uk.ac.starlink.table

Class StoragePolicy

java.lang.Object

uk.ac.starlink.table.StoragePolicy

Direct Known Subclasses:

ByteStoreStoragePolicy, MonitorStoragePolicy

public abstract class StoragePolicy
extends Object

Defines storage methods for bulk data. If the table handling system needs to cache bulk data somewhere, for instance because it is reading a table from a stream but needs to make it available for random access, it will use a StoragePolicy object to work out how to do it.

Code which has no preferences about how to store data can obtain an instance of this class using the getDefaultPolicy() method. The initial value of this may be selected by setting the system property named by the string PREF_PROPERTY ("startable.storage"). You may also use the name of a class which extends StoragePolicy and has a no-arg constructor, in which case one of these will be instantiated and used. The default, if not otherwise set, corresponds to "adaptive".

Code which wants to store data in a particular way may use one of the predefined policies ADAPTIVE, PREFER_MEMORY, PREFER_DISK SIDEWAYS or DISCARD, or may implement their own policy by extending this class. If you want more control, you can always create instances of the public RowStore implementations directly.

Author:

Mark Taylor (Starlink)

Field Summary

Fields

Modifier and Type	Field and Description
static StoragePolicy	ADAPTIVE Storage policy which will store small amounts of data in an array in memory, and larger amounts in a scratch disk file.
static StoragePolicy	DISCARD Storage policy which just throws away the rows it is given.
static String	PREF_PROPERTY Name of the system property which can be set to indicate the initial setting of the default storage policy ("startable.storage").
static StoragePolicy	PREFER_DISK Storage policy which will normally store table data in a scratch disk file.
static StoragePolicy	PREFER_MEMORY Storage policy which will always store table data in memory.
static StoragePolicy	SIDEWAYS Storage policy which will normally store table data in scratch disk files in such a way that cells from the same column are contiguous on disk.

Constructor Summary

Constructors

Constructor and Description
StoragePolicy()

Method Summary

Modifier and Type	Method and Description
StarTable	copyTable(StarTable table) Returns a random-access deep copy of the given table.
static StoragePolicy	getDefaultPolicy() Returns the default storage policy for this JVM.
abstract ByteStore	makeByteStore() Returns a new ByteStore object which can be used to provide a destination for general purpose data storage.
abstract RowStore	makeConfiguredRowStore(StarTable meta) Creates a new RowStore and primes it by calling TableSink.acceptMetadata(uk.ac.starlink.table.StarTable) on it.
abstract RowStore	makeRowStore() Returns a new RowStore object which can be used to provide a destination for random-access table storage.
StarTable	randomTable(StarTable table) Returns a table based on a given table and guaranteed to have random access.
static void	setDefaultPolicy(StoragePolicy policy) Sets the default storage policy used for this JVM.

Methods inherited from class java.lang.Object

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

Field Detail

PREF_PROPERTY

public static final String PREF_PROPERTY

Name of the system property which can be set to indicate the initial setting of the default storage policy ("startable.storage"). Currently recognised values are "adaptive", "memory", "disk", "sideways", and "discard". Alternatively, the classname of a StoragePolicy implementation with a no-arg constructor may be supplied.

See Also:

Constant Field Values

PREFER_MEMORY

public static final StoragePolicy PREFER_MEMORY

Storage policy which will always store table data in memory. Table cells are stored as objects, which will be fast to write, and can cope with any object type, but may be expensive on memory.

PREFER_DISK

public static final StoragePolicy PREFER_DISK

Storage policy which will normally store table data in a scratch disk file. If it's impossible for some reason (I/O error, security restrictions) then it will fall back to using memory. It might also use memory if it thinks it's got a small table to deal with. Temporary disk files are written in the default temporary directory, which is the value of the java.io.tmpdir system property. These files will be deleted when the JVM exits, if not before. They will probably be deleted around the time they are no longer needed (when the RowStore in question is garbage collected), though this cannot be guaranteed since it depends on the details of the JVM's GC implementation.

SIDEWAYS

public static final StoragePolicy SIDEWAYS

Storage policy which will normally store table data in scratch disk files in such a way that cells from the same column are contiguous on disk. This may be more efficient for certain access patterns for tables which are very large and, in particular, very wide. It's generally more expensive on system resources than PREFER_DISK however, so it is only the best choice in rather specialised circumstances. If it's impossible for some reason to store the data in this way, or if the number of cells requested is small, it will fall back to using memory storage. Temporary disk files (at least one per column) are written in the default temporary directory, which is the value of the java.io.tmpdir system property. These files will be deleted when the JVM exits, if not before. They will probably be deleted around the time they are no longer needed (when the RowStore in question is garbage collected), though this cannot be guaranteed since it depends on the details of the JVM's GC implementation.

DISCARD

public static final StoragePolicy DISCARD

Storage policy which just throws away the rows it is given. Tables obtained from its row stores will have no rows. Obviously, this has rather limited application.

ADAPTIVE

public static final StoragePolicy ADAPTIVE

Storage policy which will store small amounts of data in an array in memory, and larger amounts in a scratch disk file. Temporary disk files are written in the default temporary directory, which is the value of the java.io.tmpdir system property. These files will be deleted when the JVM exits, if not before. They will probably be deleted around the time they are no longer needed (when the RowStore in question is garbage collected), though this cannot be guaranteed since it depends on the details of the JVM's GC implementation.

Constructor Detail

StoragePolicy

public StoragePolicy()

Method Detail

getDefaultPolicy

public static StoragePolicy getDefaultPolicy()

Returns the default storage policy for this JVM.

Returns:

default storage policy

setDefaultPolicy

public static void setDefaultPolicy(StoragePolicy policy)

Sets the default storage policy used for this JVM.

Parameters:

policy - new default storage policy

makeByteStore

public abstract ByteStore makeByteStore()

Returns a new ByteStore object which can be used to provide a destination for general purpose data storage.

Returns:

new byte store

makeRowStore

public abstract RowStore makeRowStore()

Returns a new RowStore object which can be used to provide a destination for random-access table storage.

Returns:

a RowStore object

makeConfiguredRowStore

public abstract RowStore makeConfiguredRowStore(StarTable meta)

Creates a new RowStore and primes it by calling TableSink.acceptMetadata(uk.ac.starlink.table.StarTable) on it.

Parameters:

meta - template giving the metadata which describes the rows that will have to be stored

Returns:

a RowStore on which acceptMetadata(meta) has been called

randomTable

public StarTable randomTable(StarTable table)
                      throws IOException

Returns a table based on a given table and guaranteed to have random access. If the original table table has random access then it is returned, otherwise a new random access table is built using its data.

Parameters:

table - original table

Returns:

a table with the same data as table and with isRandom()==true

Throws:

IOException

copyTable

public StarTable copyTable(StarTable table)
                    throws IOException

Returns a random-access deep copy of the given table. This utility method is like randomTable(uk.ac.starlink.table.StarTable) except that a copy is made even if the original is already random access. It can be useful if you want a copy of the table known to have the resource usage or performance characteristics defined by this policy.

Parameters:

table - input table

Returns:

deep copy of table

Throws:

IOException