Page History
<?xml version="1.0" encoding="utf-8"?>
<html>
NOTE: This is just a proposal. There is no guarantee that any of this will ever end up in the actual codebase, I just felt it was worth experimenting. --JR
Update (09-05-2007): _I've made this work successfully for
...
Collection
...
s,
...
Item
...
s, and
...
Bundle
...
s. The performance improvements aren't fully implemented, but the separation is there, and in theory, that was the hard part. The
...
ArchiveManager
...
seems to be working pretty well too._ --JR
Update (11-05-2007): _I've implemented DAOs for the
...
Community
...
class as well. The
...
ArchiveManager
...
now supports moving
...
Item
...
s,
...
Collection
...
_, and
...
_ Communities
...
between containers._ - -JR
Update (23-05-2007): I've totally reimplemented persistent identifiers in DSpace as well (see PersistentIdentifiers ). As well as removing the Handle System dependency, they also use DAOs. - -JR
Update (20-06-2007): _After a bit of a hiatus while I fixed persistent identifiers PersistentIdentifiers I've come back to DAOs, and I've now (mostly) got them in place for
...
Bitstream
...
s as well. The two major classes that still need doing are
...
EPerson
...
and
...
Group
...
_; once they're done, there are a few others (eg:
...
_ SupervisedItem
...
_,
...
_ WorkspaceItem
...
_, etc) but they should be relatively simple._ --JR
Just adding a comment that Handle/Pid management could be greatly improved by such an addition as well. currently with item caching, the DSpaceObject.getHandle method can become stale and using DAO's behind the scene for the HandleManagement might be beneficial – -- Mark Diggory 13:56, 10 May 2007 (EDT)
Update (14-08-2007): _Everything (apart from the code in
...
org.dspace.checker
...
_) has been pushed through the DAO + layer. Non-DAO classes no longer
...
_ import
...
the
...
DatabaseManager
...
or
...
throw
...
SQLException
...
s. There are interfaces for CRUD and link operations in
...
org.dspace.storage.dao
...
that I intend to write some tests to for throwing at all the implementing DAOs._ --JR
...
It has often struck me that DSpace would benefit from the use of Data Access Objects (DAO). If nothing else, it would make porting to alternative database platforms far easier; all we would need to do is provide alternative implementations for the DAO interfaces that worked for a given database. To this end, I have broken up some of the core classes in
...
org.dspace.content
...
to use DAOs.
As part of the same effort, I have done some work on making the
...
Context
...
less data-layer dependent (by having it hold a
...
org.dspace.storage.dao.GlobalDAO
...
rather than a
...
java.sql.Connection
...
, etc). I've also introduced a proxy for the
...
Item
...
that is a bit smarter about when it retrieves content from the data layer, and an
...
ArchiveManager
...
class that takes care of some core "archive operations" (so that other core classes don't need to).
...
- Incorporate the new DAO classes into the codebase
- Refactor
org.dspace.content.Item (etc) to use the DAO implementations of the data access methods internallyCode Block - Mark relevant methods in
org.dspace.content.Item asCode Block @DeprecatedCode Block - Using the compile-time deprecation warnings as a guide, refactor the rest of the code to use the DAOs explicitly rather than hiding the functionality behind existing methods
Without further ado, here is how I have refactored
...
org.dspace.content.Item
...
to use DAOs. A few important things to note:
- "old" code has been used where possible to avoid re-implementing the wheel
- I've never liked
org.dspace.content.ItemIterator so I've switched to using a "real"Code Block Iterator from aCode Block List<Item>List<Item>Code Block
For examples of both of these principles, see the implementation of
...
getItems()
...
below. It is a fairly straightforward wrapper for the current
...
Item.findAll()
...
, except that it returns a
Code Block |
---|
List<Item> |
...
Code Block |
---|
ItemIterator |
...
List<Item> rather than an ItemIterator.
Panel | ||||||
---|---|---|---|---|---|---|
Contents
|
...
org.dspace.content
The
...
Item
...
class will be broken up into the following classes:
[[#org.dspace.content.Item|Code Block org.dspace.content.Item
]]: core class that doesn't go near the database (it doesn't even know about the DAOs); behaves much like the current implementation.Code Block - [[#org.dspace.content.dao.ItemDAO|
org.dspace.content.dao.ItemDAO
]]: interface defining DAO APIcode[[#org.dspace.content.dao.ItemDAOFactory| org.dspace.content.dao.ItemDAOFactory
]]: factory for dishing out implementations of the above interface [[#org.dspace.content.dao.postgres.ItemDAOPostgres|Code Block org.dspace.content.dao.postgres.ItemDAOPostgres
]]: default implementation of the above interface for use with PostgreSQLcode[[#org.dspace.content.proxy.ItemProxy|org.dspace.content.proxy.ItemProxy
]]: subclass of Item that needs to know about the DAO. It will be used for (eg) only loading metadata on demand, to reduce the memory footprint ofCode Block
s Items etc.Code Block Item
The following classes have also been introduced:
[[#org.dspace.core.ArchiveManager|Code Block org.dspace.core.ArchiveManager
]]Code Block - [[#org.dspace.storage.dao.GlobalDAO|
org.dspace.storage.dao.GlobalDAO
]] [[#org.dspace.storage.dao.GlobalDAOFactory|Code Block org.dspace.storage.dao.GlobalDAOFactory
]]code- [[#org.dspace.storage.dao.postgres.GlobalDAOPostgres|
org.dspace.storage.dao.GlobalDAOPostgres
]]
Note that it might be preferable to have a more generic implementation of the
...
ItemDAO
...
interface that supports both PostgreSQL and Oracle, but given that one motivation for adopting DAOs is to remove db-specificities from the code making it easier to port, I thought it was sensible to start with just PostgreSQL. Eventually, it ought to be possible to drop in
...
ItemDAOHibernate
...
(etc) implementations that make db portability far easier.
...
org.dspace.content.Item
Basic implementation of the
...
Item
...
object. This class has been stripped down to remove all contact with the database, including (but not limited to) contstructors, factory methods,
...
update()
...
,
...
delete()
...
,
...
find()
...
, etc. I haven't decided exactly how the
...
Item
...
API will look, but it will probably be much the same as before, only with any of the aforementioned methods. Another key difference is that it will have actual Java objects as member variables instead of pulling everything out of a
...
TableRow
...
.
...
org.dspace.content.proxy.ItemProxy
This will be a fairly simple proxy implementation. Specifically, it will be closest to being a virtual proxy, in that it will appear to be a regular
...
Item
...
object, but will have a slightly smarter implementation (not loading metadata until requested, keeping track of what has changed to make updates more efficient etc).
Panelcode |
---|
public class ItemProxy extends Item { // Overrides relevant methods of Item. } |
...
|
org.dspace.content.dao.ItemDAO
This isn't final, but it's a good start.
Panelcode |
---|
public interface ItemDAO extends ContentDAO implements CRUD<Item>, Link<Item, Bundle> { public Item { public Item create(); throwsAuthorizeException public Item AuthorizeException; public Item retrieve(int id); public Item retrieve(UUID uuid); public void update(Item item); throwsAuthorizeException public void AuthorizeException; public void delete(int id); throwsAuthorizeException public List<Item> AuthorizeException; public List<Item> getItems(); public List<Item> getItemsBySubmitter(EPerson eperson); public List<Item> getItemsByCollection(Collection collection); public List<Item> getParentItems(Bundle bundle); } |
...
|
org.dspace.content.dao.ItemDAOFactory
Panelcode |
---|
public class ItemDAOFactory { public static ItemDAO { public static ItemDAO getInstance(Context context) { { // Eventually, the implementation that is returned willbe be // defined in the configuration. return new return new ItemDAOPostgres(context); } } |
...
}
}
|
org.dspace.content.dao.postgres.ItemDAOPostgres
This is a fairly straightforward implementation of the above interface. As much as possible, code from the original
...
Item
...
class will be used. For instance, this is how
...
getItems()
...
is implemented:
Panelcode |
---|
public List<Item> getItems() { try { TableRowIterator tri = { try { TableRowIterator tri = DatabaseManager.queryTable(context, "item", "SELECT item_id FROM item WHERE in_archive = '1'"); |
Panel |
List<Item> items = new List<Item> items = new ArrayList<Item>(); |
Panel |
for for (TableRow row : tri.toList()) { int id = { int id = row.getIntColumn("item_id"); items.add(retrieve(id)); } |
Panel |
return items; } catch (SQLException sqle) { // Need to think more carefully about how we deal with SQLExceptions throw new RuntimeException(sqle); } }
}
return items;
} catch (SQLException sqle){
// Need to think more carefully about how we deal with SQLExceptions
throw new RuntimeException(sqle);
}
}
|
Some changes have been made to eliminate
...
ItemIterator
...
s, and to generally make things a little more consistent with the rest of the code (this looks almost identical to, eg,
...
CollectionDAO.getCollections(
...
)
.
Code Block org.dspace.core
...
org.dspace.core.ArchiveManager
The idea behind this class came from the realisation that
...
Item.withdraw()
...
and
...
Item.reinstate()
...
don't really make sense. What I'd much rather do is call (eg)
...
ArchiveManager.withdrawItem(Item
...
item)
...
.
I've been thinking that the
...
ArchiveManager
...
could be used for certain maintenance operations as well, such as moving
Code Block |
---|
Item |
s between
Code Block |
---|
Collection |
sItems
between Collections
, and maybe acting as a wrapper for the
...
CommunityFiliator
...
.
Panelcode |
---|
public class ArchiveManager { public static void { public static void withdrawItem(Context context, Item item) { { // ... } |
Panel |
public static void } public static void reinstateItem(Context context, Item item) { { // ... } |
Panel |
public static void } public static void moveItem(Context context, Item item, Collection source, Collection dest) { { // ... } } |
...
}
}
|
org.dspace.storage
...
org.dspace.storage.dao.GlobalDAO
As suggested by Richard Jones, there probably ought to be a top-level general-purpose DAO interface that has implementations for the various storage mechanisms (
...
GlobalDAOPostgres
...
etc). The idea is to have this top-level object capture any implementation-specific details in a single top-level object, rather than in every Postgres DAO implementation. For example, with the current database "abstraction layer", the top-level implementation of
...
GlobalDAO
...
understands the
...
Context
...
object, whereas a Hibernate implementation would know what a
...
SessionFactory
...
is.
Panelcode |
---|
public interface GlobalDAO { { // The following methods actually currently throw SQLExceptions to // keep things simple, but in future SQLExceptions should be // eliminated from any code that doesn't directly touch a database. public void startTransaction() throws GlobalDAOException; public void endTransaction() throws GlobalDAOException; public void saveTransaction() throws GlobalDAOException; public void abortTransaction(); public boolean transactionOpen(); @Deprecated Connection getConnection(); } |
...
|
org.dspace.storage.dao.GlobalDAOFactory
Super-simple
...
GlobalDAO
...
factory.
...
org.dspace.storage.dao.GlobalDAOPostgres
Implementation of the
...
GlobalDAO
...
interface for PostgreSQL.
Panelcode |
---|
public class GlobalDAOPostgres implements GlobalDAO { private Connection connection; |
Panel |
---|
// ... |
Panel |
---|
public void startTransaction() { connection = { private Connection connection; // ... public void startTransaction() { connection = DatabaseManager.getConnection(); connection.setAutoCommit(false); } |
Panel |
// ... } |
...
}
|