com.bigdata.journal.RWStrategy Maven / Gradle / Ivy
/**
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
[email protected]
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
package com.bigdata.journal;
import java.io.File;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.io.RandomAccessFile;
import java.nio.ByteBuffer;
import java.security.DigestException;
import java.security.MessageDigest;
import java.util.UUID;
import java.util.concurrent.Future;
import java.util.concurrent.locks.Lock;
import org.apache.log4j.Logger;
import com.bigdata.cache.ConcurrentWeakValueCache;
import com.bigdata.counters.CounterSet;
import com.bigdata.ha.HAGlue;
import com.bigdata.ha.QuorumRead;
import com.bigdata.ha.QuorumService;
import com.bigdata.ha.msg.HARebuildRequest;
import com.bigdata.ha.msg.IHALogRequest;
import com.bigdata.ha.msg.IHARebuildRequest;
import com.bigdata.ha.msg.IHAWriteMessage;
import com.bigdata.io.IBufferAccess;
import com.bigdata.io.writecache.WriteCacheService;
import com.bigdata.journal.AbstractJournal.ISnapshotData;
import com.bigdata.mdi.IResourceMetadata;
import com.bigdata.quorum.Quorum;
import com.bigdata.quorum.QuorumException;
import com.bigdata.rawstore.AbstractRawStore;
import com.bigdata.rawstore.IAddressManager;
import com.bigdata.rawstore.IAllocationContext;
import com.bigdata.rawstore.IPSOutputStream;
import com.bigdata.rwstore.IRWStrategy;
import com.bigdata.rwstore.IRawTx;
import com.bigdata.rwstore.RWStore;
import com.bigdata.rwstore.RWStore.StoreCounters;
import com.bigdata.util.ChecksumError;
/**
* A highly scalable persistent {@link IBufferStrategy} wrapping the
* {@link RWStore} which may be used as the backing store for a {@link Journal}.
*
* The {@link RWStore} manages allocation slots. This can translate into an
* enormous space savings on the disk for large data sets (when compared to the
* WORM) since old revisions of B+Tree nodes and leaves may be recycled
* efficiently.
*
*
History
*
* The {@link RWStrategy} supports access to historical commit states in
* combination with the history retention policy of the
* {@link ITransactionService}.
*
* Compatibility
*
* The {@link RWStore} uses a distinct binary layout on the disk based which is
* not directly compatible with the WORM binary storage layer. The WORM and the
* {@link RWStore} uses the same file header and root blocks. However, the
* {@link RWStore} defines some fields in the root blocks which are not used by
* the WORM store such as the metabits info. In addition, some of the root block
* fields defined by the WORM store are not used by the {@link RWStore}.
*
* @see RWStore.Options
*
* @author Martyn Cutcher
*/
public class RWStrategy extends AbstractRawStore implements IBufferStrategy,
IHABufferStrategy, IRWStrategy {
private static final transient Logger log = Logger.getLogger(RWStrategy.class);
private final IAddressManager m_am;
/**
* The backing store implementation.
*/
private final RWStore m_store;
/**
* The {@link UUID} for the store.
*/
private final UUID m_uuid;
/**
* The size of the backing file when it was opened by the constructor.
*/
final private long m_initialExtent;
/**
* The HA {@link Quorum} (optional).
*/
private final Quorum m_quorum;
/**
*
* @param fileMetadata
* @param quorum The HA {@link Quorum} (optional).
*/
RWStrategy(final FileMetadata fileMetadata, final Quorum quorum) {
if (fileMetadata == null)
throw new IllegalArgumentException();
m_uuid = fileMetadata.rootBlock.getUUID();
// MAY be null.
m_quorum = quorum;
m_store = new RWStore(fileMetadata, quorum);
m_am = new RWAddressManager(m_store);
m_initialExtent = fileMetadata.file.length();
}
public ByteBuffer readRootBlock(final boolean rootBlock0) {
return m_store.readRootBlock(rootBlock0);
}
@Override
public ByteBuffer read(final long addr) {
try {
// Try reading from the local store.
return readFromLocalStore(addr);
} catch (InterruptedException e) {
// wrap and rethrow.
throw new RuntimeException(e);
} catch (ChecksumError e) {
/*
* Note: This assumes that the ChecksumError is not wrapped by
* another exception. If it is, then the ChecksumError would not be
* caught.
*/
// log the error.
try {
log.error(e + " : addr=" + toString(addr), e);
} catch (Throwable ignored) {
// ignore error in logging system.
}
// update the performance counters.
final StoreCounters c = (StoreCounters) m_store.getStoreCounters()
.acquire();
try {
c.checksumErrorCount++;
} finally {
c.release();
}
if (m_quorum != null && m_quorum.isHighlyAvailable()) {
if (m_quorum.isQuorumMet()) {
try {
// Read on another node in the quorum.
final byte[] a = ((QuorumRead) m_quorum.getMember())
.readFromQuorum(m_uuid, addr);
return ByteBuffer.wrap(a);
} catch (Throwable t) {
throw new RuntimeException("While handling: " + e, t);
}
}
}
// Otherwise rethrow the checksum error.
throw e;
}
}
public long write(final ByteBuffer data) {
return write(data, null);
}
// @Override
// public long write(ByteBuffer data, long oldAddr, IAllocationContext context) {
// return write(data, oldAddr);
// }
/**
* Overridden to integrate with the shadow allocator support of the
* {@link RWStore}. Shadow allocators may be used to isolate allocation
* changes (both allocating slots and releasing slots) across different
* processes.
*/
@Override
public long write(final ByteBuffer data, final IAllocationContext context) {
if (data == null)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_BUFFER_NULL);
if (data.hasArray() && data.arrayOffset() != 0) {
/*
* @todo [data] is not always backed by an array, the array may not
* be visible (read-only), the array offset may not be zero, etc.
* Try to drive the ByteBuffer into the RWStore.alloc() method
* instead.
*
* See https://sourceforge.net/apps/trac/bigdata/ticket/151
*/
throw new AssertionError();
}
final int nbytes = data.remaining();
if (nbytes == 0)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_BUFFER_EMPTY);
final long rwaddr = m_store.alloc(data.array(), nbytes, context);
data.position(nbytes); // update position to end of buffer
final long retaddr = encodeAddr(rwaddr, nbytes);
return retaddr;
}
private long encodeAddr(long alloc, final int nbytes) {
alloc <<= 32;
alloc += nbytes;
return alloc;
}
/** Pull the latched address out of the int64 address. */
private int decodeAddr(long addr) {
addr >>= 32;
return (int) addr;
}
private int decodeSize(final long addr) {
return (int) (addr & 0xFFFFFFFF);
}
public void delete(final long addr) {
delete(addr, null/* IAllocationContext */);
}
/**
* Must check whether there are existing transactions which may access
* this data, and if not free immediately, otherwise defer.
*/
public void delete(final long addr, final IAllocationContext context) {
final int rwaddr = decodeAddr(addr);
final int sze = decodeSize(addr);
if (rwaddr == 0L)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_ADDRESS_IS_NULL);
if (sze == 0)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_BAD_RECORD_SIZE);
m_store.free(rwaddr, sze, context);
}
public void detachContext(final IAllocationContext context) {
m_store.detachContext(context);
}
public void abortContext(final IAllocationContext context) {
m_store.abortContext(context);
}
/**
* Operation is not supported.
*
* @throws UnsupportedOperationException
* always.
*/
public void closeForWrites() {
// @todo could be implemented at some point.
throw new UnsupportedOperationException();
}
public BufferMode getBufferMode() {
return BufferMode.DiskRW;
}
public CounterSet getCounters() {
return m_store.getCounters();
}
public long getExtent() {
return m_store.getStoreFile().length();
}
public int getHeaderSize() {
return FileMetadata.headerSize0;
}
public long getInitialExtent() {
return m_initialExtent;
}
public long getMaximumExtent() {
return 0L;
}
public boolean useChecksums() {
return true;
}
public long getNextOffset() {
return m_store.getNextOffset();
}
public long getUserExtent() {
return m_store.getFileStorage();
}
/**
* Operation is not supported.
*
* @throws UnsupportedOperationException
* always.
*/
public long transferTo(RandomAccessFile out) throws IOException {
// @todo could perhaps be implemented at some point.
throw new UnsupportedOperationException();
}
public void truncate(final long extent) {
m_store.establishExtent(extent);
}
public void writeRootBlock(final IRootBlockView rootBlock,
final ForceEnum forceOnCommit) {
m_store.writeRootBlock(rootBlock, forceOnCommit);
}
private void assertOpen() {
if (!m_store.isOpen())
throw new IllegalStateException(AbstractBufferStrategy.ERR_NOT_OPEN);
}
public void close() {
// throw exception if open per the API.
assertOpen();
m_store.close();
}
public void deleteResources() {
if (m_store.isOpen())
throw new IllegalStateException(AbstractBufferStrategy.ERR_OPEN);
final File file = m_store.getStoreFile();
if (file.exists()) {
if (!file.delete()) {
// throw new RuntimeException("Unable to delete file: " + file);
log.warn("Unable to delete file: " + file);
}
}
}
public void destroy() {
// close w/o exception throw.
m_store.close();
// delete the backing store.
deleteResources();
}
@Override
public void commit() {
m_store.commit();
}
/**
* Calls through to store and then to WriteCacheService.reset
*/
@Override
public void abort() {
m_store.reset();
}
public void force(final boolean metadata) {
try {
m_store.flushWrites(metadata);
} catch (IOException e) {
throw new RuntimeException(e);
}
}
public File getFile() {
return m_store.getStoreFile();
}
/**
* Not supported - this is available on the {@link AbstractJournal}.
*
* @throws UnsupportedOperationException
* always
*/
public IResourceMetadata getResourceMetadata() {
throw new UnsupportedOperationException();
}
public UUID getUUID() {
return m_uuid;
}
public boolean isFullyBuffered() {
return false;
}
public boolean isOpen() {
return m_store.isOpen();
}
public boolean isReadOnly() {
return false;
}
public boolean isStable() {
return true;
}
/**
* {@inheritDoc}
*
* Overridden to return the #of bytes in the file rather than the user
* bytes. This is because the {@link RWStore} does not know the #of bytes of
* user data in each allocation slot. Therefore it is not able to keep
* accurrate track of the user bytes as allocation slots are cycled.
*
* @see https://sourceforge.net/apps/trac/bigdata/ticket/212 (Records must
* store the as-written length for HA failover reads to be successful.)
*/
public long size() {
return m_store.getFileStorage();
}
/*
* IAddressManager
*/
public IAddressManager getAddressManager() {
return m_am;
}
public int getByteCount(final long addr) {
return m_am.getByteCount(addr);
}
public long getOffset(final long addr) {
return m_am.getOffset(addr);
}
public long toAddr(final int nbytes, final long offset) {
return m_am.toAddr(nbytes, offset);
}
public String toString(final long addr) {
return m_am.toString(addr);
}
/**
* {@inheritDoc}
*
* The state of the provided block is not relevant since it does not hold
* information on recent allocations (the meta allocations will only effect
* the root block after a commit). This is passed through to the
* {@link RWStore} which examines its internal state.
*/
public boolean requiresCommit(final IRootBlockView block) {
return m_store.requiresCommit();
}
/**
* Supports protocol in BigdataSailConnection to check for modifications
* prior to calling rollback().
*
* @return true if store has been modified since last commit()
*/
@Override
public boolean isDirty() {
return m_store.requiresCommit();
}
public long getMetaBitsAddr() {
return m_store.getMetaBitsAddr();
}
public long getMetaStartAddr() {
return m_store.getMetaStartAddr();
}
public int getMaxRecordSize() {
return m_store.getMaxAllocSize() - 4/* checksum */;
}
/**
* Although the RW Store uses a latched addressing strategy it is not
* meaningful to make this available in this interface.
*/
public int getOffsetBits() {
return 0;
}
// /**
// * Used for unit tests, could also be used to access raw statistics.
// *
// * @return the associated RWStore
// */
// @Deprecated
// public RWStore getRWStore() {
//
// return m_store;
//
// }
public RWStore getStore() {
return m_store;
}
public long getPhysicalAddress(final long addr) {
// extract the latched address.
final int rwaddr = decodeAddr(addr);
// obtain the byte offset on the file.
return m_store.physicalAddress(rwaddr);
}
/*
* IHABufferStrategy
*/
public void writeRawBuffer(final IHAWriteMessage msg, final IBufferAccess b)
throws IOException, InterruptedException {
m_store.writeRawBuffer(msg, b);
}
@Override
public Future sendHALogBuffer(final IHALogRequest req,
final IHAWriteMessage msg, final IBufferAccess b)
throws IOException, InterruptedException {
return m_store.sendHALogBuffer(req, msg, b);
}
@Override
public Future sendRawBuffer(final IHARebuildRequest req,
// long commitCounter, long commitTime,
final long sequence, final long quorumToken, final long fileExtent,
final long offset, final int nbytes, final ByteBuffer b)
throws IOException, InterruptedException {
return m_store.sendRawBuffer(req, /* commitCounter, commitTime, */
sequence, quorumToken, fileExtent, offset, nbytes, b);
}
@Override
public void writeOnStream(final OutputStream os, final ISnapshotData snapshotData,
final Quorum> quorum, final long token)
throws IOException, QuorumException {
try {
m_store.writeOnStream(os, snapshotData, quorum, token);
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
@Override
public Object snapshotAllocators() {
// TODO Auto-generated method stub
throw new UnsupportedOperationException();
}
@Override
public void computeDigest(final Object snapshot, final MessageDigest digest)
throws DigestException, IOException {
m_store.computeDigest(snapshot, digest);
}
public ByteBuffer readFromLocalStore(final long addr)
throws InterruptedException {
final int rwaddr = decodeAddr(addr);
final int sze = decodeSize(addr);
if (rwaddr == 0L)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_ADDRESS_IS_NULL);
if (sze == 0)
throw new IllegalArgumentException(
AbstractBufferStrategy.ERR_BAD_RECORD_SIZE);
/**
* Allocate buffer to include checksum to allow single read but then
* return ByteBuffer excluding those bytes
*/
// final byte buf[] = new byte[sze + 4]; // 4 bytes for checksum
//
// m_store.getData(rwaddr, buf, 0, sze+4);
//
// return ByteBuffer.wrap(buf, 0, sze);
return m_store.getData(rwaddr, sze);
}
/**
* Called from HAGlue.receiveAndReplicate to ensure the correct file extent
* prior to any writes. For RW this is essential as the allocation blocks
* for current committed data could otherwise be overwritten and the store
* invalidated.
*
* @see com.bigdata.journal.IHABufferStrategy#setExtentForLocalStore(long)
*/
public void setExtentForLocalStore(final long extent) throws IOException,
InterruptedException {
m_store.establishExtent(extent);
}
/**
* An assert oriented method that allows a finite number of addresses
* to be monitored to ensure it is not freed.
*
* @param addr - address to be locked
* @return true - for use in assert statement
*/
public boolean lockAddress(final long addr) {
m_store.lockAddress(decodeAddr(addr));
return true;
}
// /**
// * If history is retained this returns the time for which
// * data was most recently released. No request can be made for data
// * earlier than this.
// * @return latest data release time
// */
@Override
public long getLastReleaseTime() {
return m_store.getLastReleaseTime();
}
/**
* Lifted to provide a direct interface from the Journal so that the
* CommitRecordIndex can be pruned prior to store commit.
*/
public int checkDeferredFrees(final AbstractJournal journal) {
final int totalFreed = m_store.checkDeferredFrees(/*true,*/ journal); // free now if possible
if (totalFreed > 0 && log.isInfoEnabled()) {
log.info("Freed " + totalFreed + " deferralls on commit");
}
return totalFreed;
}
/**
* Return true if the address is marked as committed in the {@link RWStore}
* in memory bit maps.
*
* @param addr
* The address.
*/
public boolean isCommitted(final long addr) {
return m_store.isCommitted(decodeAddr(addr));
}
/**
* Return true iff the address was in the write cache as of the
* moment the write cache was inspected.
*
* @param addr
* The address.
*/
public boolean inWriteCache(final long addr) {
return m_store.inWriteCache(decodeAddr(addr));
}
@Override
public IRawTx newTx() {
return m_store.newTx();
}
//@Martyn: Please review
/*
*
@Override
public void registerContext(final IAllocationContext context) {
m_store.registerContext(context);
}
*/
@Override
public void registerExternalCache(
final ConcurrentWeakValueCache historicalIndexCache,
final int byteCount) {
m_store.registerExternalCache(historicalIndexCache, byteCount);
}
@Override
public long saveDeferrals() {
return m_store.saveDeferrals();
}
@Override
public InputStream getInputStream(final long addr) {
return m_store.getInputStream(addr);
}
@Override
public IPSOutputStream getOutputStream() {
return m_store.getOutputStream();
}
@Override
public IPSOutputStream getOutputStream(final IAllocationContext context) {
return m_store.getOutputStream(context);
}
@Override
public void resetFromHARootBlock(final IRootBlockView rootBlock) {
m_store.resetFromHARootBlock(rootBlock);
}
@Override
public long getBlockSequence() {
return m_store.getBlockSequence();
}
@Override
public long getCurrentBlockSequence() {
return m_store.getCurrentBlockSequence();
}
@Override
public ByteBuffer readRaw(final long position, final ByteBuffer transfer) {
return m_store.readRaw(position, transfer);
}
@Override
public void writeRawBuffer(final HARebuildRequest req,
final IHAWriteMessage msg, final ByteBuffer transfer)
throws IOException {
if (req == null)
throw new IllegalArgumentException();
// if (m_rebuildSequence != msg.getSequence())
// throw new IllegalStateException(
// "Invalid sequence number for rebuild, expected: "
// + m_rebuildSequence + ", actual: "
// + msg.getSequence());
m_store.writeRaw(msg.getFirstOffset(), transfer);
if (log.isDebugEnabled())
log.debug("Transfer rebuild: " + msg.getSequence() + ", address: "
+ msg.getFirstOffset());
// m_rebuildSequence++;
}
@Override
public Lock getCommitLock() {
return m_store.getCommitLock();
}
@Override
public void postCommit() {
m_store.postCommit();
}
@Override
public void postHACommit(final IRootBlockView rootBlock) {
m_store.postHACommit(rootBlock);
}
@Override
public WriteCacheService getWriteCacheService() {
return m_store.getWriteCacheService();
}
@Override
public StoreState getStoreState() {
return m_store.getStoreState();
}
//@Martyn: Please review
@Override
public IAllocationContext newAllocationContext(boolean isolated) {
return m_store.newAllocationContext(isolated);
}
// @Override
// public boolean isFlushed() {
// return m_store.isFlushed();
// }
// private int m_rebuildSequence = -1;
//
// @Override
// public void prepareForRebuild(final HARebuildRequest req) {
// m_store.prepareForRebuild(req);
// m_rebuildSequence = 0;
// }
//
// @Override
// public void completeRebuild(final HARebuildRequest req, final IRootBlockView rbv) {
// m_store.completeRebuild(req, rbv);
// m_rebuildSequence = -1;
// }
}