Examples with EventIndexWriter org.apache.nifi.provenance.index.EventIndexWriter used on opensource projects

Example 6 with EventIndexWriter

use of org.apache.nifi.provenance.index.EventIndexWriter in project nifi by apache.

the class PersistentProvenanceRepository method mergeJournals.

/**
 * <p>
 * Merges all of the given Journal Files into a single, merged Provenance
 * Event Log File. As these records are merged, they will be compressed, if
 * the repository is configured to compress records, and will be indexed.
 * </p>
 *
 * <p>
 * If the repository is configured to compress the data, the file written to
 * may not be the same as the <code>suggestedMergeFile</code>, as a filename
 * extension of '.gz' may be appended. If the journals are successfully
 * merged, the file that they were merged into will be returned. If unable
 * to merge the records (for instance, because the repository has been
 * closed or because the list of journal files was empty), this method will
 * return <code>null</code>.
 * </p>
 *
 * @param journalFiles the journal files to merge
 * @param suggestedMergeFile the file to write the merged records to
 * @param eventReporter the event reporter to report any warnings or errors
 * to; may be null.
 *
 * @return the file that the given journals were merged into, or
 * <code>null</code> if no records were merged.
 *
 * @throws IOException if a problem occurs writing to the mergedFile,
 * reading from a journal, or updating the Lucene Index.
 */
File mergeJournals(final List<File> journalFiles, final File suggestedMergeFile, final EventReporter eventReporter) throws IOException {
    logger.debug("Merging {} to {}", journalFiles, suggestedMergeFile);
    if (this.closed.get()) {
        logger.info("Provenance Repository has been closed; will not merge journal files to {}", suggestedMergeFile);
        return null;
    }
    if (journalFiles.isEmpty()) {
        logger.debug("Couldn't merge journals: Journal Files is empty; won't merge journals");
        return null;
    }
    Collections.sort(journalFiles, new Comparator<File>() {

        @Override
        public int compare(final File o1, final File o2) {
            final String suffix1 = LuceneUtil.substringAfterLast(o1.getName(), ".");
            final String suffix2 = LuceneUtil.substringAfterLast(o2.getName(), ".");
            try {
                final int journalIndex1 = Integer.parseInt(suffix1);
                final int journalIndex2 = Integer.parseInt(suffix2);
                return Integer.compare(journalIndex1, journalIndex2);
            } catch (final NumberFormatException nfe) {
                return o1.getName().compareTo(o2.getName());
            }
        }
    });
    // Search for any missing files. At this point they should have been written to disk otherwise cannot continue.
    // Missing files is most likely due to incomplete cleanup of files post merge
    final List<File> availableFiles = filterUnavailableFiles(journalFiles);
    final int numAvailableFiles = availableFiles.size();
    // check if we have all of the "partial" files for the journal.
    if (numAvailableFiles > 0) {
        if (suggestedMergeFile.exists()) {
            // we have all "partial" files and there is already a merged file. Delete the data from the index
            // because the merge file may not be fully merged. We will re-merge.
            logger.warn("Merged Journal File {} already exists; however, all partial journal files also exist " + "so assuming that the merge did not finish. Repeating procedure in order to ensure consistency.");
            final DeleteIndexAction deleteAction = new DeleteIndexAction(this, indexConfig, getIndexManager());
            try {
                deleteAction.execute(suggestedMergeFile);
            } catch (final Exception e) {
                logger.warn("Failed to delete records from Journal File {} from the index; this could potentially result in duplicates. Failure was due to {}", suggestedMergeFile, e.toString());
                if (logger.isDebugEnabled()) {
                    logger.warn("", e);
                }
            }
            // file and the TOC file. Otherwise, we could get the wrong copy and have issues retrieving events.
            if (!suggestedMergeFile.delete()) {
                logger.error("Failed to delete partially written Provenance Journal File {}. This may result in events from this journal " + "file not being able to be displayed. This file should be deleted manually.", suggestedMergeFile);
            }
            final File tocFile = TocUtil.getTocFile(suggestedMergeFile);
            if (tocFile.exists() && !tocFile.delete()) {
                logger.error("Failed to delete .toc file {}; this may result in not being able to read the Provenance Events from the {} Journal File. " + "This can be corrected by manually deleting the {} file", tocFile, suggestedMergeFile, tocFile);
            }
        }
    } else {
        logger.warn("Cannot merge journal files {} because they do not exist on disk", journalFiles);
        return null;
    }
    final long startNanos = System.nanoTime();
    // Map each journal to a RecordReader
    final List<RecordReader> readers = new ArrayList<>();
    int records = 0;
    final boolean isCompress = configuration.isCompressOnRollover();
    final File writerFile = isCompress ? new File(suggestedMergeFile.getParentFile(), suggestedMergeFile.getName() + ".gz") : suggestedMergeFile;
    try {
        for (final File journalFile : availableFiles) {
            try {
                // Use MAX_VALUE for number of chars because we don't want to truncate the value as we write it
                // out. This allows us to later decide that we want more characters and still be able to retrieve
                // the entire event.
                readers.add(RecordReaders.newRecordReader(journalFile, null, Integer.MAX_VALUE));
            } catch (final EOFException eof) {
            // there's nothing here. Skip over it.
            } catch (final IOException ioe) {
                logger.warn("Unable to merge {} with other Journal Files due to {}", journalFile, ioe.toString());
                if (logger.isDebugEnabled()) {
                    logger.warn("", ioe);
                }
                if (eventReporter != null) {
                    eventReporter.reportEvent(Severity.ERROR, EVENT_CATEGORY, "Failed to merge Journal Files due to " + ioe.toString());
                }
            }
        }
        // Create a Map so that the key is the next record available from a reader and the value is the Reader from which
        // the record came. This sorted map is then used so that we are able to always get the first entry, which is the next
        // lowest record id
        final SortedMap<StandardProvenanceEventRecord, RecordReader> recordToReaderMap = new TreeMap<>(new Comparator<StandardProvenanceEventRecord>() {

            @Override
            public int compare(final StandardProvenanceEventRecord o1, final StandardProvenanceEventRecord o2) {
                return Long.compare(o1.getEventId(), o2.getEventId());
            }
        });
        long minEventId = 0L;
        long earliestTimestamp = System.currentTimeMillis();
        for (final RecordReader reader : readers) {
            StandardProvenanceEventRecord record = null;
            try {
                record = reader.nextRecord();
            } catch (final EOFException eof) {
            // record will be null and reader can no longer be used
            } catch (final Exception e) {
                logger.warn("Failed to generate Provenance Event Record from Journal due to " + e + "; it's " + "possible that the record wasn't completely written to the file. This journal will be " + "skipped.");
                if (logger.isDebugEnabled()) {
                    logger.warn("", e);
                }
                if (eventReporter != null) {
                    eventReporter.reportEvent(Severity.WARNING, EVENT_CATEGORY, "Failed to read Provenance Event " + "Record from Journal due to " + e + "; it's possible that the record wasn't " + "completely written to the file. This journal will be skipped.");
                }
            }
            if (record == null) {
                continue;
            }
            if (record.getEventTime() < earliestTimestamp) {
                earliestTimestamp = record.getEventTime();
            }
            if (record.getEventId() < minEventId) {
                minEventId = record.getEventId();
            }
            recordToReaderMap.put(record, reader);
        }
        // We want to keep track of the last 1000 events in the files so that we can add them to 'ringBuffer'.
        // However, we don't want to add them directly to ringBuffer, because once they are added to ringBuffer, they are
        // available in query results. As a result, we can have the issue where we've not finished indexing the file
        // but we try to create the lineage for events in that file. In order to avoid this, we will add the records
        // to a temporary RingBuffer and after we finish merging the records will then copy the data to the
        // ringBuffer provided as a method argument.
        final RingBuffer<ProvenanceEventRecord> latestRecords = new RingBuffer<>(1000);
        // with the next entry from the journal file from which the previous record was written.
        try (final RecordWriter writer = RecordWriters.newSchemaRecordWriter(writerFile, idGenerator, configuration.isCompressOnRollover(), true)) {
            writer.writeHeader(minEventId);
            final IndexingAction indexingAction = createIndexingAction();
            final File indexingDirectory = indexConfig.getWritableIndexDirectory(writerFile, earliestTimestamp);
            long maxId = 0L;
            final BlockingQueue<Tuple<StandardProvenanceEventRecord, Integer>> eventQueue = new LinkedBlockingQueue<>(100);
            final AtomicBoolean finishedAdding = new AtomicBoolean(false);
            final List<Future<?>> futures = new ArrayList<>();
            final EventIndexWriter indexWriter = getIndexManager().borrowIndexWriter(indexingDirectory);
            try {
                final ExecutorService exec = Executors.newFixedThreadPool(configuration.getIndexThreadPoolSize(), new ThreadFactory() {

                    @Override
                    public Thread newThread(final Runnable r) {
                        final Thread t = Executors.defaultThreadFactory().newThread(r);
                        t.setName("Index Provenance Events");
                        return t;
                    }
                });
                final AtomicInteger indexingFailureCount = new AtomicInteger(0);
                try {
                    for (int i = 0; i < configuration.getIndexThreadPoolSize(); i++) {
                        final Callable<Object> callable = new Callable<Object>() {

                            @Override
                            public Object call() throws IOException {
                                while (!eventQueue.isEmpty() || !finishedAdding.get()) {
                                    try {
                                        final Tuple<StandardProvenanceEventRecord, Integer> tuple;
                                        try {
                                            tuple = eventQueue.poll(10, TimeUnit.MILLISECONDS);
                                        } catch (final InterruptedException ie) {
                                            Thread.currentThread().interrupt();
                                            continue;
                                        }
                                        if (tuple == null) {
                                            continue;
                                        }
                                        indexingAction.index(tuple.getKey(), indexWriter.getIndexWriter(), tuple.getValue());
                                    } catch (final Throwable t) {
                                        logger.error("Failed to index Provenance Event for " + writerFile + " to " + indexingDirectory, t);
                                        if (indexingFailureCount.incrementAndGet() >= MAX_INDEXING_FAILURE_COUNT) {
                                            return null;
                                        }
                                    }
                                }
                                return null;
                            }
                        };
                        final Future<?> future = exec.submit(callable);
                        futures.add(future);
                    }
                    boolean indexEvents = true;
                    while (!recordToReaderMap.isEmpty()) {
                        final StandardProvenanceEventRecord record = recordToReaderMap.firstKey();
                        final RecordReader reader = recordToReaderMap.get(record);
                        writer.writeRecord(record);
                        final int blockIndex = writer.getTocWriter().getCurrentBlockIndex();
                        boolean accepted = false;
                        while (!accepted && indexEvents) {
                            try {
                                accepted = eventQueue.offer(new Tuple<>(record, blockIndex), 10, TimeUnit.MILLISECONDS);
                            } catch (final InterruptedException ie) {
                                Thread.currentThread().interrupt();
                            }
                            // So, if the queue is filled, we will check if this is the case.
                            if (!accepted && indexingFailureCount.get() >= MAX_INDEXING_FAILURE_COUNT) {
                                // don't add anything else to the queue.
                                indexEvents = false;
                                eventQueue.clear();
                                final String warning = String.format("Indexing Provenance Events for %s has failed %s times. This exceeds the maximum threshold of %s failures, " + "so no more Provenance Events will be indexed for this Provenance file.", writerFile, indexingFailureCount.get(), MAX_INDEXING_FAILURE_COUNT);
                                logger.warn(warning);
                                if (eventReporter != null) {
                                    eventReporter.reportEvent(Severity.WARNING, EVENT_CATEGORY, warning);
                                }
                            }
                        }
                        maxId = record.getEventId();
                        latestRecords.add(truncateAttributes(record));
                        records++;
                        // Remove this entry from the map
                        recordToReaderMap.remove(record);
                        // Get the next entry from this reader and add it to the map
                        StandardProvenanceEventRecord nextRecord = null;
                        try {
                            nextRecord = reader.nextRecord();
                        } catch (final EOFException eof) {
                        // record will be null and reader can no longer be used
                        } catch (final Exception e) {
                            logger.warn("Failed to generate Provenance Event Record from Journal due to " + e + "; it's possible that the record wasn't completely written to the file. " + "The remainder of this journal will be skipped.");
                            if (logger.isDebugEnabled()) {
                                logger.warn("", e);
                            }
                            if (eventReporter != null) {
                                eventReporter.reportEvent(Severity.WARNING, EVENT_CATEGORY, "Failed to read " + "Provenance Event Record from Journal due to " + e + "; it's possible " + "that the record wasn't completely written to the file. The remainder " + "of this journal will be skipped.");
                            }
                        }
                        if (nextRecord != null) {
                            recordToReaderMap.put(nextRecord, reader);
                        }
                    }
                } finally {
                    finishedAdding.set(true);
                    exec.shutdown();
                }
                for (final Future<?> future : futures) {
                    try {
                        future.get();
                    } catch (final ExecutionException ee) {
                        final Throwable t = ee.getCause();
                        if (t instanceof RuntimeException) {
                            throw (RuntimeException) t;
                        }
                        throw new RuntimeException(t);
                    } catch (final InterruptedException e) {
                        Thread.currentThread().interrupt();
                        throw new RuntimeException("Thread interrupted");
                    }
                }
            } finally {
                getIndexManager().returnIndexWriter(indexWriter);
            }
            indexConfig.setMaxIdIndexed(maxId);
        }
        // record should now be available in the repository. We can copy the values from latestRecords to ringBuffer.
        final RingBuffer<ProvenanceEventRecord> latestRecordBuffer = this.latestRecords;
        latestRecords.forEach(new ForEachEvaluator<ProvenanceEventRecord>() {

            @Override
            public boolean evaluate(final ProvenanceEventRecord event) {
                latestRecordBuffer.add(event);
                return true;
            }
        });
    } finally {
        for (final RecordReader reader : readers) {
            try {
                reader.close();
            } catch (final IOException ioe) {
            }
        }
    }
    // Success. Remove all of the journal files, as they're no longer needed, now that they've been merged.
    for (final File journalFile : availableFiles) {
        if (!journalFile.delete() && journalFile.exists()) {
            logger.warn("Failed to remove temporary journal file {}; this file should be cleaned up manually", journalFile.getAbsolutePath());
            if (eventReporter != null) {
                eventReporter.reportEvent(Severity.WARNING, EVENT_CATEGORY, "Failed to remove temporary journal file " + journalFile.getAbsolutePath() + "; this file should be cleaned up manually");
            }
        }
        final File tocFile = TocUtil.getTocFile(journalFile);
        if (!tocFile.delete() && tocFile.exists()) {
            logger.warn("Failed to remove temporary journal TOC file {}; this file should be cleaned up manually", tocFile.getAbsolutePath());
            if (eventReporter != null) {
                eventReporter.reportEvent(Severity.WARNING, EVENT_CATEGORY, "Failed to remove temporary journal TOC file " + tocFile.getAbsolutePath() + "; this file should be cleaned up manually");
            }
        }
    }
    if (records == 0) {
        writerFile.delete();
        logger.debug("Couldn't merge journals: No Records to merge");
        return null;
    } else {
        final long nanos = System.nanoTime() - startNanos;
        final long millis = TimeUnit.MILLISECONDS.convert(nanos, TimeUnit.NANOSECONDS);
        logger.info("Successfully merged {} journal files ({} records) into single Provenance Log File {} in {} milliseconds", numAvailableFiles, records, suggestedMergeFile, millis);
    }
    return writerFile;
}

Example 7 with EventIndexWriter

use of org.apache.nifi.provenance.index.EventIndexWriter in project nifi by apache.

the class EventIndexTask method index.

private void index(final List<StoredDocument> toIndex, final String partitionName) throws IOException {
    if (toIndex.isEmpty()) {
        return;
    }
    // Convert the IndexableDocument list into a List of Documents so that we can pass them to the Index Writer.
    final List<Document> documents = toIndex.stream().map(doc -> doc.getDocument()).collect(Collectors.toList());
    boolean requestClose = false;
    boolean requestCommit = false;
    final long minEventTime = toIndex.stream().mapToLong(doc -> doc.getDocument().getField(SearchableFields.EventTime.getSearchableFieldName()).numericValue().longValue()).min().getAsLong();
    // Synchronize on the directory manager because we don't want the active directory to change
    // while we are obtaining an index writer for it. I.e., determining the active directory
    // and obtaining an Index Writer for it need to be done atomically.
    final EventIndexWriter indexWriter;
    final File indexDirectory;
    synchronized (directoryManager) {
        indexDirectory = directoryManager.getWritableIndexingDirectory(minEventTime, partitionName);
        indexWriter = indexManager.borrowIndexWriter(indexDirectory);
    }
    try {
        // Perform the actual indexing.
        boolean writerIndicatesCommit = indexWriter.index(documents, commitThreshold);
        // If we don't need to commit index based on what index writer tells us, we will still want
        // to commit the index if it's assigned to a partition and this is no longer the active index
        // for that partition. This prevents the following case:
        // 
        // Thread T1: pulls events from queue
        // Maps events to Index Directory D1
        // Thread T2: pulls events from queue
        // Maps events to Index Directory D1, the active index for Partition P1.
        // Writes events to D1.
        // Commits Index Writer for D1.
        // Closes Index Writer for D1.
        // Thread T1: Writes events to D1.
        // Determines that Index Writer for D1 does not need to be committed or closed.
        // 
        // In the case outlined above, we would potentially lose those events from the index! To avoid this,
        // we simply decide to commit the index if this writer is no longer the active writer for the index.
        // However, if we have 10 threads, we don't want all 10 threads trying to commit the index after each
        // update. We want to commit when they've all finished. This is what the IndexManager will do if we request
        // that it commit the index. It will also close the index if requested, once all writers have finished.
        // So when this is the case, we will request that the Index Manager both commit and close the writer.
        final Optional<File> activeIndexDirOption = directoryManager.getActiveIndexDirectory(partitionName);
        if (!activeIndexDirOption.isPresent() || !activeIndexDirOption.get().equals(indexDirectory)) {
            requestCommit = true;
            requestClose = true;
        }
        if (writerIndicatesCommit) {
            commit(indexWriter);
            // we've already committed the index writer so no need to request that the index manager do so also.
            requestCommit = false;
            final boolean directoryManagerIndicatesClose = directoryManager.onIndexCommitted(indexDirectory);
            requestClose = requestClose || directoryManagerIndicatesClose;
            if (logger.isDebugEnabled()) {
                final long maxId = documents.stream().mapToLong(doc -> doc.getField(SearchableFields.Identifier.getSearchableFieldName()).numericValue().longValue()).max().orElse(-1L);
                logger.debug("Committed index {} after writing a max Event ID of {}", indexDirectory, maxId);
            }
        }
    } finally {
        indexManager.returnIndexWriter(indexWriter, requestCommit, requestClose);
    }
}

Example 8 with EventIndexWriter

use of org.apache.nifi.provenance.index.EventIndexWriter in project nifi by apache.

the class CachingIndexManager method borrowIndexWriter.

@Override
public EventIndexWriter borrowIndexWriter(final File indexDirectory) throws IOException {
    final File absoluteFile = indexDirectory.getAbsoluteFile();
    logger.trace("Borrowing index writer for {}", indexDirectory);
    lock.lock();
    try {
        IndexWriterCount writerCount = writerCounts.remove(absoluteFile);
        if (writerCount == null) {
            final List<Closeable> closeables = new ArrayList<>();
            final Directory directory = FSDirectory.open(indexDirectory);
            closeables.add(directory);
            try {
                final Analyzer analyzer = new StandardAnalyzer();
                closeables.add(analyzer);
                final IndexWriterConfig config = new IndexWriterConfig(LuceneUtil.LUCENE_VERSION, analyzer);
                config.setWriteLockTimeout(300000L);
                final IndexWriter indexWriter = new IndexWriter(directory, config);
                final EventIndexWriter eventIndexWriter = new LuceneEventIndexWriter(indexWriter, indexDirectory);
                writerCount = new IndexWriterCount(eventIndexWriter, analyzer, directory, 1);
                logger.debug("Providing new index writer for {}", indexDirectory);
            } catch (final IOException ioe) {
                for (final Closeable closeable : closeables) {
                    try {
                        closeable.close();
                    } catch (final IOException ioe2) {
                        ioe.addSuppressed(ioe2);
                    }
                }
                throw ioe;
            }
            writerCounts.put(absoluteFile, writerCount);
            // Mark any active searchers as poisoned because we are updating the index
            final List<ActiveIndexSearcher> searchers = activeSearchers.get(absoluteFile);
            if (searchers != null) {
                for (final ActiveIndexSearcher activeSearcher : searchers) {
                    logger.debug("Poisoning {} because it is searching {}, which is getting updated", activeSearcher, indexDirectory);
                    activeSearcher.poison();
                }
            }
        } else {
            logger.debug("Providing existing index writer for {} and incrementing count to {}", indexDirectory, writerCount.getCount() + 1);
            writerCounts.put(absoluteFile, new IndexWriterCount(writerCount.getWriter(), writerCount.getAnalyzer(), writerCount.getDirectory(), writerCount.getCount() + 1));
        }
        return writerCount.getWriter();
    } finally {
        lock.unlock();
    }
}

Example 9 with EventIndexWriter

use of org.apache.nifi.provenance.index.EventIndexWriter in project nifi by apache.

the class DeleteIndexAction method execute.

@Override
public File execute(final File expiredFile) throws IOException {
    // count the number of records and determine the max event id that we are deleting.
    final long numDeleted = 0;
    long maxEventId = -1L;
    try (final RecordReader reader = RecordReaders.newRecordReader(expiredFile, repository.getAllLogFiles(), Integer.MAX_VALUE)) {
        maxEventId = reader.getMaxEventId();
    } catch (final IOException ioe) {
        logger.warn("Failed to obtain max ID present in journal file {}", expiredFile.getAbsolutePath());
    }
    // remove the records from the index
    final List<File> indexDirs = indexConfiguration.getIndexDirectories(expiredFile);
    for (final File indexingDirectory : indexDirs) {
        final Term term = new Term(FieldNames.STORAGE_FILENAME, LuceneUtil.substringBefore(expiredFile.getName(), "."));
        boolean deleteDir = false;
        final EventIndexWriter writer = indexManager.borrowIndexWriter(indexingDirectory);
        try {
            final IndexWriter indexWriter = writer.getIndexWriter();
            indexWriter.deleteDocuments(term);
            indexWriter.commit();
            final int docsLeft = indexWriter.numDocs();
            deleteDir = docsLeft <= 0;
            logger.debug("After expiring {}, there are {} docs left for index {}", expiredFile, docsLeft, indexingDirectory);
        } finally {
            indexManager.returnIndexWriter(writer);
        }
        // we've confirmed that all documents have been removed. Delete the index directory.
        if (deleteDir) {
            indexManager.removeIndex(indexingDirectory);
            indexConfiguration.removeIndexDirectory(indexingDirectory);
            deleteDirectory(indexingDirectory);
            logger.info("Removed empty index directory {}", indexingDirectory);
        }
    }
    // Update the minimum index to 1 more than the max Event ID in this file.
    if (maxEventId > -1L) {
        indexConfiguration.setMinIdIndexed(maxEventId + 1L);
    }
    logger.info("Deleted Indices for Expired Provenance File {} from {} index files; {} documents removed", expiredFile, indexDirs.size(), numDeleted);
    return expiredFile;
}

Example 10 with EventIndexWriter

use of org.apache.nifi.provenance.index.EventIndexWriter in project nifi by apache.

the class SimpleIndexManager method borrowIndexSearcher.

@Override
public EventIndexSearcher borrowIndexSearcher(final File indexDir) throws IOException {
    final File absoluteFile = indexDir.getAbsoluteFile();
    final IndexWriterCount writerCount;
    synchronized (writerCounts) {
        writerCount = writerCounts.remove(absoluteFile);
        if (writerCount != null) {
            // Increment writer count and create an Index Searcher based on the writer
            writerCounts.put(absoluteFile, new IndexWriterCount(writerCount.getWriter(), writerCount.getAnalyzer(), writerCount.getDirectory(), writerCount.getCount() + 1, writerCount.isCloseableWhenUnused()));
        }
    }
    final DirectoryReader directoryReader;
    if (writerCount == null) {
        logger.trace("Creating index searcher for {}", indexDir);
        final Directory directory = FSDirectory.open(indexDir);
        directoryReader = DirectoryReader.open(directory);
    } else {
        final EventIndexWriter eventIndexWriter = writerCount.getWriter();
        directoryReader = DirectoryReader.open(eventIndexWriter.getIndexWriter(), false);
    }
    final IndexSearcher searcher = new IndexSearcher(directoryReader, this.searchExecutor);
    logger.trace("Created index searcher {} for {}", searcher, indexDir);
    return new LuceneEventIndexSearcher(searcher, indexDir, null, directoryReader);
}