|
BTree Indexes with Locking
Implements classes used by the language layer to implement
SQL secondary indexes. The classes here extend and use the classes in
{@link org.apache.derby.impl.store.access.btree} to implement
a locked btree index.
The key to understanding the class layout is to understand the public
store interfaces in
{@link org.apache.derby.iapi.store.access.conglomerate}, which contains
the shared interfaces that must be implemented by all access methods.
Currently Derby implements heap and btree index access methods. Users of
access methods use the same interface no matter what the underlying
type or particular implementation of the access method. Therefore, Derby
can support multiple types of btree index implementations, which if done
right should require no changes to actual users of the access methods.
In reality the interfaces would have to change in some ways to
support a radically different kind of access method, such as
GiST. But
the implementor should enhance the interfaces in the conglomerate
package so that these can then be supported by all existing access
methods.
Isolation Levels
Isolation level implementation in the B-Tree index is done with data
only locking, i.e., locks on secondary index rows are actually locks on the
data rows that they point to. The specifics of particular isolation levels
are hidden in various implementations of the {@link org.apache.derby.impl.store.access.btree.BTreeLockingPolicy BTreeLockingPolicy} class.
The classes which do scans, deletes, and inserts do not have isolation
specific code, instead they make lock calls using BTreeLockingPolicy
interfaces, and then depending on the isolation level one of the implmentations
does the actual locking.
|