NIFI-556: Refactored TOC's to include min event id for each block

This commit is contained in:
Mark Payne 2015-04-29 13:22:09 -04:00
parent 4baf48ae95
commit 20f11b1a77
9 changed files with 161 additions and 29 deletions

View File

@ -87,11 +87,13 @@ import org.apache.nifi.provenance.serialization.RecordReader;
import org.apache.nifi.provenance.serialization.RecordReaders;
import org.apache.nifi.provenance.serialization.RecordWriter;
import org.apache.nifi.provenance.serialization.RecordWriters;
import org.apache.nifi.provenance.toc.TocReader;
import org.apache.nifi.provenance.toc.TocUtil;
import org.apache.nifi.reporting.Severity;
import org.apache.nifi.util.FormatUtils;
import org.apache.nifi.util.NiFiProperties;
import org.apache.nifi.util.RingBuffer;
import org.apache.nifi.util.RingBuffer.ForEachEvaluator;
import org.apache.nifi.util.StopWatch;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
@ -328,7 +330,7 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
final File journalFile = new File(journalDirectory, String.valueOf(initialRecordId) + ".journal." + i);
writers[i] = RecordWriters.newRecordWriter(journalFile, false, false);
writers[i].writeHeader();
writers[i].writeHeader(initialRecordId);
}
logger.info("Created new Provenance Event Writers for events starting with ID {}", initialRecordId);
@ -361,6 +363,19 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
for (final Path path : paths) {
try (RecordReader reader = RecordReaders.newRecordReader(path.toFile(), getAllLogFiles())) {
// if this is the first record, try to find out the block index and jump directly to
// the block index. This avoids having to read through a lot of data that we don't care about
// just to get to the first record that we want.
if ( records.isEmpty() ) {
final TocReader tocReader = reader.getTocReader();
if ( tocReader != null ) {
final Integer blockIndex = tocReader.getBlockIndexForEventId(firstRecordId);
if (blockIndex != null) {
reader.skipToBlock(blockIndex);
}
}
}
StandardProvenanceEventRecord record;
while (records.size() < maxRecords && ((record = reader.nextRecord()) != null)) {
if (record.getEventId() >= firstRecordId) {
@ -1231,6 +1246,7 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
}
});
long minEventId = 0L;
long earliestTimestamp = System.currentTimeMillis();
for (final RecordReader reader : readers) {
StandardProvenanceEventRecord record = null;
@ -1256,13 +1272,26 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
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);
// loop over each entry in the map, persisting the records to the merged file in order, and populating the map
// with the next entry from the journal file from which the previous record was written.
try (final RecordWriter writer = RecordWriters.newRecordWriter(writerFile, configuration.isCompressOnRollover(), true)) {
writer.writeHeader();
writer.writeHeader(minEventId);
final IndexingAction indexingAction = new IndexingAction(this);
@ -1282,7 +1311,7 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
indexingAction.index(record, indexWriter, blockIndex);
maxId = record.getEventId();
ringBuffer.add(record);
latestRecords.add(record);
records++;
// Remove this entry from the map
@ -1307,6 +1336,15 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
indexManager.returnIndexWriter(indexingDirectory, indexWriter);
}
}
// record should now be available in the repository. We can copy the values from latestRecords to ringBuffer.
latestRecords.forEach(new ForEachEvaluator<ProvenanceEventRecord>() {
@Override
public boolean evaluate(final ProvenanceEventRecord event) {
ringBuffer.add(event);
return true;
}
});
} finally {
for (final RecordReader reader : readers) {
try {
@ -1694,7 +1732,7 @@ public class PersistentProvenanceRepository implements ProvenanceEventRepository
case FORK:
case CLONE:
case REPLAY:
return submitLineageComputation(event.getParentUuids(), LineageComputationType.EXPAND_PARENTS, eventId, 0L, event.getEventTime());
return submitLineageComputation(event.getParentUuids(), LineageComputationType.EXPAND_PARENTS, eventId, event.getLineageStartDate(), event.getEventTime());
default: {
final AsyncLineageSubmission submission = new AsyncLineageSubmission(LineageComputationType.EXPAND_PARENTS, eventId, Collections.<String>emptyList(), 1);
lineageSubmissionMap.put(submission.getLineageIdentifier(), submission);

View File

@ -86,16 +86,22 @@ public class StandardRecordWriter implements RecordWriter {
}
@Override
public synchronized void writeHeader() throws IOException {
public synchronized void writeHeader(final long firstEventId) throws IOException {
lastBlockOffset = rawOutStream.getBytesWritten();
resetWriteStream();
resetWriteStream(firstEventId);
out.writeUTF(PersistentProvenanceRepository.class.getName());
out.writeInt(PersistentProvenanceRepository.SERIALIZATION_VERSION);
out.flush();
}
private void resetWriteStream() throws IOException {
/**
* Resets the streams to prepare for a new block
* @param eventId the first id that will be written to the new block
* @throws IOException if unable to flush/close the current streams properly
*/
private void resetWriteStream(final long eventId) throws IOException {
if ( out != null ) {
out.flush();
}
@ -112,13 +118,13 @@ public class StandardRecordWriter implements RecordWriter {
}
if ( tocWriter != null ) {
tocWriter.addBlockOffset(rawOutStream.getBytesWritten());
tocWriter.addBlockOffset(rawOutStream.getBytesWritten(), eventId);
}
writableStream = new BufferedOutputStream(new GZIPOutputStream(new NonCloseableOutputStream(rawOutStream), 1), 65536);
} else {
if ( tocWriter != null ) {
tocWriter.addBlockOffset(rawOutStream.getBytesWritten());
tocWriter.addBlockOffset(rawOutStream.getBytesWritten(), eventId);
}
writableStream = new BufferedOutputStream(rawOutStream, 65536);
@ -130,7 +136,7 @@ public class StandardRecordWriter implements RecordWriter {
@Override
public synchronized long writeRecord(final ProvenanceEventRecord record, long recordIdentifier) throws IOException {
public synchronized long writeRecord(final ProvenanceEventRecord record, final long recordIdentifier) throws IOException {
final ProvenanceEventType recordType = record.getEventType();
final long startBytes = byteCountingOut.getBytesWritten();
@ -142,7 +148,7 @@ public class StandardRecordWriter implements RecordWriter {
// because of the way that GZIPOutputStream works, we need to call close() on it in order for it
// to write its trailing bytes. But we don't want to close the underlying OutputStream, so we wrap
// the underlying OutputStream in a NonCloseableOutputStream
resetWriteStream();
resetWriteStream(recordIdentifier);
}
}

View File

@ -28,9 +28,10 @@ public interface RecordWriter extends Closeable {
/**
* Writes header information to the underlying stream
*
* @param firstEventId the ID of the first provenance event that will be written to the stream
* @throws IOException if unable to write header information to the underlying stream
*/
void writeHeader() throws IOException;
void writeHeader(long firstEventId) throws IOException;
/**
* Writes the given record out to the underlying stream

View File

@ -37,6 +37,7 @@ import java.io.IOException;
public class StandardTocReader implements TocReader {
private final boolean compressed;
private final long[] offsets;
private final long[] firstEventIds;
public StandardTocReader(final File file) throws IOException {
try (final FileInputStream fis = new FileInputStream(file);
@ -60,11 +61,32 @@ public class StandardTocReader implements TocReader {
throw new IOException("Table of Contents appears to be corrupt: could not read 'compression flag' from header; expected value of 0 or 1 but got " + compressionFlag);
}
final int numBlocks = (int) ((file.length() - 2) / 8);
final int blockInfoBytes;
switch (version) {
case 1:
blockInfoBytes = 8;
break;
case 2:
default:
blockInfoBytes = 16;
break;
}
final int numBlocks = (int) ((file.length() - 2) / blockInfoBytes);
offsets = new long[numBlocks];
if ( version > 1 ) {
firstEventIds = new long[numBlocks];
} else {
firstEventIds = new long[0];
}
for (int i=0; i < numBlocks; i++) {
offsets[i] = dis.readLong();
if ( version > 1 ) {
firstEventIds[i] = dis.readLong();
}
}
}
}
@ -98,11 +120,36 @@ public class StandardTocReader implements TocReader {
public int getBlockIndex(final long blockOffset) {
for (int i=0; i < offsets.length; i++) {
if ( offsets[i] > blockOffset ) {
// if the offset is less than the offset of our first block,
// just return 0 to indicate the first block. Otherwise,
// return i-1 because i represents the first block whose offset is
// greater than 'blockOffset'.
return (i == 0) ? 0 : i-1;
}
}
// None of the blocks have an offset greater than the provided offset.
// Therefore, if the event is present, it must be in the last block.
return offsets.length - 1;
}
@Override
public Integer getBlockIndexForEventId(final long eventId) {
// if we don't have event ID's stored in the TOC (which happens for version 1 of the TOC),
// or if the event ID is less than the first Event ID in this TOC, then the Event ID
// is unknown -- return null.
if ( firstEventIds.length == 0 || eventId < firstEventIds[0] ) {
return null;
}
for (int i=1; i < firstEventIds.length; i++) {
if ( firstEventIds[i] > eventId ) {
return i-1;
}
}
return offsets.length - 1;
// None of the blocks start with an Event ID greater than the provided ID.
// Therefore, if the event is present, it must be in the last block.
return firstEventIds.length - 1;
}
}

View File

@ -40,7 +40,7 @@ import org.slf4j.LoggerFactory;
public class StandardTocWriter implements TocWriter {
private static final Logger logger = LoggerFactory.getLogger(StandardTocWriter.class);
public static final byte VERSION = 1;
public static final byte VERSION = 2;
private final File file;
private final FileOutputStream fos;
@ -75,10 +75,11 @@ public class StandardTocWriter implements TocWriter {
}
@Override
public void addBlockOffset(final long offset) throws IOException {
public void addBlockOffset(final long offset, final long firstEventId) throws IOException {
final BufferedOutputStream bos = new BufferedOutputStream(fos);
final DataOutputStream dos = new DataOutputStream(bos);
dos.writeLong(offset);
dos.writeLong(firstEventId);
dos.flush();
index++;
logger.debug("Adding block {} at offset {}", index, offset);

View File

@ -59,4 +59,13 @@ public interface TocReader extends Closeable {
* @return the index of the block that contains the given offset
*/
int getBlockIndex(long blockOffset);
/**
* Returns the block index where the given event ID should be found
*
* @param eventId the ID of the provenance event of interest
* @return the block index where the given event ID should be found, or <code>null</code> if
* the block index is not known
*/
Integer getBlockIndexForEventId(long eventId);
}

View File

@ -21,16 +21,19 @@ import java.io.File;
import java.io.IOException;
/**
* Writes a .toc file
* Writes a Table-of-Contents (.toc) file
*/
public interface TocWriter extends Closeable {
/**
* Adds the given block offset as the next Block Offset in the Table of Contents
*
* @param offset the byte offset at which the block begins
* @param firstEventId the ID of the first Provenance Event that will be in the block
*
* @throws IOException if unable to persist the block index
*/
void addBlockOffset(long offset) throws IOException;
void addBlockOffset(long offset, long firstEventId) throws IOException;
/**
* @return the index of the current Block

View File

@ -67,7 +67,7 @@ public class TestStandardRecordReaderWriter {
final TocWriter tocWriter = new StandardTocWriter(tocFile, false, false);
final StandardRecordWriter writer = new StandardRecordWriter(journalFile, tocWriter, false, 1024 * 1024);
writer.writeHeader();
writer.writeHeader(1L);
writer.writeRecord(createEvent(), 1L);
writer.close();
@ -77,7 +77,7 @@ public class TestStandardRecordReaderWriter {
final StandardRecordReader reader = new StandardRecordReader(fis, journalFile.getName(), tocReader)) {
assertEquals(0, reader.getBlockIndex());
reader.skipToBlock(0);
StandardProvenanceEventRecord recovered = reader.nextRecord();
final StandardProvenanceEventRecord recovered = reader.nextRecord();
assertNotNull(recovered);
assertEquals("nifi://unit-test", recovered.getTransitUri());
@ -95,7 +95,7 @@ public class TestStandardRecordReaderWriter {
final TocWriter tocWriter = new StandardTocWriter(tocFile, false, false);
final StandardRecordWriter writer = new StandardRecordWriter(journalFile, tocWriter, true, 100);
writer.writeHeader();
writer.writeHeader(1L);
writer.writeRecord(createEvent(), 1L);
writer.close();
@ -105,7 +105,7 @@ public class TestStandardRecordReaderWriter {
final StandardRecordReader reader = new StandardRecordReader(fis, journalFile.getName(), tocReader)) {
assertEquals(0, reader.getBlockIndex());
reader.skipToBlock(0);
StandardProvenanceEventRecord recovered = reader.nextRecord();
final StandardProvenanceEventRecord recovered = reader.nextRecord();
assertNotNull(recovered);
assertEquals("nifi://unit-test", recovered.getTransitUri());
@ -124,7 +124,7 @@ public class TestStandardRecordReaderWriter {
// new record each 1 MB of uncompressed data
final StandardRecordWriter writer = new StandardRecordWriter(journalFile, tocWriter, true, 1024 * 1024);
writer.writeHeader();
writer.writeHeader(1L);
for (int i=0; i < 10; i++) {
writer.writeRecord(createEvent(), i);
}
@ -143,7 +143,7 @@ public class TestStandardRecordReaderWriter {
reader.skipToBlock(0);
}
StandardProvenanceEventRecord recovered = reader.nextRecord();
final StandardProvenanceEventRecord recovered = reader.nextRecord();
assertNotNull(recovered);
assertEquals("nifi://unit-test", recovered.getTransitUri());
}
@ -163,7 +163,7 @@ public class TestStandardRecordReaderWriter {
// new block each 10 bytes
final StandardRecordWriter writer = new StandardRecordWriter(journalFile, tocWriter, true, 100);
writer.writeHeader();
writer.writeHeader(1L);
for (int i=0; i < 10; i++) {
writer.writeRecord(createEvent(), i);
}
@ -174,7 +174,7 @@ public class TestStandardRecordReaderWriter {
try (final FileInputStream fis = new FileInputStream(journalFile);
final StandardRecordReader reader = new StandardRecordReader(fis, journalFile.getName(), tocReader)) {
for (int i=0; i < 10; i++) {
StandardProvenanceEventRecord recovered = reader.nextRecord();
final StandardProvenanceEventRecord recovered = reader.nextRecord();
System.out.println(recovered);
assertNotNull(recovered);
assertEquals((long) i, recovered.getEventId());

View File

@ -64,11 +64,11 @@ public class TestStandardTocReader {
@Test
public void testGetBlockIndex() throws IOException {
public void testGetBlockIndexV1() throws IOException {
final File file = new File("target/" + UUID.randomUUID().toString());
try (final OutputStream out = new FileOutputStream(file);
final DataOutputStream dos = new DataOutputStream(out)) {
out.write(0);
out.write(1);
out.write(0);
for (int i=0; i < 1024; i++) {
@ -88,4 +88,31 @@ public class TestStandardTocReader {
file.delete();
}
}
@Test
public void testGetBlockIndexV2() throws IOException {
final File file = new File("target/" + UUID.randomUUID().toString());
try (final OutputStream out = new FileOutputStream(file);
final DataOutputStream dos = new DataOutputStream(out)) {
out.write(2);
out.write(0);
for (int i=0; i < 1024; i++) {
dos.writeLong(i * 1024L);
dos.writeLong(0L);
}
}
try {
try(final StandardTocReader reader = new StandardTocReader(file)) {
assertFalse(reader.isCompressed());
for (int i=0; i < 1024; i++) {
assertEquals(i * 1024, reader.getBlockOffset(i));
}
}
} finally {
file.delete();
}
}
}