Examples with MMapDirectory org.apache.lucene.store.MMapDirectory used on opensource projects

Example 6 with MMapDirectory

use of org.apache.lucene.store.MMapDirectory in project lucene-solr by apache.

the class BaseStoredFieldsFormatTestCase method testBigDocuments.

@Nightly
public void testBigDocuments() throws IOException {
    assumeWorkingMMapOnWindows();
    // "big" as "much bigger than the chunk size"
    // for this test we force a FS dir
    // we can't just use newFSDirectory, because this test doesn't really index anything.
    // so if we get NRTCachingDir+SimpleText, we make massive stored fields and OOM (LUCENE-4484)
    Directory dir = new MockDirectoryWrapper(random(), new MMapDirectory(createTempDir("testBigDocuments")));
    IndexWriterConfig iwConf = newIndexWriterConfig(new MockAnalyzer(random()));
    iwConf.setMaxBufferedDocs(RandomNumbers.randomIntBetween(random(), 2, 30));
    RandomIndexWriter iw = new RandomIndexWriter(random(), dir, iwConf);
    if (dir instanceof MockDirectoryWrapper) {
        ((MockDirectoryWrapper) dir).setThrottling(Throttling.NEVER);
    }
    // emptyDoc
    final Document emptyDoc = new Document();
    // lot of small fields
    final Document bigDoc1 = new Document();
    // 1 very big field
    final Document bigDoc2 = new Document();
    final Field idField = new StringField("id", "", Store.NO);
    emptyDoc.add(idField);
    bigDoc1.add(idField);
    bigDoc2.add(idField);
    final FieldType onlyStored = new FieldType(StringField.TYPE_STORED);
    onlyStored.setIndexOptions(IndexOptions.NONE);
    final Field smallField = new Field("fld", randomByteArray(random().nextInt(10), 256), onlyStored);
    final int numFields = RandomNumbers.randomIntBetween(random(), 500000, 1000000);
    for (int i = 0; i < numFields; ++i) {
        bigDoc1.add(smallField);
    }
    final Field bigField = new Field("fld", randomByteArray(RandomNumbers.randomIntBetween(random(), 1000000, 5000000), 2), onlyStored);
    bigDoc2.add(bigField);
    final int numDocs = atLeast(5);
    final Document[] docs = new Document[numDocs];
    for (int i = 0; i < numDocs; ++i) {
        docs[i] = RandomPicks.randomFrom(random(), Arrays.asList(emptyDoc, bigDoc1, bigDoc2));
    }
    for (int i = 0; i < numDocs; ++i) {
        idField.setStringValue("" + i);
        iw.addDocument(docs[i]);
        if (random().nextInt(numDocs) == 0) {
            iw.commit();
        }
    }
    iw.commit();
    // look at what happens when big docs are merged
    iw.forceMerge(1);
    final DirectoryReader rd = DirectoryReader.open(dir);
    final IndexSearcher searcher = new IndexSearcher(rd);
    for (int i = 0; i < numDocs; ++i) {
        final Query query = new TermQuery(new Term("id", "" + i));
        final TopDocs topDocs = searcher.search(query, 1);
        assertEquals("" + i, 1, topDocs.totalHits);
        final Document doc = rd.document(topDocs.scoreDocs[0].doc);
        assertNotNull(doc);
        final IndexableField[] fieldValues = doc.getFields("fld");
        assertEquals(docs[i].getFields("fld").length, fieldValues.length);
        if (fieldValues.length > 0) {
            assertEquals(docs[i].getFields("fld")[0].binaryValue(), fieldValues[0].binaryValue());
        }
    }
    rd.close();
    iw.close();
    dir.close();
}

Example 7 with MMapDirectory

use of org.apache.lucene.store.MMapDirectory in project Anserini by castorini.

the class SearchTweets method main.

public static void main(String[] args) throws Exception {
    long curTime = System.nanoTime();
    SearchArgs searchArgs = new SearchArgs();
    CmdLineParser parser = new CmdLineParser(searchArgs, ParserProperties.defaults().withUsageWidth(90));
    try {
        parser.parseArgument(args);
    } catch (CmdLineException e) {
        System.err.println(e.getMessage());
        parser.printUsage(System.err);
        System.err.println("Example: SearchTweets" + parser.printExample(OptionHandlerFilter.REQUIRED));
        return;
    }
    LOG.info("Reading index at " + searchArgs.index);
    Directory dir;
    if (searchArgs.inmem) {
        LOG.info("Using MMapDirectory with preload");
        dir = new MMapDirectory(Paths.get(searchArgs.index));
        ((MMapDirectory) dir).setPreload(true);
    } else {
        LOG.info("Using default FSDirectory");
        dir = FSDirectory.open(Paths.get(searchArgs.index));
    }
    IndexReader reader = DirectoryReader.open(dir);
    IndexSearcher searcher = new IndexSearcher(reader);
    if (searchArgs.ql) {
        LOG.info("Using QL scoring model");
        searcher.setSimilarity(new LMDirichletSimilarity(searchArgs.mu));
    } else if (searchArgs.bm25) {
        LOG.info("Using BM25 scoring model");
        searcher.setSimilarity(new BM25Similarity(searchArgs.k1, searchArgs.b));
    } else {
        LOG.error("Error: Must specify scoring model!");
        System.exit(-1);
    }
    RerankerCascade cascade = new RerankerCascade();
    if (searchArgs.rm3) {
        cascade.add(new Rm3Reranker(IndexTweets.ANALYZER, StatusField.TEXT.name, "src/main/resources/io/anserini/rerank/rm3/rm3-stoplist.twitter.txt"));
        cascade.add(new RemoveRetweetsTemporalTiebreakReranker());
    } else {
        cascade.add(new RemoveRetweetsTemporalTiebreakReranker());
    }
    if (!searchArgs.model.isEmpty() && searchArgs.extractors != null) {
        LOG.debug(String.format("Ranklib model used, modeled loaded from %s", searchArgs.model));
        cascade.add(new RankLibReranker(searchArgs.model, StatusField.TEXT.name, searchArgs.extractors));
    }
    FeatureExtractors extractorChain = null;
    if (searchArgs.extractors != null) {
        extractorChain = FeatureExtractors.loadExtractor(searchArgs.extractors);
    }
    if (searchArgs.dumpFeatures) {
        PrintStream out = new PrintStream(searchArgs.featureFile);
        Qrels qrels = new Qrels(searchArgs.qrels);
        cascade.add(new TweetsLtrDataGenerator(out, qrels, extractorChain));
    }
    MicroblogTopicSet topics = MicroblogTopicSet.fromFile(new File(searchArgs.topics));
    PrintStream out = new PrintStream(new FileOutputStream(new File(searchArgs.output)));
    LOG.info("Writing output to " + searchArgs.output);
    LOG.info("Initialized complete! (elapsed time = " + (System.nanoTime() - curTime) / 1000000 + "ms)");
    long totalTime = 0;
    int cnt = 0;
    for (MicroblogTopic topic : topics) {
        long curQueryTime = System.nanoTime();
        Query filter = LongPoint.newRangeQuery(StatusField.ID.name, 0L, topic.getQueryTweetTime());
        Query query = AnalyzerUtils.buildBagOfWordsQuery(StatusField.TEXT.name, IndexTweets.ANALYZER, topic.getQuery());
        BooleanQuery.Builder builder = new BooleanQuery.Builder();
        builder.add(filter, BooleanClause.Occur.FILTER);
        builder.add(query, BooleanClause.Occur.MUST);
        Query q = builder.build();
        TopDocs rs = searcher.search(q, searchArgs.hits);
        List<String> queryTokens = AnalyzerUtils.tokenize(IndexTweets.ANALYZER, topic.getQuery());
        RerankerContext context = new RerankerContext(searcher, query, topic.getId(), topic.getQuery(), queryTokens, StatusField.TEXT.name, filter);
        ScoredDocuments docs = cascade.run(ScoredDocuments.fromTopDocs(rs, searcher), context);
        for (int i = 0; i < docs.documents.length; i++) {
            String qid = topic.getId().replaceFirst("^MB0*", "");
            out.println(String.format("%s Q0 %s %d %f %s", qid, docs.documents[i].getField(StatusField.ID.name).numericValue(), (i + 1), docs.scores[i], searchArgs.runtag));
        }
        long qtime = (System.nanoTime() - curQueryTime) / 1000000;
        LOG.info("Query " + topic.getId() + " (elapsed time = " + qtime + "ms)");
        totalTime += qtime;
        cnt++;
    }
    LOG.info("All queries completed!");
    LOG.info("Total elapsed time = " + totalTime + "ms");
    LOG.info("Average query latency = " + (totalTime / cnt) + "ms");
    reader.close();
    out.close();
}

Example 8 with MMapDirectory

use of org.apache.lucene.store.MMapDirectory in project Anserini by castorini.

the class SearchWebCollection method main.

public static void main(String[] args) throws Exception {
    SearchArgs searchArgs = new SearchArgs();
    CmdLineParser parser = new CmdLineParser(searchArgs, ParserProperties.defaults().withUsageWidth(90));
    try {
        parser.parseArgument(args);
    } catch (CmdLineException e) {
        System.err.println(e.getMessage());
        parser.printUsage(System.err);
        System.err.println("Example: SearchWebCollection" + parser.printExample(OptionHandlerFilter.REQUIRED));
        return;
    }
    LOG.info("Reading index at " + searchArgs.index);
    Directory dir;
    if (searchArgs.inmem) {
        LOG.info("Using MMapDirectory with preload");
        dir = new MMapDirectory(Paths.get(searchArgs.index));
        ((MMapDirectory) dir).setPreload(true);
    } else {
        LOG.info("Using default FSDirectory");
        dir = FSDirectory.open(Paths.get(searchArgs.index));
    }
    Similarity similarity = null;
    if (searchArgs.ql) {
        LOG.info("Using QL scoring model");
        similarity = new LMDirichletSimilarity(searchArgs.mu);
    } else if (searchArgs.bm25) {
        LOG.info("Using BM25 scoring model");
        similarity = new BM25Similarity(searchArgs.k1, searchArgs.b);
    } else {
        LOG.error("Error: Must specify scoring model!");
        System.exit(-1);
    }
    RerankerCascade cascade = new RerankerCascade();
    boolean useQueryParser = false;
    if (searchArgs.rm3) {
        cascade.add(new Rm3Reranker(new EnglishAnalyzer(), FIELD_BODY, "src/main/resources/io/anserini/rerank/rm3/rm3-stoplist.gov2.txt"));
        useQueryParser = true;
    } else {
        cascade.add(new IdentityReranker());
    }
    FeatureExtractors extractors = null;
    if (searchArgs.extractors != null) {
        extractors = FeatureExtractors.loadExtractor(searchArgs.extractors);
    }
    if (searchArgs.dumpFeatures) {
        PrintStream out = new PrintStream(searchArgs.featureFile);
        Qrels qrels = new Qrels(searchArgs.qrels);
        cascade.add(new WebCollectionLtrDataGenerator(out, qrels, extractors));
    }
    Path topicsFile = Paths.get(searchArgs.topics);
    if (!Files.exists(topicsFile) || !Files.isRegularFile(topicsFile) || !Files.isReadable(topicsFile)) {
        throw new IllegalArgumentException("Topics file : " + topicsFile + " does not exist or is not a (readable) file.");
    }
    TopicReader tr = (TopicReader) Class.forName("io.anserini.search.query." + searchArgs.topicReader + "TopicReader").getConstructor(Path.class).newInstance(topicsFile);
    SortedMap<Integer, String> topics = tr.read();
    final long start = System.nanoTime();
    SearchWebCollection searcher = new SearchWebCollection(searchArgs.index);
    searcher.search(topics, searchArgs.output, similarity, searchArgs.hits, cascade, useQueryParser, searchArgs.keepstop);
    searcher.close();
    final long durationMillis = TimeUnit.MILLISECONDS.convert(System.nanoTime() - start, TimeUnit.NANOSECONDS);
    LOG.info("Total " + topics.size() + " topics searched in " + DurationFormatUtils.formatDuration(durationMillis, "HH:mm:ss"));
}

Example 9 with MMapDirectory

use of org.apache.lucene.store.MMapDirectory in project elasticsearch by elastic.

the class FsDirectoryServiceTests method doTestPreload.

private void doTestPreload(String... preload) throws IOException {
    Settings build = Settings.builder().put(IndexModule.INDEX_STORE_TYPE_SETTING.getKey(), "mmapfs").putArray(IndexModule.INDEX_STORE_PRE_LOAD_SETTING.getKey(), preload).build();
    IndexSettings settings = IndexSettingsModule.newIndexSettings("foo", build);
    IndexStore store = new IndexStore(settings);
    Path tempDir = createTempDir().resolve(settings.getUUID()).resolve("0");
    Files.createDirectories(tempDir);
    ShardPath path = new ShardPath(false, tempDir, tempDir, new ShardId(settings.getIndex(), 0));
    FsDirectoryService fsDirectoryService = new FsDirectoryService(settings, store, path);
    Directory directory = fsDirectoryService.newDirectory();
    assertFalse(directory instanceof SleepingLockWrapper);
    if (preload.length == 0) {
        assertTrue(directory.toString(), directory instanceof MMapDirectory);
        assertFalse(((MMapDirectory) directory).getPreload());
    } else if (Arrays.asList(preload).contains("*")) {
        assertTrue(directory.toString(), directory instanceof MMapDirectory);
        assertTrue(((MMapDirectory) directory).getPreload());
    } else {
        assertTrue(directory.toString(), directory instanceof FileSwitchDirectory);
        FileSwitchDirectory fsd = (FileSwitchDirectory) directory;
        assertTrue(fsd.getPrimaryDir() instanceof MMapDirectory);
        assertTrue(((MMapDirectory) fsd.getPrimaryDir()).getPreload());
        assertTrue(fsd.getSecondaryDir() instanceof MMapDirectory);
        assertFalse(((MMapDirectory) fsd.getSecondaryDir()).getPreload());
    }
}

Example 10 with MMapDirectory

use of org.apache.lucene.store.MMapDirectory in project lucene-solr by apache.

the class TestIndexWriter method testDeleteUnusedFiles.

public void testDeleteUnusedFiles() throws Exception {
    assumeFalse("test relies on exact filenames", Codec.getDefault() instanceof SimpleTextCodec);
    assumeWorkingMMapOnWindows();
    for (int iter = 0; iter < 2; iter++) {
        // relies on windows semantics
        Path path = createTempDir();
        FileSystem fs = new WindowsFS(path.getFileSystem()).getFileSystem(URI.create("file:///"));
        Path indexPath = new FilterPath(path, fs);
        // NOTE: on Unix, we cannot use MMapDir, because WindowsFS doesn't see/think it keeps file handles open.  Yet, on Windows, we MUST use
        // MMapDir because the windows OS will in fact prevent file deletion for us, and fails otherwise:
        FSDirectory dir;
        if (Constants.WINDOWS) {
            dir = new MMapDirectory(indexPath);
        } else {
            dir = new NIOFSDirectory(indexPath);
        }
        MergePolicy mergePolicy = newLogMergePolicy(true);
        // This test expects all of its segments to be in CFS
        mergePolicy.setNoCFSRatio(1.0);
        mergePolicy.setMaxCFSSegmentSizeMB(Double.POSITIVE_INFINITY);
        IndexWriter w = new IndexWriter(dir, newIndexWriterConfig(new MockAnalyzer(random())).setMergePolicy(mergePolicy).setUseCompoundFile(true));
        Document doc = new Document();
        doc.add(newTextField("field", "go", Field.Store.NO));
        w.addDocument(doc);
        DirectoryReader r;
        if (iter == 0) {
            // use NRT
            r = w.getReader();
        } else {
            // don't use NRT
            w.commit();
            r = DirectoryReader.open(dir);
        }
        assertTrue(Files.exists(indexPath.resolve("_0.cfs")));
        assertTrue(Files.exists(indexPath.resolve("_0.cfe")));
        assertTrue(Files.exists(indexPath.resolve("_0.si")));
        if (iter == 1) {
            // we run a full commit so there should be a segments file etc.
            assertTrue(Files.exists(indexPath.resolve("segments_1")));
        } else {
            // this is an NRT reopen - no segments files yet
            assertFalse(Files.exists(indexPath.resolve("segments_1")));
        }
        w.addDocument(doc);
        w.forceMerge(1);
        if (iter == 1) {
            w.commit();
        }
        IndexReader r2 = DirectoryReader.openIfChanged(r);
        assertNotNull(r2);
        assertTrue(r != r2);
        // NOTE: here we rely on "Windows" behavior, ie, even
        // though IW wanted to delete _0.cfs since it was
        // merged away, because we have a reader open
        // against this file, it should still be here:
        assertTrue(Files.exists(indexPath.resolve("_0.cfs")));
        // forceMerge created this
        //assertTrue(files.contains("_2.cfs"));
        w.deleteUnusedFiles();
        // r still holds this file open
        assertTrue(Files.exists(indexPath.resolve("_0.cfs")));
        //assertTrue(files.contains("_2.cfs"));
        r.close();
        if (iter == 0) {
            // on closing NRT reader, it calls writer.deleteUnusedFiles
            assertFalse(Files.exists(indexPath.resolve("_0.cfs")));
        } else {
            // now FSDir can remove it
            dir.deletePendingFiles();
            assertFalse(Files.exists(indexPath.resolve("_0.cfs")));
        }
        w.close();
        r2.close();
        dir.close();
    }
}