com.bigdata.journal.WORMStrategy 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.OutputStream;
import java.io.RandomAccessFile;
import java.nio.ByteBuffer;
import java.nio.channels.Channel;
import java.nio.channels.FileChannel;
import java.security.DigestException;
import java.security.MessageDigest;
import java.util.Map;
import java.util.UUID;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicReference;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.ReadLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.WriteLock;
import com.bigdata.btree.BTree.Counter;
import com.bigdata.counters.AbstractStatisticsCollector;
import com.bigdata.counters.CounterSet;
import com.bigdata.counters.Instrument;
import com.bigdata.counters.striped.StripedCounters;
import com.bigdata.ha.HAGlue;
import com.bigdata.ha.HAPipelineGlue;
import com.bigdata.ha.QuorumPipeline;
import com.bigdata.ha.QuorumRead;
import com.bigdata.ha.QuorumService;
import com.bigdata.ha.msg.HARebuildRequest;
import com.bigdata.ha.msg.HAWriteMessage;
import com.bigdata.ha.msg.IHALogRequest;
import com.bigdata.ha.msg.IHARebuildRequest;
import com.bigdata.ha.msg.IHAWriteMessage;
import com.bigdata.io.ChecksumUtility;
import com.bigdata.io.DirectBufferPool;
import com.bigdata.io.FileChannelUtility;
import com.bigdata.io.IBufferAccess;
import com.bigdata.io.IReopenChannel;
import com.bigdata.io.MergeStreamWithSnapshotData;
import com.bigdata.io.compression.CompressorRegistry;
import com.bigdata.io.compression.IRecordCompressor;
import com.bigdata.io.writecache.IBackingReader;
import com.bigdata.io.writecache.WriteCache;
import com.bigdata.io.writecache.WriteCacheCounters;
import com.bigdata.io.writecache.WriteCacheService;
import com.bigdata.journal.AbstractJournal.ISnapshotData;
import com.bigdata.quorum.Quorum;
import com.bigdata.quorum.QuorumException;
import com.bigdata.rawstore.IRawStore;
import com.bigdata.util.ChecksumError;
/**
* Disk-based Write Once Read Many (WORM) journal strategy. The phsyical layout
* on the disk is the journal header, the root blocks, and then the user extent.
*
* Writes are buffered in a write cache. The cache is flushed when it would
* overflow. As a result only large sequential writes are performed on the
* store. Reads read through the write cache for consistency.
*
* Note: This is used to realize both the {@link BufferMode#Disk} and the
* {@link BufferMode#Temporary} {@link BufferMode}s. When configured for the
* {@link BufferMode#Temporary} mode: the root blocks will not be written onto
* the disk, writes will not be forced, and the backing file will be created the
* first time the {@link DiskOnlyStrategy} attempts to write through to the
* disk. For many scenarios, the backing file will never be created unless the
* write cache overflows. This provides very low latency on start-up, the same
* MRMW capability, and allows very large temporary stores.
*
* FIXME Examine behavior when write caching is enabled/disabled for the OS.
* This has a profound impact. Asynchronous writes of multiple buffers, and the
* use of smaller buffers, may be absolutely when the write cache is disabled.
* It may be that swapping sets in because the Windows write cache is being
* overworked, in which case doing incremental and async IO would help. Compare
* with behavior on server platforms. See
* http://support.microsoft.com/kb/259716,
* http://www.accucadd.com/TechNotes/Cache/WriteBehindCache.htm,
* http://msdn2.microsoft.com/en-us/library/aa365165.aspx,
* http://www.jasonbrome.com/blog/archives/2004/04/03/writecache_enabled.html,
* http://support.microsoft.com/kb/811392,
* http://mail-archives.apache.org/mod_mbox
* /db-derby-dev/200609.mbox/%[email protected]%3E
*
*
* /sbin/hdparm -W 0 /dev/hda 0 Disable write caching
* /sbin/hdparm -W 1 /dev/hda 1 Enable write caching
*
*
* @todo report whether or not the on-disk write cache is enabled for each
* platform in {@link AbstractStatisticsCollector}. offer guidance on how
* to disable that write cache.
*
* @todo test verifying that the write cache can be disabled.
*
* @todo test verifying that the buffer position and limit are updated correctly
* by {@link #write(ByteBuffer)} regardless of the code path.
*
* @todo Retro fit the concept of a write cache into the
* {@link DirectBufferStrategy} so that we defer writes onto the disk
* until (a) a threshold of data has been buffered; or (b)
* {@link #force(boolean)} is invoked. Note that the implementation will
* be a bit different since the Direct mode is already fully buffered so
* we do not need to allocate a separate writeCache. However, we will
* still need to track the {@link #writeCacheOffset} and maintain a
* {@link #writeCacheIndex}.
*
* @author Bryan Thompson
* @version $Id$
*
* @see BufferMode#Disk
* @see BufferMode#Temporary
*/
public class WORMStrategy extends AbstractBufferStrategy implements
IDiskBasedStrategy, IHABufferStrategy, IBackingReader {
/**
* The file.
*/
private final File file;
/**
* The mode used to open that file.
*/
private final String fileMode;
/**
* true iff configured as a {@link BufferMode#Temporary} store.
*/
private final boolean temporaryStore;
/**
* The backing file for a {@link BufferMode#Temporary} store is not opened
* until the {@link #writeCache} is flushed to disk for the first time. In
* these scenarios this field will be false until the
* {@link #writeCache} is flushed and true thereafter. For
* {@link BufferMode#Disk}, this field is initially true.
*
* The value of this field determines the behavior of
* {@link #reopenChannel()}.
*/
private boolean fileOpened;
/**
* The IO interface for the file - use
* {@link #getRandomAccessFile()} rather than this field.
*/
private /*final*/ volatile RandomAccessFile raf;
/**
* The size of the journal header, including MAGIC, version, and both root
* blocks. This is used as an offset when computing the address of a record
* in an underlying file and is ignored by buffer modes that are not backed
* by a file (e.g., transient) or that are memory mapped (since the map is
* setup to skip over the header)
*/
private final int headerSize;
/**
* Extent of the file. This value should be valid since we obtain an
* exclusive lock on the file when we open it.
*
* @todo Atomic long to ensure visibility of changes?
*/
private long extent;
private long userExtent;
private final long minimumExtension;
private final Quorum quorum;
// private final AtomicReference> quorumRef;
/**
* This lock is used to exclude readers/writers performing IOs against the
* backing file when the extent of the backing file is about to be changed.
* Readers and writers take the {@link ReadLock}. The {@link WriteLock} is
* taken when the file extent must be changed. This is a workaround for an
* old (an unresolved as of February 2010) Sun bug.
*
* @see http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6371642
*/
final private ReentrantReadWriteLock extensionLock = new ReentrantReadWriteLock();
/**
* The service responsible for migrating dirty records onto the backing file
* and (for HA) onto the other members of the {@link Quorum}.
*
* This MAY be null for a read-only store or if the write cache
* is disabled. It is required for HA.
*
* TODO This should not really be volatile. For HA, we wind up needing to
* set a new value on this field in {@link #abort()}. The field used to
* be final. Perhaps an {@link AtomicReference} would be appropriate now?
*/
private volatile WORMWriteCacheService writeCacheService;
@Override
public WORMWriteCacheService getWriteCacheService() {
return writeCacheService;
}
/**
* true iff the backing store has record level checksums.
*/
private final boolean useChecksums;
/**
* The #of write cache buffers to use.
*
* @see FileMetadata#writeCacheBufferCount
*/
private final int writeCacheBufferCount;
/**
* The #of read cache buffers to use.
*
* @see com.bigdata.journal.Options#READ_CACHE_BUFFER_COUNT
*/
private final int readCacheBufferCount;
/**
* The threshold at which readCache records are moved to the
* hotCache
*
* @see com.bigdata.journal.Options#HOT_CACHE_THRESHOLD
*/
private final int hotCacheThreshold;
/**
* The number of hotCache buffers
*
* @see com.bigdata.journal.Options#HOT_CACHE_SIZE
*/
private final int hotCacheSize;
/**
* The key for the {@link CompressorRegistry} which identifies the
* {@link IRecordCompressor} to be applied (optional).
*
* @see com.bigdata.journal.Options#HALOG_COMPRESSOR
*/
private final String compressorKey;
/**
* true if the backing store will be used in an HA
* {@link Quorum} (this is passed through to the {@link WriteCache} objects
* which use this flag to conditionally track the checksum of the entire
* write cache buffer).
*/
private final boolean isQuorumUsed;
/**
* The {@link UUID} which identifies the journal (this is the same for each
* replicated journal is a quorum, so it is really a logical store UUID).
*
* Note: This can not be final since it is potentially changed (in HA) when
* we install new root blocks onto a store.
*/
private final AtomicReference storeUUIDRef = new AtomicReference();
@Override
public boolean useChecksums() {
return useChecksums;
}
/**
* Issues the disk writes for the write cache and recycles the write cache
* to receive new writes.
*/
private void flushWriteCache() {
if (writeCacheService != null) {
try {
/*
* Issue the disk writes (does not force to the disk).
*
* Note: This will wind up calling writeOnDisk().
*
* Note: It is critical that this operation is atomic with
* regard to writes on the cache. Otherwise new writes can enter
* the cache after it was flushed to the backing channel but
* before it is reset. Those writes will then be lost. This
* issue does not arise for the {@link WriteCacheService} since
* it atomically moves the full buffer onto a dirty list.
*/
writeCacheService.flush(false/* force */);
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
}
final public int getHeaderSize() {
return headerSize;
}
final public File getFile() {
return file;
}
/**
* Note: This MAY be null. If {@link BufferMode#Temporary}
* is used then it WILL be null until the {@link #writeCache}
* is flushed to disk for the first time.
*/
final public RandomAccessFile getRandomAccessFile() {
return raf;
}
/**
* Note: This MAY be null. If {@link BufferMode#Temporary}
* is used then it WILL be null until the {@link #writeCache}
* is flushed to disk for the first time.
*/
final public FileChannel getChannel() {
final RandomAccessFile raf = getRandomAccessFile();
if (raf == null)
return null;
return raf.getChannel();
}
/**
* Striped performance counters for {@link IRawStore} access, including
* operations that read or write through to the underlying media.
*
* @author Bryan
* Thompson
* @param
*
* @todo report elapsed time and average latency for force, reopen, and
* writeRootBlock.
*
* FIXME CAT may be much faster than striped locks (2-3x faster).
*/
static public class StoreCounters> extends
StripedCounters {
/**
* #of read requests.
*/
public volatile long nreads;
/**
* #of read requests that read through to the backing file.
*/
public volatile long ndiskRead;
/**
* #of bytes read.
*/
public volatile long bytesRead;
/**
* #of bytes that have been read from the disk.
*/
public volatile long bytesReadFromDisk;
/**
* Total elapsed time for reads.
*/
public volatile long elapsedReadNanos;
/**
* Total elapsed time for reading on the disk.
*/
public volatile long elapsedDiskReadNanos;
/**
* The #of checksum errors while reading on the local disk.
*/
public volatile long checksumErrorCount;
/**
* #of write requests.
*/
public volatile long nwrites;
/**
* #of write requests that write through to the backing file.
*/
public volatile long ndiskWrite;
/**
* The size of the largest record read.
*/
public volatile long maxReadSize;
/**
* The size of the largest record written.
*/
public volatile long maxWriteSize;
/**
* #of bytes written.
*/
public volatile long bytesWritten;
/**
* #of bytes that have been written on the disk.
*/
public volatile long bytesWrittenOnDisk;
/**
* Total elapsed time for writes.
*/
public volatile long elapsedWriteNanos;
/**
* Total elapsed time for writing on the disk.
*/
public volatile long elapsedDiskWriteNanos;
/**
* #of times the data were forced to the disk.
*/
public volatile long nforce;
/**
* #of times the length of the file was changed (typically, extended).
*/
public volatile long ntruncate;
/**
* #of times the file has been reopened after it was closed by an
* interrupt.
*/
public volatile long nreopen;
/**
* #of times one of the root blocks has been written.
*/
public volatile long nwriteRootBlock;
/**
* {@inheritDoc}
*/
public StoreCounters() {
super();
}
/**
* {@inheritDoc}
*/
public StoreCounters(final int batchSize) {
super(batchSize);
}
/**
* {@inheritDoc}
*/
public StoreCounters(final int nstripes, final int batchSize) {
super(nstripes, batchSize);
}
@Override
public void add(final T o) {
super.add(o);
nreads += o.nreads;
ndiskRead += o.ndiskRead;
bytesRead += o.bytesRead;
bytesReadFromDisk += o.bytesReadFromDisk;
maxReadSize = Math.max(maxReadSize, o.maxReadSize);
elapsedReadNanos += o.elapsedReadNanos;
elapsedDiskReadNanos += o.elapsedDiskReadNanos;
checksumErrorCount += o.checksumErrorCount;
nwrites += o.nwrites;
ndiskWrite += o.ndiskWrite;
maxWriteSize = Math.max(maxWriteSize, o.maxWriteSize);
bytesWritten += o.bytesWritten;
bytesWrittenOnDisk += o.bytesWrittenOnDisk;
elapsedWriteNanos += o.elapsedWriteNanos;
elapsedDiskWriteNanos += o.elapsedDiskWriteNanos;
nforce += o.nforce;
ntruncate += o.ntruncate;
nreopen += o.nreopen;
nwriteRootBlock += o.nwriteRootBlock;
}
@Override
public T subtract(final T o) {
// make a copy of the current counters.
final T t = super.subtract(o);
// subtract out the given counters.
t.nreads -= o.nreads;
t.ndiskRead -= o.ndiskRead;
t.bytesRead -= o.bytesRead;
t.bytesReadFromDisk -= o.bytesReadFromDisk;
t.maxReadSize -= o.maxReadSize; // @todo report max? min?
t.elapsedReadNanos -= o.elapsedReadNanos;
t.elapsedDiskReadNanos -= o.elapsedDiskReadNanos;
t.checksumErrorCount -= o.checksumErrorCount;
t.nwrites -= o.nwrites;
t.ndiskWrite -= o.ndiskWrite;
t.maxWriteSize -= o.maxWriteSize; // @todo report max? min?
t.bytesWritten -= o.bytesWritten;
t.bytesWrittenOnDisk -= o.bytesWrittenOnDisk;
t.elapsedWriteNanos -= o.elapsedWriteNanos;
t.elapsedDiskWriteNanos -= o.elapsedDiskWriteNanos;
t.nforce -= o.nforce;
t.ntruncate -= o.ntruncate;
t.nreopen -= o.nreopen;
t.nwriteRootBlock -= o.nwriteRootBlock;
return t;
}
@Override
public void clear() {
// subtract out the given counters.
nreads = 0;
ndiskRead = 0;
bytesRead = 0;
bytesReadFromDisk = 0;
maxReadSize = 0;
elapsedReadNanos = 0;
elapsedDiskReadNanos = 0;
checksumErrorCount = 0;
nwrites = 0;
ndiskWrite = 0;
maxWriteSize = 0;
bytesWritten = 0;
bytesWrittenOnDisk = 0;
elapsedWriteNanos = 0;
elapsedDiskWriteNanos = 0;
nforce = 0;
ntruncate = 0;
nreopen = 0;
nwriteRootBlock = 0;
}
@Override
public CounterSet getCounters() {
final CounterSet root = super.getCounters();
// IRawStore API
{
/*
* reads
*/
root.addCounter("nreads", new Instrument() {
public void sample() {
setValue(nreads);
}
});
root.addCounter("bytesRead", new Instrument() {
public void sample() {
setValue(bytesRead);
}
});
root.addCounter("readSecs", new Instrument() {
public void sample() {
final double elapsedReadSecs = (elapsedReadNanos / 1000000000.);
setValue(elapsedReadSecs);
}
});
root.addCounter("bytesReadPerSec", new Instrument() {
public void sample() {
final double readSecs = (elapsedReadNanos / 1000000000.);
final double bytesReadPerSec = (readSecs == 0L ? 0d
: (bytesRead / readSecs));
setValue(bytesReadPerSec);
}
});
root.addCounter("maxReadSize", new Instrument() {
public void sample() {
setValue(maxReadSize);
}
});
root.addCounter("checksumErrorCount", new Instrument() {
public void sample() {
setValue(checksumErrorCount);
}
});
/*
* writes
*/
root.addCounter("nwrites", new Instrument() {
public void sample() {
setValue(nwrites);
}
});
root.addCounter("bytesWritten", new Instrument() {
public void sample() {
setValue(bytesWritten);
}
});
root.addCounter("writeSecs", new Instrument() {
public void sample() {
final double writeSecs = (elapsedWriteNanos / 1000000000.);
setValue(writeSecs);
}
});
root.addCounter("bytesWrittenPerSec", new Instrument() {
public void sample() {
final double writeSecs = (elapsedWriteNanos / 1000000000.);
final double bytesWrittenPerSec = (writeSecs == 0L ? 0d
: (bytesWritten / writeSecs));
setValue(bytesWrittenPerSec);
}
});
root.addCounter("maxWriteSize", new Instrument() {
public void sample() {
setValue(maxWriteSize);
}
});
} // IRawStore
// disk statistics
{
final CounterSet disk = root.makePath("disk");
/*
* read
*/
disk.addCounter("nreads", new Instrument() {
public void sample() {
setValue(ndiskRead);
}
});
disk.addCounter("bytesRead", new Instrument() {
public void sample() {
setValue(bytesReadFromDisk);
}
});
disk.addCounter("bytesPerRead", new Instrument() {
public void sample() {
final double bytesPerDiskRead = (ndiskRead == 0 ? 0d
: (bytesReadFromDisk / (double) ndiskRead));
setValue(bytesPerDiskRead);
}
});
disk.addCounter("readSecs", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
setValue(diskReadSecs);
}
});
disk.addCounter("bytesReadPerSec", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
final double bytesReadPerSec = (diskReadSecs == 0L ? 0d
: bytesReadFromDisk / diskReadSecs);
setValue(bytesReadPerSec);
}
});
disk.addCounter("secsPerRead", new Instrument() {
public void sample() {
final double diskReadSecs = (elapsedDiskReadNanos / 1000000000.);
final double readLatency = (diskReadSecs == 0 ? 0d
: diskReadSecs / ndiskRead);
setValue(readLatency);
}
});
/*
* write
*/
disk.addCounter("nwrites", new Instrument() {
public void sample() {
setValue(ndiskWrite);
}
});
disk.addCounter("bytesWritten", new Instrument() {
public void sample() {
setValue(bytesWrittenOnDisk);
}
});
disk.addCounter("bytesPerWrite", new Instrument() {
public void sample() {
final double bytesPerDiskWrite = (ndiskWrite == 0 ? 0d
: (bytesWrittenOnDisk / (double) ndiskWrite));
setValue(bytesPerDiskWrite);
}
});
disk.addCounter("writeSecs", new Instrument() {
public void sample() {
final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
setValue(diskWriteSecs);
}
});
disk.addCounter("bytesWrittenPerSec", new Instrument() {
public void sample() {
final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
final double bytesWrittenPerSec = (diskWriteSecs == 0L ? 0d
: bytesWrittenOnDisk / diskWriteSecs);
setValue(bytesWrittenPerSec);
}
});
disk.addCounter("secsPerWrite", new Instrument() {
public void sample() {
final double diskWriteSecs = (elapsedDiskWriteNanos / 1000000000.);
final double writeLatency = (diskWriteSecs == 0 ? 0d
: diskWriteSecs / ndiskWrite);
setValue(writeLatency);
}
});
/*
* other
*/
disk.addCounter("nforce", new Instrument() {
public void sample() {
setValue(nforce);
}
});
disk.addCounter("nextend", new Instrument() {
public void sample() {
setValue(ntruncate);
}
});
disk.addCounter("nreopen", new Instrument() {
public void sample() {
setValue(nreopen);
}
});
disk.addCounter("rootBlockWrites", new Instrument() {
public void sample() {
setValue(nwriteRootBlock);
}
});
} // disk
return root;
} // getCounters()
} // class StoreCounters
/**
* Striped performance counters for this class.
*/
private final AtomicReference storeCounters = new AtomicReference();
/**
* Returns the striped performance counters for the store.
*/
public StoreCounters getStoreCounters() {
return storeCounters.get();
}
/**
* Replaces the {@link StoreCounters} object.
*
* @param storeCounters
* The new {@link Counter}s.
*
* @throws IllegalArgumentException
* if the argument is null.
*/
public void setStoreCounters(final StoreCounters storeCounters) {
if (storeCounters == null)
throw new IllegalArgumentException();
this.storeCounters.set(storeCounters);
}
/**
* Return interesting information about the write cache and file operations.
*/
public CounterSet getCounters() {
final CounterSet root = new CounterSet();
root.addCounter("nextOffset", new Instrument() {
public void sample() {
setValue(nextOffset.get());
}
});
root.addCounter("extent", new Instrument() {
public void sample() {
setValue(extent);
}
});
// attach the most recently updated values from the striped counters.
root.attach(storeCounters.get().getCounters());
if (writeCacheService != null) {
final CounterSet tmp = root.makePath("writeCache");
tmp.attach(writeCacheService.getCounters());
}
return root;
}
/**
*
* @param maximumExtent
* @param fileMetadata
*/
WORMStrategy(final long maximumExtent, final long minimumExtension,
final FileMetadata fileMetadata,
final Quorum quorum) {
super(
// @see https://sourceforge.net/apps/trac/bigdata/ticket/236
Long.valueOf(fileMetadata.getProperty(Options.INITIAL_EXTENT,
Options.DEFAULT_INITIAL_EXTENT)),
// fileMetadata.extent,
maximumExtent, fileMetadata.offsetBits,
fileMetadata.nextOffset, fileMetadata.getBufferMode(),
fileMetadata.readOnly);
this.file = fileMetadata.file;
this.fileMode = fileMetadata.fileMode;
this.temporaryStore = (fileMetadata.getBufferMode()==BufferMode.Temporary);
this.raf = fileMetadata.raf;
this.fileOpened = raf != null;
if (!temporaryStore && !fileOpened) {
throw new RuntimeException(
"File not open and not a temporary store");
}
this.extent = fileMetadata.extent;
this.headerSize = FileMetadata.headerSize0;
this.userExtent = extent - headerSize;
this.minimumExtension = minimumExtension;
this.quorum = quorum;
this.useChecksums = fileMetadata.useChecksums;
this.storeUUIDRef.set(fileMetadata.rootBlock.getUUID());
// initialize striped performance counters for this store.
this.storeCounters.set(new StoreCounters(10/* batchSize */));
/*
* Enable the write cache?
*
* Note: Do NOT enable the write cache if the file is being opened in a
* read-only mode.
*
* Note: If the file has been closed for writes (closeTime != 0L), then
* the file is read-only regardless of the mode in which it was opened.
*
* Note: NIO always (at least up to Java 6) allocates a "temporary"
* direct byte buffer for disk read/write operations on a heap buffer
* AND there is a bug in the release of those buffers. Therefore do NOT
* pass in a heap byte buffer for the write cache!!!
*
* Note: HA MUST use a write cache service (the write cache service
* handles the write pipeline to the downstream quorum members).
*/
// final Quorum quorum = quorumRef.get();
this.writeCacheBufferCount = fileMetadata.writeCacheBufferCount;
this.readCacheBufferCount = Integer.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.READ_CACHE_BUFFER_COUNT,
com.bigdata.journal.Options.DEFAULT_READ_CACHE_BUFFER_COUNT));
this.hotCacheThreshold = Integer.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.HOT_CACHE_THRESHOLD,
com.bigdata.journal.Options.DEFAULT_HOT_CACHE_THRESHOLD));
this.hotCacheSize = Integer.valueOf(fileMetadata.getProperty(
com.bigdata.journal.Options.HOT_CACHE_SIZE,
com.bigdata.journal.Options.DEFAULT_HOT_CACHE_SIZE));
this.compressorKey = fileMetadata.getProperty(
com.bigdata.journal.Options.HALOG_COMPRESSOR,
com.bigdata.journal.Options.DEFAULT_HALOG_COMPRESSOR);
isQuorumUsed = quorum != null; // && quorum.isHighlyAvailable();
final boolean useWriteCacheService = fileMetadata.writeCacheEnabled
&& !fileMetadata.readOnly && fileMetadata.closeTime == 0L
|| isQuorumUsed;
if (useWriteCacheService) {
/*
* WriteCacheService.
*/
this.writeCacheService = newWriteCacheService();
this._checkbuf = null;
} else {
this.writeCacheService = null;
this._checkbuf = useChecksums ? ByteBuffer.allocateDirect(4) : null;
}
}
private WORMWriteCacheService newWriteCacheService() {
try {
return new WORMWriteCacheService(writeCacheBufferCount,
useChecksums, extent, opener, quorum);
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
private class WORMWriteCacheService extends WriteCacheService {
WORMWriteCacheService(final int nbuffers, final boolean useChecksum,
final long fileExtent,
final IReopenChannel opener,
final Quorum quorum) throws InterruptedException {
// Note: Compaction explicitly disabled for the WORM.
super(writeCacheBufferCount, 0/* minCleanListSize */, readCacheBufferCount,
false/* prefixWrites */, 100/* compactionThreshold */, hotCacheSize, hotCacheThreshold,
useChecksums, extent, opener, quorum, WORMStrategy.this /*reader*/);
}
@Override
public WriteCacheImpl newWriteCache(final IBufferAccess buf,
final boolean useChecksum, final boolean bufferHasData,
final IReopenChannel opener,
final long fileExtent)
throws InterruptedException {
return new WriteCacheImpl(0/* baseOffset */, buf, useChecksum,
bufferHasData, (IReopenChannel) opener,
fileExtent);
}
}
/**
* Implementation coordinates writes using the read lock of the
* {@link DiskOnlyStrategy#extensionLock}. This is necessary in order to
* avoid corrupt reads or writes with concurrent changes to the file size.
*
* @author Bryan
* Thompson
*/
private class WriteCacheImpl extends WriteCache.FileChannelWriteCache {
public WriteCacheImpl(final long baseOffset, final IBufferAccess buf,
final boolean useChecksum,
final boolean bufferHasData,
final IReopenChannel opener,
final long fileExtent)
throws InterruptedException {
super(baseOffset, buf, useChecksum, isQuorumUsed,
bufferHasData, opener, fileExtent);
}
@Override
protected String getCompressorKey() {
return compressorKey;
}
/**
* {@inheritDoc}
*
* Overridden to expose this method to the {@link WORMStrategy} class.
*/
@Override
protected void setFirstOffset(final long firstOffset) {
super.setFirstOffset(firstOffset);
}
@Override
protected boolean writeOnChannel(final ByteBuffer data,
final long firstOffset,
final Map recordMapIsIgnored,
final long nanos) throws InterruptedException, IOException {
final long begin = System.nanoTime();
long remaining = nanos;
final Lock readLock = extensionLock.readLock();
if (!readLock.tryLock(remaining, TimeUnit.NANOSECONDS)) {
return false;
}
try {
remaining = nanos - (System.nanoTime() - begin);
final int dpos = data.position();
final int nbytes = data.remaining();
/*
* Note: We are holding the readLock (above). This is Ok since
* file extension occurs when the record is accepted for write
* while only the readLock is required to actually write on the
* file.
*/
final int nwrites = writeOnDisk(data, firstOffset);
final WriteCacheCounters counters = this.counters.get();
counters.nchannelWrite += nwrites;
counters.bytesWritten += nbytes;
counters.elapsedWriteNanos += (System.nanoTime() - begin);
if (WriteCache.log.isTraceEnabled()) {
WriteCache.log.trace("wroteOnDisk: dpos=" + dpos
+ ", nbytes=" + nbytes + ", firstOffset="
+ firstOffset + ", nrecords="
+ recordMapIsIgnored.size());
}
return true;
} finally {
readLock.unlock();
}
}
}
final public boolean isStable() {
return true;
}
public boolean isFullyBuffered() {
return false;
}
public void force(final boolean metadata) {
assertOpen();
try {
// flush all pending writes to disk.
flushWriteCache();
if (!temporaryStore) {
// sync the disk.
getChannel().force(metadata);
}
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nforce++;
} finally {
c.release();
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
}
/**
* {@inheritDoc}
*
* This implementation flushes the write cache (if enabled).
*/
@Override
public void commit() {
flushWriteCache();
if (writeCacheService != null) {
// Reset the write cache block counter.
lastBlockSequence = writeCacheService.resetSequence();
}
super.commit();
}
@Override
public long getBlockSequence() {
return lastBlockSequence;
}
/**
* The sequence number for the last {@link WriteCache} block written in the
* current commit. This is set in {@link #commit()} when we flush the
* {@link WriteCacheService}. This is used to prepare the new
* {@link IRootBlockView}. The value of this field is NOT incremented as
* writes are made, but only after they have been flushed.
*
* TODO This pattern arose because {@link WriteCacheService#resetSequence()}
* is called from {@link #commit()}. It might be more pleasant to call that
* from {@link AbstractJournal#commitNow(long)} so we could consolidate this
* logic. That requires us to expose the {@link WriteCacheService} to the
* {@link AbstractJournal} through the {@link IHABufferStrategy}.
*/
private long lastBlockSequence = 0;
@Override
public long getCurrentBlockSequence() {
final WriteCacheService tmp = writeCacheService;
if(tmp == null) {
/*
* Either this is not an HA strategy mode -or- we are in abort() and
* the value temporarily [null]. If there is an abort(), then the
* counter will be reset to 0L.
*/
return 0L;
}
return tmp.getSequence();
}
/**
* Resets the {@link WriteCacheService} (if enabled).
*
* Note: This assumes the caller is synchronized appropriately otherwise
* writes belonging to other threads will be discarded from the cache!
*/
@Override
public void abort() {
super.abort();
if (writeCacheService != null) {
try {
if (quorum != null) {
/**
* When the WORMStrategy is part of an HA quorum, we need to
* close out and then reopen the WriteCacheService every
* time the quorum token is changed. For convenience, this
* is handled by extending the semantics of abort() on the
* Journal and reset() on the WORMStrategy.
*
* @see
* HA Journal
*/
writeCacheService.close();
writeCacheService = newWriteCacheService();
} else {
writeCacheService.reset();
writeCacheService.setExtent(extent);
}
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
}
/**
* Closes the file immediately (without flushing any pending writes).
*
* @todo This is synchronized solely to coordinate the service shutdown
* state, which SHOULD use a different lock.
*/
synchronized public void close() {
/*
* Note: this clears the [open] flag. It is important to do this first
* so that we do not re-open the channel once it has been closed.
*/
super.close();
// Release the write cache.
releaseWriteCache();
try {
if (raf != null) {
synchronized (opener) {
if (raf != null && raf.getChannel().isOpen()) {
raf.close();
}
}
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
if (!bufferMode.isStable() && file.exists()) {
if (!file.delete()) {
log.warn("Unable to delete temporary file: " + file);
}
}
}
public void deleteResources() {
if (isOpen()) {
throw new IllegalStateException();
}
if (fileOpened && file.exists() && !file.delete()) {
log.warn("Could not delete file: " + file.getAbsoluteFile());
}
}
final public long getExtent() {
return extent;
}
final public long getUserExtent() {
return userExtent;
}
/**
* Extended to handle {@link ChecksumError}s by reading on another node when
* the {@link Quorum} (iff the quorum is highly available).
*
* {@inheritDoc}
*
* @todo hook for monitoring (nagios, etc). bad reads indicate a problem
* with the disk which should be tracked over time.
*
* @todo If we see a read error from a checksum and want to update the
* record on the backing file then we would have to go around the
* write cache to do a direct disk write since (at least for the WORM)
* the assumption is pure append for the write cache.
*
* An attempt to overwrite a bad record on the disk could itself be a
* bad idea. If it was just a high write, then it might be Ok. But
* many other kinds of errors are likely to have long pauses while the
* OS attempts to get a good read/write from the file system.
*
* @todo We might want to maintain a set of known bad records and fail the
* node when the size of that set grows too large. That would also
* help us to avoid "hanging" on a bad read when we know that we have
* to get the data from another node based on past experience for that
* record.
*/
@Override
public ByteBuffer read(final long addr) {
try {
// Try reading from the local store.
final ByteBuffer ret = readFromLocalStore(addr);
return ret; // ret != null ? ret.duplicate() : null;
} 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) storeCounters.get()
.acquire();
try {
c.checksumErrorCount++;
} finally {
c.release();
}
// final Quorum quorum = quorumRef.get();
if (quorum != null && quorum.isHighlyAvailable()) {
if (quorum.isQuorumMet()) {
try {
// Read on another node in the quorum.
final byte[] a = ((QuorumRead) quorum.getMember())
.readFromQuorum(storeUUIDRef.get(), addr);
return ByteBuffer.wrap(a);
} catch (Throwable t) {
throw new RuntimeException("While handling: " + e, t);
}
}
}
// Otherwise rethrow the checksum error.
throw e;
}
}
/**
* {@inheritDoc}
*
* This implementation tests the {@link WriteCacheService} first
* and then reads through to the local disk on a cache miss. This is
* automatically invoked by {@link #read(long)}.
*/
public ByteBuffer readFromLocalStore(final long addr)
throws InterruptedException {
final long begin = System.nanoTime();
if (addr == 0L)
throw new IllegalArgumentException(ERR_ADDRESS_IS_NULL);
final long offset = getOffset(addr);
final long paddr = offset2PhysicalAddress(offset);
final int nbytes = getByteCount(addr);
if (nbytes == 0) {
throw new IllegalArgumentException(ERR_RECORD_LENGTH_ZERO);
}
if (offset + nbytes > nextOffset.get()) {
throw new IllegalArgumentException(ERR_ADDRESS_NOT_WRITTEN);
}
{
final StoreCounters storeCounters = (StoreCounters) this.storeCounters
.get().acquire();
try {
if (nbytes > storeCounters.maxReadSize) {
storeCounters.maxReadSize = nbytes;
}
} finally {
storeCounters.release();
}
}
if (writeCacheService != null) {
/*
* Test the write cache for a hit. The WriteCacheService handles
* synchronization internally.
*
* Note: WriteCacheService#read(long) DOES NOT throw an
* IllegalStateException for an asynchronous close. However, it will
* throw a RuntimeException if there is a checksum error on the
* record.
*/
// Note: Can throw ChecksumError, InterruptedException
ByteBuffer tmp = writeCacheService.read(paddr, nbytes);
if (tmp != null) {
/*
* Hit on the write cache.
*
* Update the store counters.
*/
final StoreCounters c = (StoreCounters) storeCounters
.get().acquire();
try {
c.nreads++;
c.bytesRead += nbytes;
c.elapsedReadNanos += (System.nanoTime() - begin);
} finally {
c.release();
}
// if (log.isTraceEnabled())
// log.trace("cacheRead: addr=" + toString(addr));
return tmp;
}
} // if(writeCacheService!=null)
/*
* Read through to the disk.
*
* Note: Strip off the checksum from the end of the record and validate
* it.
*/
{
final long beginDisk = System.nanoTime();
// Allocate a new buffer of the exact capacity.
final ByteBuffer dst = ByteBuffer.allocate(nbytes);
// Read through to the disk.
readRaw(/* nbytes, */paddr, dst);
if (useChecksums) {
// extract the checksum.
final int chk = dst.getInt(nbytes - 4);
// adjust the record length to exclude the checksum.
dst.limit(nbytes - 4);
if (chk != ChecksumUtility.threadChk.get().checksum(dst)) {
throw new ChecksumError("address=" + paddr + ", nbytes="
+ nbytes);
}
}
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nreads++;
c.bytesRead += nbytes;
c.bytesReadFromDisk += nbytes;
c.elapsedReadNanos += (System.nanoTime() - begin);
c.elapsedDiskReadNanos += (System.nanoTime() - beginDisk);
} finally {
c.release();
}
if (log.isTraceEnabled())
log.trace("diskRead: addr=" + toString(addr));
// return the buffer.
return dst;
}
}
/**
* Adjusts the offset by the headerSize, such that writing to a zero offset
* would not corrupt the header.
*
* @param offset - the WORMStore address
* @return the physical address of the offset provided
*/
private long offset2PhysicalAddress(final long offset) {
return offset + headerSize;
}
/**
* Read on the backing file. {@link ByteBuffer#remaining()} bytes will be
* read into the caller's buffer, starting at the specified offset in the
* backing file.
*
* @param offset
* The offset of the first byte (now absolute, not relative to the
* start of the data region).
* @param dst
* Where to put the data. Bytes will be written at position until
* limit.
*
* @return The caller's buffer, prepared for reading.
*/
public ByteBuffer readRaw(final long offset, final ByteBuffer dst) {
final Lock readLock = extensionLock.readLock();
readLock.lock();
try {
final int startPos = dst.position();
try {
// the offset into the disk file.
// final long pos = headerSize + offset;
final long pos = offset; // offset is physical disk address
// read on the disk.
final int ndiskRead = FileChannelUtility.readAll(opener, dst,
pos);
// update performance counters.
final StoreCounters c = (StoreCounters) storeCounters
.get().acquire();
try {
c.ndiskRead += ndiskRead;
} finally {
c.release();
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
// Reset start position - do not flip()
dst.position(startPos);
return dst;
} finally {
readLock.unlock();
}
}
/**
* Used to re-open the {@link FileChannel} in this class.
*/
private final IReopenChannel opener = new IReopenChannel() {
public String toString() {
return file.toString();
}
public FileChannel reopenChannel() throws IOException {
return WORMStrategy.this.reopenChannel();
}
};
/**
* This method transparently re-opens the channel for the backing file.
*
* Note: This method is synchronized (internally) so that concurrent readers
* do not try to all open the store at the same time.
*
* Note: This method is invoked by both readers and writers. It WILL NOT be
* invoked by {@link FileChannelUtility} if the channel was closed by an
* interrupt in the current thread (a different exception is thrown).
*/
private FileChannel reopenChannel() throws IOException {
/*
* Note: This is basically a double-checked locking pattern. It is
* used to avoid synchronizing when the backing channel is already
* open.
*/
{
final RandomAccessFile tmp = raf;
if (tmp != null) {
final FileChannel channel = tmp.getChannel();
if (channel.isOpen()) {
// The channel is still open.
return channel;
}
}
}
synchronized (opener) {
assertOpen();
if (raf != null && raf.getChannel().isOpen()) {
/*
* The channel is still open. If you are allowing concurrent
* reads on the channel, then this could indicate that two
* readers each found the channel closed and that one was able
* to re-open the channel before the other such that the channel
* was open again by the time the 2nd reader got here.
*/
return raf.getChannel();
}
if (temporaryStore && !fileOpened) {
/*
* The backing file has not been opened.
*
* Note: Without this case this method would create the backing
* store for a Temporary store if anyone happened to invoke it.
* In fact, this method will never get invoked for a Temporary
* store without a backing store since the reads never read
* against the channel because it does not exist. So, really,
* this is just here to be paranoid.
*/
throw new AssertionError("TemporaryStore not yet open: " + file);
}
// open the file.
this.raf = new RandomAccessFile(file, fileMode);
if (log.isInfoEnabled())
log.info("(Re-)opened file: " + file);
try {
/*
* Request a shared file lock.
*/
final boolean readOnly = "r".equals(fileMode);
if (raf.getChannel()
.tryLock(0, Long.MAX_VALUE, readOnly/* shared */) == null) {
/*
* Note: A null return indicates that someone else holds the
* lock. This can happen if the platform does not support
* shared locks or if someone requested an exclusive file
* lock.
*/
try {
raf.close();
} catch (Throwable t) {
// ignore.
}
throw new IOException("File already locked? file=" + file);
}
} catch (IOException ex) {
/*
* Note: This is true of NFS volumes. This is Ok and should be
* ignored. However the backing file is not protected against
* accidental deletes or overwrites.
*/
if (log.isInfoEnabled())
log.info("FileLock not supported: file=" + file, ex);
}
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nreopen++;
} finally {
c.release();
}
return raf.getChannel();
}
}
/**
* Note: Synchronization is provided by the caller (
* {@link #write(ByteBuffer)}
*
* @param nbytes
* The #of bytes in the record.
*
* @return The address at which that record may be written.
*/
private long allocate(final int nbytes) {
// if (isReadOnly())
// throw new IllegalStateException(ERR_READ_ONLY);
//
// if (nbytes <= 0)
// throw new IllegalArgumentException(ERR_BAD_RECORD_SIZE);
// final long addr; // address in the store.
// synchronized(nextOffset) {
/*
* The offset at which the record will be written on the disk file
* (not adjusted for the root blocks).
*/
final long offset = nextOffset.get();
/*
* Make sure that the allocated region of the file exists (this will
* acquire the extensionLock's write lock in truncate(long) if it
* has to change the size of the file).
*/
overflow(offset, nbytes);
/*
* Formulate the address that can be used to recover that record.
*/
final long addr = toAddr(nbytes, offset);
/*
* Increment the offset of the next address to be assigned by the
* #of bytes in the record.
*/
nextOffset.addAndGet(nbytes);
// }
return addr;
}
public long write(final ByteBuffer data) {
if (data == null)
throw new IllegalArgumentException(ERR_BUFFER_NULL);
if (isReadOnly())
throw new IllegalStateException(ERR_READ_ONLY);
// #of bytes in the record.
final int remaining = data.remaining();
// #of bytes to write onto the file (includes the optional checksum).
final int nwrite = remaining + (useChecksums ? 4 : 0);
if (remaining == 0)
throw new IllegalArgumentException(ERR_BUFFER_EMPTY);
final long begin = System.nanoTime();
// final StoreCounters storeCounters = this.storeCounters.get();
// get checksum for the buffer contents.
final int chk = useChecksums ? ChecksumUtility.threadChk.get()
.checksum(data) : 0;
final long addr; // address in the store.
try {
/*
* The offset at which the record will be written on the disk file
* (not adjusted for the root blocks).
*/
final long offset;
synchronized (writeOnCacheLock) {
/*
* Allocate address for a new record with [nbytes] of data.
*
* Note: [writeOnCacheLock] imposes a total ordering over the
* allocations AND also over those records which get written
* through to the write cache.
*
* Note: Records written onto the cache for the WORM store must
* be single threaded in order to ensure that the records are
* laid out on the cache buffer in their allocation order.
* Otherwise they could be in a jumbled order. We could of
* course use the ordered write version of the cache to work
* around that, but writes on the cache should be very fast
* (especially if the cache is a service since it will hand us a
* new buffer if the old one is full).
*
* Note: We do not need to protect against concurrent close of
* the write cache since it is only closed by close() and
* closeForWrites() and the caller is responsible for ensuring
* that there are no concurrent writes during those methods
* (concurrent readers are allowed, but will be interrupted by
* close()).
*/
// Note: allocation must include the optional checksum.
addr = allocate(nwrite);
offset = getOffset(addr);
final long paddr = offset2PhysicalAddress(offset);
boolean wroteOnCache = false;
if (writeCacheService != null) {
if (!writeCacheService.write(paddr, data, chk))
throw new AssertionError();
wroteOnCache = true;
}
if (!wroteOnCache) {
/*
* The writeCache is disabled or the record is too large for
* the write cache, so just write the record directly on the
* disk.
*
* Note: At this point the backing file is already extended.
*
* Note: Unlike writes on the cache, the order in which we
* lay down this write onto the disk does not matter. We
* have already made the allocation and now the caller will
* block until the record is on the disk.
*/
final Lock readLock = extensionLock.readLock();
readLock.lock();
try {
writeOnDisk(data, paddr);
if (useChecksums) {
/*
* Note: If [useChecksums] is enabled but we are not
* using the WriteCacheService then we also need to
* write the checksum on the file here.
*/
final ByteBuffer b = _checkbuf;
b.clear();
b.putInt(chk);
b.flip();
writeOnDisk(b, paddr + remaining);
}
} finally {
readLock.unlock();
}
} // if(!wroteOnCache)
} // synchronized(writeOnCacheLock)
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nwrites++;
c.bytesWritten += nwrite;
c.elapsedWriteNanos += (System.nanoTime() - begin);
if (nwrite > c.maxWriteSize) {
c.maxWriteSize = nwrite;
}
} finally {
c.release();
}
} catch(InterruptedException ex) {
throw new RuntimeException(ex);
}
return addr;
}
/**
* An object used to serialize the allocation of record addresses and their
* writes onto the cache. See write() above for why. This is also relied on
* to make {@link #flushWriteCache()} atomic.
*/
private final Object writeOnCacheLock = new Object();
/**
* A small direct {@link ByteBuffer} used if we need to write the checksum
* on the backing file directly because the {@link WriteCacheService} is not
* in use.
*/
private final ByteBuffer _checkbuf;
// private HARebuildRequest m_rebuildRequest;
//
// private int m_rebuildSequence;
/**
* Make sure that the file is large enough to accept a write of
* nbytes starting at offset bytes into the file. This is only
* invoked from {@link #allocate(int)}, which is responsible for ensuring
* that the store file is large enough on the disk for any address which it
* returns to its caller. This way {@link #writeOnDisk(ByteBuffer, long)}
* does not have to check the extent on the disk.
*
* @param offset
* The offset into the file (NOT adjusted for the root blocks).
* @param nbytes
* The #of bytes to be written at that offset.
*/
private void overflow(final long offset, final int nbytes) {
final long needed = (offset + nbytes) - userExtent;
if (needed > 0) {
if (!overflow(needed)) {
throw new OverflowException();
}
}
}
/**
* Overridden to use the value specified to the constructor.
*/
@Override
protected long getMinimumExtension() {
return minimumExtension;
}
/**
* Create/open the backing file for a {@link BufferMode#Temporary} store iff
* it has not been created/opened.
*/
final private void createBackingFile() {
if (!fileOpened && temporaryStore) {
try {
// // open the file for the first time (create).
// raf = FileLockUtility.openFile(file, fileMode,
// bufferMode != BufferMode.Mapped/*useTryLock*/);
// note: set true so that reopenChannel will create the file.
fileOpened = true;
reopenChannel();
if (log.isInfoEnabled())
log.info("Opened backing file for temporary store: "
+ file);
} catch (IOException e) {
throw new RuntimeException("Could not open temp file: file="
+ file, e);
}
}
}
/**
* Write the data on the disk (synchronous).
*
* Note: The caller MUST hold either the read lock or the write lock on the
* {@link #extensionLock}. Normal writes should hold the read lock. File
* size changes MUST hold the write lock.
*
* @param data
* The data. The bytes from the current
* {@link ByteBuffer#position()} to the
* {@link ByteBuffer#limit()} will be written and the
* {@link ByteBuffer#position()} will be advanced to the
* {@link ByteBuffer#limit()} . The caller may subsequently
* modify the contents of the buffer without side effects (i.e.,
* the data are copied onto the disk).
* @param offset
* The offset in the user extent at which the data will be
* written (it is automatically adjusted for the size of the root
* blocks).
*
* @return The #of write operations against the disk.
*/
private int writeOnDisk(final ByteBuffer data, final long offset) {
assert offset >= 0 : "offset=" + offset;
// Thread MUST have either the read or write lock.
assert extensionLock.getReadHoldCount() > 0
|| extensionLock.getWriteHoldCount() > 0;
final long begin = System.nanoTime();
// final StoreCounters storeCounters = this.storeCounters.get();
createBackingFile();
final int nbytes = data.remaining();
/*
* The position in the file at which the record will be written
* (this is now the absolute position).
*/
// final long pos = headerSize + offset;
assert offset >= headerSize;
final long pos = offset;
final int nwrites;
try {
/*
* Write bytes in [data] from position to limit onto the channel.
*
* Note: Since readers ARE NOT synchronized it is possible for a
* reader is to be interrupted during a concurrent NIO operation and
* thus the channel MAY be asynchronously closed while we are
* writing on it, in which case we need to retry the write until it
* succeeds.
*/
nwrites = FileChannelUtility.writeAll(opener, data, pos);
// Update counters.
final long elapsed = (System.nanoTime() - begin);
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.ndiskWrite += nwrites;
c.bytesWrittenOnDisk += nbytes;
c.elapsedDiskWriteNanos += elapsed;
} finally {
c.release();
}
if (log.isTraceEnabled()) {
/*
* Note: There are only two places where the journal writes on the
* disk using this backing buffer implementation. Here and when it
* updates the root blocks. It only syncs the disk at the commit.
*/
log.trace("wrote on disk: bytes="
+ nbytes
+ ", elapsed="
+ TimeUnit.NANOSECONDS.toMillis(elapsed)
+ "ms; totals: write="
+ TimeUnit.NANOSECONDS
.toMillis(storeCounters.get().elapsedDiskWriteNanos)
+ "ms, read="
+ TimeUnit.NANOSECONDS
.toMillis(storeCounters.get().elapsedDiskReadNanos)
+ "ms");
}
return nwrites;
} catch (IOException ex) {
throw new RuntimeException(ex);
}
}
public ByteBuffer readRootBlock(final boolean rootBlock0) {
if(!isOpen()) throw new IllegalStateException();
final ByteBuffer tmp = ByteBuffer
.allocate(RootBlockView.SIZEOF_ROOT_BLOCK);
try {
FileChannelUtility.readAll(opener, tmp,
rootBlock0 ? FileMetadata.OFFSET_ROOT_BLOCK0
: FileMetadata.OFFSET_ROOT_BLOCK1);
tmp.position(0); // resets the position.
} catch (IOException ex) {
throw new RuntimeException(ex);
}
return tmp;
}
public void writeRootBlock(final IRootBlockView rootBlock,
final ForceEnum forceOnCommit) {
/*
* Note: Root blocks are written for a temporary store in support of
* rollback().
*/
// if(temporaryStore) {
//
// /*
// * Note: There are NO ROOT BLOCKS for a temporary store. Root blocks
// * are only useful for stores that can be re-opened, and you can not
// * re-open a temporary store - the backing file is always deleted
// * when the store is closed. The AbstractJournal still formats the
// * root blocks and retains a reference to the current root block,
// * but it is NOT written onto the file.
// */
//
// return;
//
// }
if (rootBlock == null)
throw new IllegalArgumentException();
try {
final ByteBuffer data = rootBlock.asReadOnlyBuffer();
final long pos = rootBlock.isRootBlock0()
? FileMetadata.OFFSET_ROOT_BLOCK0
: FileMetadata.OFFSET_ROOT_BLOCK1;
/*
* Note: This uses the [opener] to automatically retry the operation
* in case concurrent readers are interrupting, causing an
* asynchronous close of the backing channel.
*
* @todo Consider using the read lock vs the write lock of the
* extensionLock here. The advantage of the read lock is higher
* concurrency. The advantage of the write lock is that it locks out
* readers when we are writing the root blocks, which could help to
* ensure timely updates of the root blocks even if readers are
* behaving badly (lots of interrupts).
*
* FIXME Modify AbstractInterruptsTestCase to test for correct
* handling of root block writes where concurrent readers cause the
* backing store to be closed asynchronously. This code block SHOULD
* cause the root block write to eventually succeed.
*/
final Lock lock = extensionLock.readLock();
lock.lock();
try {
// Update the root block.
FileChannelUtility.writeAll(opener, data, pos);
if (!temporaryStore) {
/*
* Generally, you want to force the file data to the disk
* here. The file metadata MIGHT not matter since we always
* force it to the disk when we change the file size (unless
* the file system updates other aspects of file metadata
* during normal writes).
*
* @todo make sure the journal has already forced the
* writes, that forcing an empty cache buffer is a NOP, and
* that we want to just force the channel after we write the
* root blocks since writes were already forced on each node
* in the quorum before we wrote the root blocks and the
* root blocks are transmitted using RMI not the write
* pipeline.
*/
// sync the disk.
getChannel().force(forceOnCommit == ForceEnum.ForceMetadata);
// force(forceOnCommit == ForceEnum.ForceMetadata);
}
// Update counters.
final StoreCounters c = (StoreCounters) storeCounters.get()
.acquire();
try {
c.nwriteRootBlock++;
} finally {
c.release();
}
} finally {
lock.unlock();
}
} catch (IOException ex) {
throw new RuntimeException(ex);
}
if (log.isDebugEnabled())
log.debug("wrote root block: "+rootBlock);
}
@Override
public void truncate(final long newExtent) {
final long newUserExtent = newExtent - headerSize;
if (newUserExtent < getNextOffset()) {
throw new IllegalArgumentException(ERR_TRUNCATE);
}
if (newUserExtent == getUserExtent()) {
// NOP.
return;
}
final Lock writeLock = extensionLock.writeLock();
writeLock.lock();
try {
/*
* Note: This handles the case for a Temporary store where the write
* cache is the same size as the initial extent and everything
* written so far has been absorbed by the write cache.
*/
createBackingFile();
/*
* Extend (or truncate) the file.
*
* FIXME I could see how this might fail with a concurrent interrupt
* of a reader. This "extend" needs to be robust just writeAll() on
* FileChannelUtility. It must use the opener and retry if there is
* a ClosedByInterruptException. [See the notes below in the catch
* clause.]
*/
getRandomAccessFile().setLength(newExtent);
if (writeCacheService != null) {
/*
* Inform the write cache service that the file extent has
* changed. It will propagate this message along the write
* pipeline when HA is enabled.
*/
try {
writeCacheService.setExtent(newExtent);
} catch (InterruptedException t) {
throw new RuntimeException(t);
}
}
// Update fields and counters while holding the lock.
this.userExtent = newUserExtent;
this.extent = newExtent;
storeCounters.get().ntruncate++;
/*
* Since we just changed the file length we force the data to disk
* and update the file metadata. This is a relatively expensive
* operation but we want to make sure that we do not loose track of
* a change in the length of the file.
*
* @todo Per the link below, consider modifying this to force for a
* temporary store as well. At issue is whether the Java bug is
* linked to the update of the file metadata or just a concurrency
* issue. If it is only a concurrency issue, then there is no reason
* to force the file metadata for a temporary store.
*
* http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6371642
*/
if (!temporaryStore) {
/*
* We need to force the file data and metadata to the disk. When
* integrated with the WriteCacheService the FileChannel#force()
* request will be executed in a different thread and would
* deadlock unless we first release the WriteLock since
* writeOnChannel needs to acquire the ReadLock to proceed.
*
* We address this by doing acquiring the ReadLock (we are
* already holding the WriteLock so this will not block) and
* then releasing the WriteLock so other threads may now also
* acquire the ReadLock.
*
* Note: An alternative would be to directly invoke force(true)
* on the FileChannel.
*/
extensionLock.readLock().lock();
extensionLock.writeLock().unlock();
force(true/*metadata*/);
// opener.reopenChannel().force(true/*metadata*/);
}
if (log.isInfoEnabled())
log.info("newLength=" + cf.format(newExtent) + ", file="+ file);
if (log.isInfoEnabled())
log.info(getCounters().toString());
} catch (IOException ex) {
/*
* FIXME I've see an IOException "The handle is invalid" tossed here
* (just once). A bit of searching around suggests that perhaps the
* RandomAccessFile was concurrently closed? Something to look out
* for if it happens again. [@todo probably a concurrent reader was
* interrupted, in which case this method should just try the
* setLength() operation again.] [@todo the MRMW test can throw this
* during test shutdown, which simulates interrupt of NIO
* operations].
*/
throw new RuntimeException(ex);
} finally {
// writeLock.unlock();
extensionLock.readLock().unlock();
}
}
/*
* @todo why is this synchronized? the operation should be safe. maybe
* against a concurrent close?
*/
@Override
synchronized public long transferTo(final RandomAccessFile out)
throws IOException {
if (out == null)
throw new IllegalArgumentException();
/*
* Note: Force the write cache to the disk so that all the data we want
* to transfer from channel to channel are actually on the source
* channel!
*
* Note: This also handles the case for a Temporary store where the
* backing file has not even been created yet.
*/
flushWriteCache();
return super.transferFromDiskTo(this, out);
}
/**
* Extended to reset the write cache.
*
* Note: The file is NOT closed and re-opened in a read-only mode in order
* to avoid causing difficulties for concurrent readers.
*/
@Override
public void closeForWrites() {
// sets the [readOnly] flag.
super.closeForWrites();
// do not discard the write cache, just reset it to preserve
// read cache
// releaseWriteCache();
try {
writeCacheService.reset();
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
}
/**
* {@inheritDoc}
*
* Note: {@link #read(long)} has logic to handle the concurrent close of the
* {@link WriteCacheService}, passing through the operation to the disk.
*
* @todo Should this get invoked from {@link #closeForWrites()} for HA? If
* read failover is handled by the {@link WriteCacheService} then we
* can not close it out here.
*/
private final void releaseWriteCache() {
if (writeCacheService != null) {
writeCacheService.close();
}
}
/**
* This implementation can not release storage allocations and invocations
* of this method are ignored.
*/
@Override
public void delete(final long addr) {
if (writeCacheService != null) {
writeCacheService.clearWrite(addr, 0);
}
}
@Override
public void writeRawBuffer(final IHAWriteMessage msg, final IBufferAccess b)
throws IOException, InterruptedException {
// expand buffer before writing on the store.
final ByteBuffer xb = msg.expand(b.buffer());
if (true || log.isTraceEnabled()) {
log.warn("Buffer, position: " + xb.position()
+ ", limit: " + xb.limit());
}
final IBufferAccess ba = new IBufferAccess() {
@Override
public ByteBuffer buffer() {
return xb;
}
@Override
public void release() throws InterruptedException {
}
@Override
public void release(long timeout, TimeUnit unit)
throws InterruptedException {
}
};
/*
* Wrap up the data from the message as a WriteCache object. This will
* build up a RecordMap containing the allocations to be made, and
* including a ZERO (0) data length if any offset winds up being deleted
* (released).
*
* Note: We do not need to pass in the compressorKey here. It is ignored
* by WriteCache.flush(). We have expanded the payload above. Now we are
* just flushing the write cache onto the disk.
*/
final WriteCacheImpl writeCache = writeCacheService.newWriteCache(ba,
useChecksums, true/* bufferHasData */, opener,
msg.getFileExtent());
// Ensure that replicated buffers are not compacted.
writeCache.closeForWrites();
final long firstOffset = msg.getFirstOffset();
if (firstOffset < getHeaderSize())
throw new IllegalArgumentException(
"firstOffset must be beyond header: firstOffset="
+ firstOffset + ", headerSize=" + getHeaderSize());
if (firstOffset < getNextOffset())
throw new IllegalArgumentException(
"firstOffset must be beyond nextOffset: firstOffset="
+ firstOffset + ", nextOffset=" + getNextOffset());
writeCache.setFirstOffset(firstOffset);
/*
* Setup buffer for writing. We receive the buffer with pos=0,
* limit=#ofbyteswritten. However, flush() expects pos=limit, will clear
* pos to zero and then write bytes up to the limit. So, we set the
* position to the limit before calling flush.
*/
final ByteBuffer bb = ba.buffer();
final int limit = bb.limit();
bb.position(limit);
/* Flush the write in the write cache to the backing store.
*
* Note: writeOnChannel() takes the extensionLock for us.
*/
// final Lock readLock = extensionLock.readLock();
// readLock.lock();
// try {
writeCache.flush(false/* force */);
// } finally {
// readLock.unlock();
// }
// install reads into readCache (if any)
writeCacheService.installReads(writeCache);
}
@Override
public Future sendHALogBuffer(final IHALogRequest req,
final IHAWriteMessage msg, final IBufferAccess b)
throws IOException, InterruptedException {
// Buffer now contains data directly from log, DO NOT read direct from store
final ByteBuffer clientBuffer = b.buffer();
// final int nbytes = msg.getSize();
// clientBuffer.position(0);
// clientBuffer.limit(nbytes);
//
// readRaw(/*nbytes, */msg.getFirstOffset(), clientBuffer);
assert clientBuffer.remaining() > 0 : "Empty buffer: " + clientBuffer;
@SuppressWarnings("unchecked")
final QuorumPipeline quorumMember = (QuorumPipeline) quorum
.getMember();
final Future remoteWriteFuture = quorumMember.replicate(req, msg,
clientBuffer);
return remoteWriteFuture;
}
@Override
public Future sendRawBuffer(final IHARebuildRequest req,
// final long commitCounter, final long commitTime,
final long sequence, final long quorumToken, final long fileExtent,
final long offset, final int nbytes, final ByteBuffer b)
throws IOException, InterruptedException {
// read direct from store
final ByteBuffer clientBuffer = b;
clientBuffer.position(0);
clientBuffer.limit(nbytes);
readRaw(/*nbytes,*/ offset, clientBuffer);
assert clientBuffer.remaining() > 0 : "Empty buffer: " + clientBuffer;
@SuppressWarnings("unchecked")
final QuorumPipeline quorumMember = (QuorumPipeline) quorum
.getMember();
final int chk = ChecksumUtility.threadChk.get().checksum(b);
final IHAWriteMessage msg = new HAWriteMessage(storeUUIDRef.get(),
-1L/* commitCounter */, -1L/* commitTime */, sequence, nbytes,
chk, StoreTypeEnum.WORM, quorumToken, fileExtent, offset/* firstOffset */);
final Future remoteWriteFuture = quorumMember.replicate(req, msg,
clientBuffer);
return remoteWriteFuture;
}
@Override
public void writeOnStream(final OutputStream os, final ISnapshotData snapshotData,
final Quorum> quorum, final long token)
throws IOException, QuorumException {
final FileChannelUtility.ReopenerInputStream instr = new FileChannelUtility.ReopenerInputStream(
opener);
try {
MergeStreamWithSnapshotData.process(instr, snapshotData, os);
if (quorum!=null && !quorum.getClient().isJoinedMember(token)) {
// See #1172
throw new QuorumException();
}
} finally {
instr.close();
}
}
@Override
public void setExtentForLocalStore(final long extent) throws IOException,
InterruptedException {
truncate(extent);
}
@Override
public void resetFromHARootBlock(final IRootBlockView rootBlock) {
final long rbNextOffset = rootBlock.getNextOffset();
nextOffset.set(rbNextOffset);
commitOffset.set(rbNextOffset);
// Note: Potentially updated (if root blocks were reinstalled).
storeUUIDRef.set(rootBlock.getUUID());
}
@Override
public void postHACommit(final IRootBlockView rootBlock) {
final long rbNextOffset = rootBlock.getNextOffset();
nextOffset.set(rbNextOffset);
commitOffset.set(rbNextOffset);
}
@Override
public Object snapshotAllocators() {
/*
* Note: The WORM does not have a concept of allocators. There is
* nothing that we need to snapshot since there is no state that is
* overwritten.
*/
return null;
}
@Override
public void computeDigest(final Object snapshot, final MessageDigest digest)
throws DigestException, IOException {
if (snapshot != null)
throw new UnsupportedOperationException();
IBufferAccess buf = null;
try {
try {
// Acquire a buffer.
buf = DirectBufferPool.INSTANCE.acquire();
} catch (InterruptedException ex) {
// Wrap and re-throw.
throw new IOException(ex);
}
// The backing ByteBuffer.
final ByteBuffer b = buf.buffer();
// // A byte[] with the same capacity as that ByteBuffer.
// final byte[] a = new byte[b.capacity()];
// The capacity of that buffer (typically 1MB).
final int bufferCapacity = b.capacity();
/*
* FIXME computeDigest(): This probably should be userExtent +
* header, not fileExtent. By choosing userExtent, we are
* effectively snapshotting the region on which we will compute the
* digest. However, there might be bytes not yet written onto the
* backing file (e.g., in the write cache service). Those bytes
* would not show up in a read from the file so we would need to do
* more work to make this digest computation safe for concurrent
* writes. What is safe is to compute the digest from the last
* commit point. That "snapshop" is always valid for the WORM.
*/
// The size of the file at the moment we begin.
final long fileExtent = getExtent();
// The #of bytes to be transmitted.
final long totalBytes = fileExtent;
// The #of bytes remaining.
long remaining = totalBytes;
// The offset of the current block.
long offset = 0L;
// The digest block sequence.
long sequence = 0L;
if (log.isInfoEnabled())
log.info("Computing digest: nbytes=" + totalBytes);
while (remaining > 0) {
final int nbytes = (int) Math.min((long) bufferCapacity,
remaining);
if (log.isDebugEnabled())
log.debug("Computing digest: sequence=" + sequence
+ ", offset=" + offset + ", nbytes=" + nbytes);
// Setup for read.
b.position(0);
b.limit(nbytes);
// read block
readRaw(/*nbytes,*/ offset, b);
// // Copy data into our byte[].
// final byte[] c = BytesUtil.toArray(b, false/* forceCopy */, a);
// update digest
// digest.update(c, 0/* off */, nbytes/* len */);
digest.update(b);
remaining -= nbytes;
offset += nbytes;
sequence++;
}
if (log.isInfoEnabled())
log.info("Computed digest: #blocks=" + sequence + ", #bytes="
+ totalBytes);
// Done.
return;
} finally {
if (buf != null) {
try {
// Release the direct buffer.
buf.release();
} catch (InterruptedException e) {
log.warn(e);
}
}
}
}
@Override
public void writeRawBuffer(HARebuildRequest req, IHAWriteMessage msg,
ByteBuffer transfer) throws IOException {
// expand buffer before writing on the store.
final ByteBuffer xtransfer = msg.expand(transfer);
// if (m_rebuildRequest == null)
// throw new IllegalStateException("Store is not in rebuild state");
//
// if (m_rebuildSequence != msg.getSequence())
// throw new IllegalStateException("Invalid sequence number for rebuild, expected: " + m_rebuildSequence + ", actual: " + msg.getSequence());
FileChannelUtility.writeAll(this.opener, xtransfer, msg.getFirstOffset());
// m_rebuildSequence++;
}
@Override
public StoreState getStoreState() {
return new WormStoreState();
}
public static class WormStoreState implements StoreState {
private static final long serialVersionUID = 1L;
@Override
public boolean equals(final Object obj) {
if (obj == null || !(obj instanceof WormStoreState))
return false;
final WormStoreState other = (WormStoreState) obj;
// Nothing to compare.
return true;
}
}
// @Override
// public void prepareForRebuild(HARebuildRequest req) {
// assert m_rebuildRequest == null;
//
// m_rebuildRequest = req;
// m_rebuildSequence = 0;
// }
//
// @Override
// public void completeRebuild(final HARebuildRequest req, final IRootBlockView rbv) {
// assert m_rebuildRequest != null;
//
// assert m_rebuildRequest.equals(req);
//
// // TODO: reinit from file
// this.resetFromHARootBlock(rbv);
//
// m_rebuildRequest = null;
// }
}