com.mckoi.store
Class LoggingBufferManager

java.lang.Object
  extended by com.mckoi.store.LoggingBufferManager

public class LoggingBufferManager
extends java.lang.Object

A paged random access buffer manager that caches access between a Store and the underlying filesystem and that also handles check point logging and crash recovery (via a JournalledSystem object).

Author:
Tobias Downer

Nested Class Summary
static interface LoggingBufferManager.StoreDataAccessorFactory
          A factory interface for creating StoreDataAccessor objects from resource names.
 
Constructor Summary
LoggingBufferManager(java.io.File journal_path, boolean read_only, int max_pages, int page_size, LoggingBufferManager.StoreDataAccessorFactory sda_factory, DebugLogger debug, boolean enable_logging)
          Constructs the manager.
LoggingBufferManager(java.io.File resource_path, java.io.File journal_path, boolean read_only, int max_pages, int page_size, java.lang.String file_ext, long max_slice_size, DebugLogger debug, boolean enable_logging)
          Constructs the manager with a scattering store implementation that converts the resource to a file in the given path.
 
Method Summary
 void lockForWrite()
          Obtains a write lock on the buffer.
 void setCheckPoint(boolean flush_journals)
          Sets a check point in the log.
 void start()
          Starts the buffer manager.
 void stop()
          Stops the buffer manager.
 void unlockForWrite()
          Releases a write lock on the buffer.
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

LoggingBufferManager

public LoggingBufferManager(java.io.File journal_path,
                            boolean read_only,
                            int max_pages,
                            int page_size,
                            LoggingBufferManager.StoreDataAccessorFactory sda_factory,
                            DebugLogger debug,
                            boolean enable_logging)
Constructs the manager.


LoggingBufferManager

public LoggingBufferManager(java.io.File resource_path,
                            java.io.File journal_path,
                            boolean read_only,
                            int max_pages,
                            int page_size,
                            java.lang.String file_ext,
                            long max_slice_size,
                            DebugLogger debug,
                            boolean enable_logging)
Constructs the manager with a scattering store implementation that converts the resource to a file in the given path.

Method Detail

start

public void start()
           throws java.io.IOException
Starts the buffer manager.

Throws:
java.io.IOException

stop

public void stop()
          throws java.io.IOException
Stops the buffer manager.

Throws:
java.io.IOException

lockForWrite

public void lockForWrite()
                  throws java.lang.InterruptedException
Obtains a write lock on the buffer. This will block if a 'setCheckPoint' is in progress, otherwise it will always succeed.

Throws:
java.lang.InterruptedException

unlockForWrite

public void unlockForWrite()
Releases a write lock on the buffer. This MUST be called if the 'lockForWrite' method is called. This should be called from a 'finally' clause.


setCheckPoint

public void setCheckPoint(boolean flush_journals)
                   throws java.io.IOException,
                          java.lang.InterruptedException
Sets a check point in the log. This logs a point in which a recovery process should at least be able to be rebuild back to. This will block if there are any write locks.

Some things to keep in mind when using this. You must ensure that no writes can occur while this operation is occuring. Typically this will happen at the end of a commit but you need to ensure that nothing can happen in the background, such as records being deleted or items being inserted. It is required that the 'no write' restriction is enforced at a high level. If care is not taken then the image written will not be clean and if a crash occurs the image that is recovered will not be stable.

Throws:
java.io.IOException
java.lang.InterruptedException