduler) conf.getMergeScheduler()).setSuppressExceptions();
IndexWriter writer = new IndexWriter(dir, conf);
((LogMergePolicy) writer.getConfig().getMergePolicy()).setMergeFactor(4);
IndexerThread[] threads = new IndexerThread[NUM_THREADS];
for(int i=0;i<NUM_THREADS;i++)
threads[i] = new IndexerThread(writer, true);
for(int i=0;i<NUM_THREADS;i++)
threads[i].start();
Thread.sleep(10);
dir.failOn(failure);
failure.setDoFail();
for(int i=0;i<NUM_THREADS;i++) {
threads[i].join();
assertTrue("hit unexpected Throwable", threads[i].error == null);
}
boolean success = false;
try {
writer.close(false);
success = true;
} catch (IOException ioe) {
failure.clearDoFail();
writer.close(false);
}
if (success) {
IndexReader reader = IndexReader.open(dir, true);
final Bits delDocs = MultiFields.getDeletedDocs(reader);
for(int j=0;j<reader.maxDoc();j++) {
if (delDocs == null || !delDocs.get(j)) {
reader.document(j);
reader.getTermFreqVectors(j);
}
}
reader.close();
}
dir.close();
}
}
// Runs test, with one thread, using the specific failure
// to trigger an IOException
public void _testSingleThreadFailure(MockDirectoryWrapper.Failure failure) throws IOException {
MockDirectoryWrapper dir = newDirectory();
IndexWriter writer = new IndexWriter(dir, newIndexWriterConfig( TEST_VERSION_CURRENT, new MockAnalyzer())
.setMaxBufferedDocs(2).setMergeScheduler(new ConcurrentMergeScheduler()));
final Document doc = new Document();
doc.add(newField("field", "aaa bbb ccc ddd eee fff ggg hhh iii jjj", Field.Store.YES, Field.Index.ANALYZED, Field.TermVector.WITH_POSITIONS_OFFSETS));
for(int i=0;i<6;i++)
writer.addDocument(doc);
dir.failOn(failure);
failure.setDoFail();
try {
writer.addDocument(doc);
writer.addDocument(doc);
writer.commit();
fail("did not hit exception");
} catch (IOException ioe) {
}
failure.clearDoFail();
writer.addDocument(doc);
writer.close(false);
dir.close();
}
// Throws IOException during FieldsWriter.flushDocument and during DocumentsWriter.abort
private static class FailOnlyOnAbortOrFlush extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyOnAbortOrFlush(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("abort".equals(trace[i].getMethodName()) ||
"flushDocument".equals(trace[i].getMethodName())) {
if (onlyOnce)
doFail = false;
//System.out.println(Thread.currentThread().getName() + ": now fail");
//new Throwable().printStackTrace(System.out);
throw new IOException("now failing on purpose");
}
}
}
}
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), is OK:
public void testIOExceptionDuringAbort() throws IOException {
_testSingleThreadFailure(new FailOnlyOnAbortOrFlush(false));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), is OK:
public void testIOExceptionDuringAbortOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyOnAbortOrFlush(true));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), with multiple threads, is OK:
public void testIOExceptionDuringAbortWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyOnAbortOrFlush(false));
}
// LUCENE-1130: make sure initial IOException, and then 2nd
// IOException during rollback(), with multiple threads, is OK:
public void testIOExceptionDuringAbortWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyOnAbortOrFlush(true));
}
// Throws IOException during DocumentsWriter.closeDocStore
private static class FailOnlyInCloseDocStore extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyInCloseDocStore(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("closeDocStore".equals(trace[i].getMethodName())) {
if (onlyOnce)
doFail = false;
throw new IOException("now failing on purpose");
}
}
}
}
}
// LUCENE-1130: test IOException in closeDocStore
public void testIOExceptionDuringCloseDocStore() throws IOException {
_testSingleThreadFailure(new FailOnlyInCloseDocStore(false));
}
// LUCENE-1130: test IOException in closeDocStore
public void testIOExceptionDuringCloseDocStoreOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyInCloseDocStore(true));
}
// LUCENE-1130: test IOException in closeDocStore, with threads
public void testIOExceptionDuringCloseDocStoreWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInCloseDocStore(false));
}
// LUCENE-1130: test IOException in closeDocStore, with threads
public void testIOExceptionDuringCloseDocStoreWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInCloseDocStore(true));
}
// Throws IOException during DocumentsWriter.writeSegment
private static class FailOnlyInWriteSegment extends MockDirectoryWrapper.Failure {
private boolean onlyOnce;
public FailOnlyInWriteSegment(boolean onlyOnce) {
this.onlyOnce = onlyOnce;
}
@Override
public void eval(MockDirectoryWrapper dir) throws IOException {
if (doFail) {
StackTraceElement[] trace = new Exception().getStackTrace();
for (int i = 0; i < trace.length; i++) {
if ("flush".equals(trace[i].getMethodName()) && "org.apache.lucene.index.DocFieldProcessor".equals(trace[i].getClassName())) {
if (onlyOnce)
doFail = false;
throw new IOException("now failing on purpose");
}
}
}
}
}
// LUCENE-1130: test IOException in writeSegment
public void testIOExceptionDuringWriteSegment() throws IOException {
_testSingleThreadFailure(new FailOnlyInWriteSegment(false));
}
// LUCENE-1130: test IOException in writeSegment
public void testIOExceptionDuringWriteSegmentOnlyOnce() throws IOException {
_testSingleThreadFailure(new FailOnlyInWriteSegment(true));
}
// LUCENE-1130: test IOException in writeSegment, with threads
public void testIOExceptionDuringWriteSegmentWithThreads() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInWriteSegment(false));
}
// LUCENE-1130: test IOException in writeSegment, with threads
public void testIOExceptionDuringWriteSegmentWithThreadsOnlyOnce() throws Exception {
_testMultipleThreadsFailure(new FailOnlyInWriteSegment(true));
}
}