com.bigdata.journal
Class MappedBufferStrategy

java.lang.Object
  com.bigdata.rawstore.AbstractRawStore
      com.bigdata.rawstore.AbstractRawWormStore
          com.bigdata.journal.AbstractBufferStrategy
              com.bigdata.journal.BasicBufferStrategy
                  com.bigdata.journal.DiskBackedBufferStrategy
                      com.bigdata.journal.MappedBufferStrategy

All Implemented Interfaces:

ICounterSetAccess, IBufferStrategy, IDiskBasedStrategy, IAddressManager, IMRMW, IMROW, IRawStore, IWORM

public class MappedBufferStrategy
extends DiskBackedBufferStrategy

Memory-mapped journal strategy (this mode is NOT recommended).

Note: the use of FileLock with a memory-mapped file is NOT recommended by the JDK as this combination is not permitted on some platforms.

Note: Extension and truncation of a mapped file are not possible with the JDK since there is no way to guarentee that the mapped file will be unmapped in a timely manner. Journals that handle IJournal#overflow() should trigger overflow just a bit earlier for a MappedByteBuffer in an attempt to avoid running out of space in the journal. If a transaction can not be committed due to overflow, it could be re-committed after handling the overflow event (e.g., throw a "CommitRetryException").

The mapped mode has nearly the performance of the transient mode in some tests. However the use of mapped files might not prove worth the candle due to the difficulties with resource deallocation for this strategy and the good performance of some alternative strategies. There are also some other issues that have shown up in some of the tests suites -- look carefully if you are going to pursue this!

Version:

$Id: MappedBufferStrategy.java 2265 2009-10-26 12:51:06Z thompsonbry $

Author:

Bryan Thompson

See Also:

http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4724038, BufferMode.Mapped

Field Summary

Fields inherited from class com.bigdata.journal.DiskBackedBufferStrategy

opener

Fields inherited from class com.bigdata.journal.BasicBufferStrategy

extent, userExtent

Fields inherited from class com.bigdata.journal.AbstractBufferStrategy

bufferMode, ERR_ADDRESS_IS_NULL, ERR_ADDRESS_NOT_WRITTEN, ERR_BAD_RECORD_SIZE, ERR_BUFFER_EMPTY, ERR_BUFFER_NULL, ERR_BUFFER_OVERRUN, ERR_MAX_EXTENT, ERR_NOT_OPEN, ERR_OPEN, ERR_READ_ONLY, ERR_RECORD_LENGTH_ZERO, ERR_TRUNCATE, initialExtent, log, maximumExtent, nextOffset, WARN

Fields inherited from class com.bigdata.rawstore.AbstractRawWormStore

am

Fields inherited from interface com.bigdata.rawstore.IAddressManager

NULL

Method Summary

void	close() The file channel is closed, but according to http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4724038 there is no way to guarentee when the mapped file will be released.
void	deleteResources() Overrides the default behavior so that an exception is NOT thrown if the file can not be deleted.
void	force(boolean metadata) Force the data to disk.
boolean	isFullyBuffered() True iff the store is fully buffered (all reads are against memory).
void	truncate(long newExtent) Note: Extension and truncation of a mapped file are not possible with the JDK since there is no way to guarentee that the mapped file will be unmapped in a timely manner.

Methods inherited from class com.bigdata.journal.DiskBackedBufferStrategy

delete, getChannel, getFile, getHeaderSize, getRandomAccessFile, isStable, read, readRootBlock, setNextOffset, transferTo, writeRootBlock

Methods inherited from class com.bigdata.journal.BasicBufferStrategy

getBufferView, getCounters, getExtent, getUserExtent, write

Methods inherited from class com.bigdata.journal.AbstractBufferStrategy

abort, assertOpen, closeForWrites, commit, destroy, getBufferMode, getInitialExtent, getMaximumExtent, getMaxRecordSize, getMetaBitsAddr, getMetaStartAddr, getMinimumExtension, getNextOffset, getResourceMetadata, getUUID, isOpen, isReadOnly, overflow, requiresCommit, size, transferFromDiskTo, useChecksums

Methods inherited from class com.bigdata.rawstore.AbstractRawWormStore

getAddressManager, getByteCount, getOffset, getOffsetBits, getPhysicalAddress, toAddr, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface com.bigdata.journal.IBufferStrategy

abort, closeForWrites, commit, getAddressManager, getBufferMode, getCounters, getExtent, getInitialExtent, getMaximumExtent, getMaxRecordSize, getMetaBitsAddr, getMetaStartAddr, getNextOffset, getOffsetBits, getUserExtent, requiresCommit, useChecksums

Methods inherited from interface com.bigdata.rawstore.IRawStore

destroy, getResourceMetadata, getUUID, isOpen, isReadOnly, size, write

Methods inherited from interface com.bigdata.rawstore.IAddressManager

getByteCount, getOffset, getPhysicalAddress, toAddr, toString

Method Detail

isFullyBuffered

public boolean isFullyBuffered()

Description copied from interface: IRawStore

True iff the store is fully buffered (all reads are against memory). Implementations MAY change the value returned by this method over the life cycle of the store, e.g., to conserve memory a store may drop or decrease its buffer if it is backed by disk.

Note: This does not guarantee that the OS will not swap the buffer onto disk.

force

public void force(boolean metadata)

Force the data to disk.

Specified by:

force in interface IRawStore

Overrides:

force in class DiskBackedBufferStrategy

Parameters:

metadata - If true, then force both the file contents and the file metadata to disk.

See Also:

MappedByteBuffer.force()

close

public void close()

The file channel is closed, but according to http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4724038 there is no way to guarentee when the mapped file will be released.

Specified by:

close in interface IRawStore

Overrides:

close in class DiskBackedBufferStrategy

deleteResources

public void deleteResources()

Overrides the default behavior so that an exception is NOT thrown if the file can not be deleted. Since Java can not provide for synchronous unmap of memory-mapped files, we can not delete the backing file immediately. Instead, we mark the file for "deleteOnExit" and let the VM attempt to clean it up when it exits.

Specified by:

deleteResources in interface IRawStore

Overrides:

deleteResources in class DiskBackedBufferStrategy

truncate

public void truncate(long newExtent)

Note: Extension and truncation of a mapped file are not possible with the JDK since there is no way to guarentee that the mapped file will be unmapped in a timely manner.

Specified by:

truncate in interface IBufferStrategy

Overrides:

truncate in class BasicBufferStrategy

Parameters:

newExtent - The new extent of the journal. This value represent the total extent of the journal, including any root blocks together with the user extent.

Throws:

UnsupportedOperationException - Always thrown.