use of herddb.log.LogSequenceNumber in project herddb by diennea.
the class FileDataStorageManager method writeTables.
@Override
public Collection<PostCheckpointAction> writeTables(String tableSpace, LogSequenceNumber sequenceNumber, List<Table> tables, List<Index> indexlist) throws DataStorageManagerException {
if (sequenceNumber.isStartOfTime() && !tables.isEmpty()) {
throw new DataStorageManagerException("impossible to write a non empty table list at start-of-time");
}
Path tableSpaceDirectory = getTablespaceDirectory(tableSpace);
try {
Files.createDirectories(tableSpaceDirectory);
Path fileTables = getTablespaceTablesMetadataFile(tableSpace, sequenceNumber);
Path fileIndexes = getTablespaceIndexesMetadataFile(tableSpace, sequenceNumber);
Path parent = getParent(fileTables);
Files.createDirectories(parent);
LOGGER.log(Level.FINE, "writeTables for tableSpace " + tableSpace + " sequenceNumber " + sequenceNumber + " to " + fileTables.toAbsolutePath().toString());
try (ManagedFile file = ManagedFile.open(fileTables);
SimpleBufferedOutputStream buffer = new SimpleBufferedOutputStream(file.getOutputStream(), COPY_BUFFERS_SIZE);
ExtendedDataOutputStream dout = new ExtendedDataOutputStream(buffer)) {
// version
dout.writeVLong(1);
// flags for future implementations
dout.writeVLong(0);
dout.writeUTF(tableSpace);
dout.writeZLong(sequenceNumber.ledgerId);
dout.writeZLong(sequenceNumber.offset);
dout.writeInt(tables.size());
for (Table t : tables) {
byte[] tableSerialized = t.serialize();
dout.writeArray(tableSerialized);
}
dout.flush();
file.sync();
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
try (ManagedFile file = ManagedFile.open(fileIndexes);
SimpleBufferedOutputStream buffer = new SimpleBufferedOutputStream(file.getOutputStream(), COPY_BUFFERS_SIZE);
ExtendedDataOutputStream dout = new ExtendedDataOutputStream(buffer)) {
// version
dout.writeVLong(1);
// flags for future implementations
dout.writeVLong(0);
dout.writeUTF(tableSpace);
dout.writeZLong(sequenceNumber.ledgerId);
dout.writeZLong(sequenceNumber.offset);
if (indexlist != null) {
dout.writeInt(indexlist.size());
for (Index t : indexlist) {
byte[] indexSerialized = t.serialize();
dout.writeArray(indexSerialized);
}
} else {
dout.writeInt(0);
}
dout.flush();
file.sync();
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
Collection<PostCheckpointAction> result = new ArrayList<>();
try (DirectoryStream<Path> stream = Files.newDirectoryStream(tableSpaceDirectory)) {
for (Path p : stream) {
if (isTablespaceIndexesMetadataFile(p)) {
try {
LogSequenceNumber logPositionInFile = readLogSequenceNumberFromIndexMetadataFile(tableSpace, p);
if (sequenceNumber.after(logPositionInFile)) {
LOGGER.log(Level.FINEST, "indexes metadata file " + p.toAbsolutePath() + ". will be deleted after checkpoint end");
result.add(new DeleteFileAction("indexes", "delete indexesmetadata file " + p.toAbsolutePath(), p));
}
} catch (DataStorageManagerException ignore) {
LOGGER.log(Level.SEVERE, "Unparsable indexesmetadata file " + p.toAbsolutePath(), ignore);
result.add(new DeleteFileAction("indexes", "delete unparsable indexesmetadata file " + p.toAbsolutePath(), p));
}
} else if (isTablespaceTablesMetadataFile(p)) {
try {
LogSequenceNumber logPositionInFile = readLogSequenceNumberFromTablesMetadataFile(tableSpace, p);
if (sequenceNumber.after(logPositionInFile)) {
LOGGER.log(Level.FINEST, "tables metadata file " + p.toAbsolutePath() + ". will be deleted after checkpoint end");
result.add(new DeleteFileAction("tables", "delete tablesmetadata file " + p.toAbsolutePath(), p));
}
} catch (DataStorageManagerException ignore) {
LOGGER.log(Level.SEVERE, "Unparsable tablesmetadata file " + p.toAbsolutePath(), ignore);
result.add(new DeleteFileAction("transactions", "delete unparsable tablesmetadata file " + p.toAbsolutePath(), p));
}
}
}
} catch (IOException err) {
LOGGER.log(Level.SEVERE, "Could not list dir " + tableSpaceDirectory, err);
}
return result;
}
use of herddb.log.LogSequenceNumber in project herddb by diennea.
the class RoutedClientSideConnection method messageReceived.
@Override
@SuppressFBWarnings(value = "SF_SWITCH_NO_DEFAULT")
public void messageReceived(Message message, Channel _channel) {
switch(message.type) {
case Message.TYPE_TABLESPACE_DUMP_DATA:
{
String dumpId = (String) message.parameters.get("dumpId");
TableSpaceDumpReceiver receiver = dumpReceivers.get(dumpId);
LOGGER.log(Level.FINE, "receiver for {0}: {1}", new Object[] { dumpId, receiver });
if (receiver == null) {
if (_channel != null) {
_channel.sendReplyMessage(message, Message.ERROR(clientId, new Exception("no such dump receiver " + dumpId)));
}
return;
}
try {
Map<String, Object> values = (Map<String, Object>) message.parameters.get("values");
String command = (String) values.get("command") + "";
boolean sendAck = true;
switch(command) {
case "start":
{
long ledgerId = (long) values.get("ledgerid");
long offset = (long) values.get("offset");
receiver.start(new LogSequenceNumber(ledgerId, offset));
break;
}
case "beginTable":
{
byte[] tableDefinition = (byte[]) values.get("table");
Table table = Table.deserialize(tableDefinition);
Long estimatedSize = (Long) values.get("estimatedSize");
long dumpLedgerId = (Long) values.get("dumpLedgerid");
long dumpOffset = (Long) values.get("dumpOffset");
List<byte[]> indexesDef = (List<byte[]>) values.get("indexes");
List<Index> indexes = indexesDef.stream().map(Index::deserialize).collect(Collectors.toList());
Map<String, Object> stats = new HashMap<>();
stats.put("estimatedSize", estimatedSize);
stats.put("dumpLedgerId", dumpLedgerId);
stats.put("dumpOffset", dumpOffset);
receiver.beginTable(new DumpedTableMetadata(table, new LogSequenceNumber(dumpLedgerId, dumpOffset), indexes), stats);
break;
}
case "endTable":
{
receiver.endTable();
break;
}
case "finish":
{
long ledgerId = (long) values.get("ledgerid");
long offset = (long) values.get("offset");
receiver.finish(new LogSequenceNumber(ledgerId, offset));
sendAck = false;
break;
}
case "data":
{
List<KeyValue> data = (List<KeyValue>) values.get("records");
List<Record> records = new ArrayList<>(data.size());
for (KeyValue kv : data) {
records.add(new Record(new Bytes(kv.key), new Bytes(kv.value)));
}
receiver.receiveTableDataChunk(records);
break;
}
case "txlog":
{
List<KeyValue> data = (List<KeyValue>) values.get("records");
List<DumpedLogEntry> records = new ArrayList<>(data.size());
for (KeyValue kv : data) {
records.add(new DumpedLogEntry(LogSequenceNumber.deserialize(kv.key), kv.value));
}
receiver.receiveTransactionLogChunk(records);
break;
}
case "transactions":
{
String tableSpace = (String) values.get("tableSpace");
List<byte[]> data = (List<byte[]>) values.get("transactions");
List<Transaction> transactions = data.stream().map(array -> {
return Transaction.deserialize(tableSpace, array);
}).collect(Collectors.toList());
receiver.receiveTransactionsAtDump(transactions);
break;
}
default:
throw new DataStorageManagerException("invalid dump command:" + command);
}
if (_channel != null && sendAck) {
_channel.sendReplyMessage(message, Message.ACK(clientId));
}
} catch (DataStorageManagerException error) {
LOGGER.log(Level.SEVERE, "error while handling dump data", error);
if (_channel != null) {
_channel.sendReplyMessage(message, Message.ERROR(clientId, error));
}
}
}
break;
}
}
use of herddb.log.LogSequenceNumber in project herddb by diennea.
the class TableSpaceManager method checkpoint.
// visible for testing
public TableSpaceCheckpoint checkpoint(boolean full, boolean pin, boolean alreadLocked) throws DataStorageManagerException, LogNotAvailableException {
if (virtual) {
return null;
}
if (recoveryInProgress) {
LOGGER.log(Level.INFO, "Checkpoint for tablespace {0} skipped. Recovery is still in progress", tableSpaceName);
return null;
}
long _start = System.currentTimeMillis();
LogSequenceNumber logSequenceNumber = null;
LogSequenceNumber _logSequenceNumber = null;
Map<String, LogSequenceNumber> checkpointsTableNameSequenceNumber = new HashMap<>();
try {
List<PostCheckpointAction> actions = new ArrayList<>();
long lockStamp = 0;
if (!alreadLocked) {
lockStamp = acquireWriteLock("checkpoint");
}
try {
logSequenceNumber = log.getLastSequenceNumber();
if (logSequenceNumber.isStartOfTime()) {
LOGGER.log(Level.INFO, "{0} checkpoint {1} at {2}. skipped (no write ever issued to log)", new Object[] { nodeId, tableSpaceName, logSequenceNumber });
return new TableSpaceCheckpoint(logSequenceNumber, checkpointsTableNameSequenceNumber);
}
LOGGER.log(Level.INFO, "{0} checkpoint start {1} at {2}", new Object[] { nodeId, tableSpaceName, logSequenceNumber });
if (actualLogSequenceNumber == null) {
throw new DataStorageManagerException("actualLogSequenceNumber cannot be null");
}
// TODO: transactions checkpoint is not atomic
Collection<Transaction> currentTransactions = new ArrayList<>(transactions.values());
for (Transaction t : currentTransactions) {
LogSequenceNumber txLsn = t.lastSequenceNumber;
if (txLsn != null && txLsn.after(logSequenceNumber)) {
LOGGER.log(Level.SEVERE, "Found transaction {0} with LSN {1} in the future", new Object[] { t.transactionId, txLsn });
}
}
actions.addAll(dataStorageManager.writeTransactionsAtCheckpoint(tableSpaceUUID, logSequenceNumber, currentTransactions));
actions.addAll(writeTablesOnDataStorageManager(new CommitLogResult(logSequenceNumber, false, true), true));
// we checkpoint all data to disk and save the actual log sequence number
for (AbstractTableManager tableManager : tables.values()) {
if (!tableManager.isSystemTable()) {
TableCheckpoint checkpoint = full ? tableManager.fullCheckpoint(pin) : tableManager.checkpoint(pin);
if (checkpoint != null) {
LOGGER.log(Level.INFO, "checkpoint done for table {0}.{1} (pin: {2})", new Object[] { tableSpaceName, tableManager.getTable().name, pin });
actions.addAll(checkpoint.actions);
checkpointsTableNameSequenceNumber.put(checkpoint.tableName, checkpoint.sequenceNumber);
if (afterTableCheckPointAction != null) {
afterTableCheckPointAction.run();
}
}
}
}
// we are sure that all data as been flushed. upon recovery we will replay the log starting from this position
actions.addAll(dataStorageManager.writeCheckpointSequenceNumber(tableSpaceUUID, logSequenceNumber));
/* Indexes checkpoint is handled by TableManagers */
if (leader) {
log.dropOldLedgers(logSequenceNumber);
}
_logSequenceNumber = log.getLastSequenceNumber();
} finally {
if (!alreadLocked) {
releaseWriteLock(lockStamp, "checkpoint");
}
}
for (PostCheckpointAction action : actions) {
try {
action.run();
} catch (Exception error) {
LOGGER.log(Level.SEVERE, "postcheckpoint error:" + error, error);
}
}
return new TableSpaceCheckpoint(logSequenceNumber, checkpointsTableNameSequenceNumber);
} finally {
long _stop = System.currentTimeMillis();
LOGGER.log(Level.INFO, "{0} checkpoint finish {1} started ad {2}, finished at {3}, total time {4} ms", new Object[] { nodeId, tableSpaceName, logSequenceNumber, _logSequenceNumber, Long.toString(_stop - _start) });
checkpointTimeStats.registerSuccessfulEvent(_stop, TimeUnit.MILLISECONDS);
}
}
use of herddb.log.LogSequenceNumber in project herddb by diennea.
the class FileDataStorageManager method tableCheckpoint.
@Override
public List<PostCheckpointAction> tableCheckpoint(String tableSpace, String tableName, TableStatus tableStatus, boolean pin) throws DataStorageManagerException {
LogSequenceNumber logPosition = tableStatus.sequenceNumber;
Path dir = getTableDirectory(tableSpace, tableName);
Path checkpointFile = getTableCheckPointsFile(dir, logPosition);
try {
if (Files.isRegularFile(checkpointFile)) {
TableStatus actualStatus = readTableStatusFromFile(checkpointFile);
if (actualStatus != null && actualStatus.equals(tableStatus)) {
LOGGER.log(Level.FINE, "tableCheckpoint " + tableSpace + ", " + tableName + ": " + tableStatus + " (pin:" + pin + ") already saved on file " + checkpointFile);
return Collections.emptyList();
}
}
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
Path parent = getParent(checkpointFile);
Path checkpointFileTemp = parent.resolve(checkpointFile.getFileName() + ".tmp");
LOGGER.log(Level.FINE, "tableCheckpoint " + tableSpace + ", " + tableName + ": " + tableStatus + " (pin:" + pin + ") to file " + checkpointFile);
try (ManagedFile file = ManagedFile.open(checkpointFileTemp, requirefsync);
SimpleBufferedOutputStream buffer = new SimpleBufferedOutputStream(file.getOutputStream(), COPY_BUFFERS_SIZE);
XXHash64Utils.HashingOutputStream oo = new XXHash64Utils.HashingOutputStream(buffer);
ExtendedDataOutputStream dataOutputKeys = new ExtendedDataOutputStream(oo)) {
// version
dataOutputKeys.writeVLong(1);
// flags for future implementations
dataOutputKeys.writeVLong(0);
tableStatus.serialize(dataOutputKeys);
dataOutputKeys.writeLong(oo.hash());
dataOutputKeys.flush();
file.sync();
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
try {
Files.move(checkpointFileTemp, checkpointFile, StandardCopyOption.REPLACE_EXISTING, StandardCopyOption.ATOMIC_MOVE);
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
/* Checkpoint pinning */
final Map<Long, Integer> pins = pinTableAndGetPages(tableSpace, tableName, tableStatus, pin);
final Set<LogSequenceNumber> checkpoints = pinTableAndGetCheckpoints(tableSpace, tableName, tableStatus, pin);
long maxPageId = tableStatus.activePages.keySet().stream().max(Comparator.naturalOrder()).orElse(Long.MAX_VALUE);
List<PostCheckpointAction> result = new ArrayList<>();
// we can drop old page files now
List<Path> pageFiles = getTablePageFiles(tableSpace, tableName);
for (Path p : pageFiles) {
long pageId = getPageId(p);
LOGGER.log(Level.FINEST, "checkpoint file {0} pageId {1}", new Object[] { p.toAbsolutePath(), pageId });
if (pageId > 0 && !pins.containsKey(pageId) && !tableStatus.activePages.containsKey(pageId) && pageId < maxPageId) {
LOGGER.log(Level.FINEST, "checkpoint file " + p.toAbsolutePath() + " pageId " + pageId + ". will be deleted after checkpoint end");
result.add(new DeleteFileAction(tableSpace, tableName, "delete page " + pageId + " file " + p.toAbsolutePath(), p));
}
}
try (DirectoryStream<Path> stream = Files.newDirectoryStream(dir)) {
for (Path p : stream) {
if (isTableOrIndexCheckpointsFile(p) && !p.equals(checkpointFile)) {
TableStatus status = readTableStatusFromFile(p);
if (logPosition.after(status.sequenceNumber) && !checkpoints.contains(status.sequenceNumber)) {
LOGGER.log(Level.FINEST, "checkpoint metadata file " + p.toAbsolutePath() + ". will be deleted after checkpoint end");
result.add(new DeleteFileAction(tableSpace, tableName, "delete checkpoint metadata file " + p.toAbsolutePath(), p));
}
}
}
} catch (IOException err) {
LOGGER.log(Level.SEVERE, "Could not list table dir " + dir, err);
}
return result;
}
use of herddb.log.LogSequenceNumber in project herddb by diennea.
the class FileDataStorageManager method writeCheckpointSequenceNumber.
@Override
public Collection<PostCheckpointAction> writeCheckpointSequenceNumber(String tableSpace, LogSequenceNumber sequenceNumber) throws DataStorageManagerException {
Path tableSpaceDirectory = getTablespaceDirectory(tableSpace);
try {
Files.createDirectories(tableSpaceDirectory);
Path checkPointFile = getTablespaceCheckPointInfoFile(tableSpace, sequenceNumber);
Path parent = getParent(checkPointFile);
Files.createDirectories(parent);
Path checkpointFileTemp = parent.resolve(checkPointFile.getFileName() + ".tmp");
LOGGER.log(Level.INFO, "checkpoint for " + tableSpace + " at " + sequenceNumber + " to " + checkPointFile.toAbsolutePath().toString());
try (ManagedFile file = ManagedFile.open(checkpointFileTemp, requirefsync);
SimpleBufferedOutputStream buffer = new SimpleBufferedOutputStream(file.getOutputStream(), COPY_BUFFERS_SIZE);
ExtendedDataOutputStream dout = new ExtendedDataOutputStream(buffer)) {
// version
dout.writeVLong(1);
// flags for future implementations
dout.writeVLong(0);
dout.writeUTF(tableSpace);
dout.writeZLong(sequenceNumber.ledgerId);
dout.writeZLong(sequenceNumber.offset);
dout.flush();
file.sync();
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
// write file atomically
Files.move(checkpointFileTemp, checkPointFile, StandardCopyOption.ATOMIC_MOVE, StandardCopyOption.REPLACE_EXISTING);
} catch (IOException err) {
throw new DataStorageManagerException(err);
}
Collection<PostCheckpointAction> result = new ArrayList<>();
try (DirectoryStream<Path> stream = Files.newDirectoryStream(tableSpaceDirectory)) {
for (Path p : stream) {
if (isTablespaceCheckPointInfoFile(p)) {
try {
LogSequenceNumber logPositionInFile = readLogSequenceNumberFromCheckpointInfoFile(tableSpace, p);
if (sequenceNumber.after(logPositionInFile)) {
LOGGER.log(Level.FINEST, "checkpoint info file " + p.toAbsolutePath() + ". will be deleted after checkpoint end");
result.add(new DeleteFileAction(tableSpace, "checkpoint", "delete checkpoint info file " + p.toAbsolutePath(), p));
}
} catch (DataStorageManagerException ignore) {
LOGGER.log(Level.SEVERE, "unparsable checkpoint info file " + p.toAbsolutePath(), ignore);
// do not auto-delete checkpoint files
}
}
}
} catch (IOException err) {
LOGGER.log(Level.SEVERE, "Could not list dir " + tableSpaceDirectory, err);
}
return result;
}
Aggregations