昨天读了一篇关于JDBC4.0设计与性能提高的文章,由于小弟英语翻译水准实在有限,就不在这里献丑了,就把原文给大家转载出来供大家阅读:
转载自:http://www.javaworld.com/javaworld/jw-05-2006/jw-
0501-jdbc
.html
Design and performance improvements with JDBC 4.0
Effectively utilize JDBC'S features to get desired results with less code
Summary
Java Database Connectivity (JDBC) 4.0 is ready for release by mid 2006
as a part of Java Standard Edition 6.0. How can you leverage the new
specification to improve the design and performance of database access
and interactions in Java applications? This article discusses the new
features of JDBC 4.0, illustrates its solutions to some existing
problems, and presents its improvements in design and performance
through examples. (2,100 words; May 1, 2006)
By Shashank Tiwari
ava Database Connectivity (JDBC),
which has existed from the first public version of the core Java
language, has evolved significantly over the last 10 years. In its
current version, 4.0, which will be packaged with Java Standard Edition
6.0 (Java SE is Sun's new name for J2SE), it shows significant
improvements in design and provides a richer API, with focus on ease of
development and improvement in productivity.
This article discusses
some of the important changes in the JDBC specification that either
improve the design or facilitate better performance. The article does
not enlist or survey every single change incorporated as a part of Java Specification Request 221, the JDBC 4.0 initiative.
After reading this article, you should be ready to leverage the new features in your next set of applications.
Annotations and the generic DataSet
I assume you are already aware of annotations and generics, which were
introduced in Java with J2SE 5.0. JDBC 4.0 introduces annotations and
the generic DataSet
.
This change aims to simplify execution of SQL queries (in scenarios
that return a single result set) and SQL DML (data manipulation
language) statements (that return either a row count or nothing).
The new API defines a set of Query
and DataSet
interfaces. The Query
interface defines a set of methods decorated with the JDBC annotations.
These decorated methods describe the SQL select and update statements,
and specify how the result set should be bound to a DataSet
. The DataSet
interface is a parameterized type, as defined by generics. The DataSet
interface provides a type-safe definition for the result set data.
All Query
interfaces inherit from the BaseQuery
interface. A concrete implementation of the interface can be instantiated using either the Connection.createQueryObject()
or DataSource.createQueryObject()
methods and passing a Query
interface type as its parameter.
A DataSet
interface inherits from java.util.List
. A data class describing the columns of the result set data, returned by an annotated method of the Query
interface, is its parameter type. A DataSet
can be manipulated and operated upon both in a connected and disconnected mode. Thus, the DataSet
is implemented either as a ResultSet
or a CachedRowSet
, depending on its operating mode: connected or disconnected. DataSet
, being a sub-interface of the java.util.List
, allows access of its data rows with the Iterator pattern, using the java.util.Iterator
interface.
The data class or the user-defined class, which is a parameter type of the DataSet
interface, can be specified in two ways: as a structure or as a
JavaBeans object. Either method achieves the goal of binding result set
data columns to user-defined class definitions, but the JavaBeans
component model is more elegant and facilitates object definition reuse
within other frameworks that support the JavaBeans model.
Listing 1 illustrates
code snippets for a simple example to show how the new API is used to
create and run SQL queries, define result set data using a user-defined
class, and bind the returned result set to the user-defined
specifications.
Listing 1. Employee user-defined type and employeeQueries
pubic class Employee {
private int employeeId;
private String firstName;
private String lastName;
public int getEmployeeId() {
return employeeId;
}
public setEmployeeId(int employeeId) {
this.employeeId = employeeId;
}
public String getFirstName() {
return firstName;
}
public setFirstName(String firstName) {
this.firstName = firstName;
}
pubic String lastName() {
return lastName;
}
public setLastName(String lastName) {
this.lastName = lastName;
}
}
interface EmployeeQueries extends BaseQuery {
@Select (sql="SELECT employeeId, firstName, lastName FROM employee")
DataSet<Employee> getAllEmployees ();
@Update (sql="delete from employee")
int deleteAllEmployees ();
}
Connection con = ...
EmployeeQueries empQueries = con.createQueryObject (EmployeeQueries.class);
DataSet<Employee> empData = empQueries.getAllEmployees ();
Exception-handling enhancements
The exception-handling functionality in the JDBC API prior to version 4.0 is limited and often insufficient. SQLException
is thrown for all types of errors. There is no classification of
exceptions, and no hierarchy defines them. The only way to get some
meaningful information is to retrieve and analyze the SQLState
value. SQLState
values and their corresponding meanings change from datasource to
datasource; hence, getting to the root of the problem and efficiently
handling exceptions proves to be a tedious task.
JDBC 4.0 has enhanced
exception-handling capability and alleviates some of the mentioned
problems. The key changes are as follows:
- Classification of
SQLException
into transient and non-transient types
- Support for chained exceptions
- Implementation of the
Iterable
interface
The SQLTransientException
is thrown where a previously failed operation may succeed on retrial. The SQLNonTransientException
is thrown where retrial will not lead to a successful operation unless the cause of the SQLException
is corrected.
Figure 1 illustrates the subclasses of SQLTransientException
and SQLNonTransientException
.
Figure 1. SQL exception classes: Transient and non-transient
|
Support for chained
exceptions are now included. New constructors add extra parameters to
capture possible causes for the exception. Multiple SQLException
s could be iterated over in a loop, and getCause()
could be called to determine the exception's possible cause. The getCause()
method can return non-SQLException
s if they are the underlying cause of the exceptions.
The SQLException
class now implements the Iterable
interface and supports the J2SE 5.0 for each loop for easier and more elegant looping.
Listing 2 depicts the usage of the new for-each-loop construct:
Listing 2. For each loop
catch(SQLException ex) {
for(Throwable t : ex) {
System.out.println("exception:" + t);
}
}
SQL/XML
A large amount of data now exists in the XML format. Databases have
extended support for the XML data type by defining a standard XML type
in the SQL 2003 specification. Most database vendors have an
implementation of the XML data type in their new releases. With the
inclusion of such a type, an XML dataset or document could be one of
the fields or column values in a row of a database table. Prior to JDBC
4.0, perhaps the best way to manipulate such data within the JDBC
framework is to use proprietary extensions from the driver vendors or
access it as a CLOB
type.
JDBC 4.0 now defines SQLXML
as the Java data type that maps the database SQL XML type. The API
supports processing of an XML type as a string or as a StAX stream.
Streaming API for XML, which for Java has been adopted via JSR 173, is
based on the Iterator pattern, as opposed to the Simple API for XML
Processing (SAX), which is based on the Observer pattern.
Invoking the Connection
object's createSQLXML()
method can create a SQLXML
object. This is an empty object, so the data can be attached to it by either using the setString()
method or by associating an XML stream using the createXMLStreamWriter()
method with the object. Similarly, XML data can be retrieved from a SQLXML
object using getString()
or associating an XML stream using createXMLStreamReader()
with the object.
The ResultSet
, the PreparedStatement
, and the CallableStatement
interfaces have getSQLXML()
methods for retrieving a SQLXML
data type. PreparedStatement
and CallableStatement
also have setSQLXML()
methods to add SQLXML
objects as parameters.
The SQLXML
resources can be released by calling their free()
methods, which might prove pertinent where the objects are valid in long-running transactions. DatabaseMetaData
's getTypeInfo()
method can be called on a datasource to check if the database supports the SQLXML
data type, since this method returns all the data types it supports.
Connections and Statements
The Connection
interface definitions have been enhanced to analyze connection state and usage to facilitate efficiency.
Sometimes database
connections are unusable though they may not necessarily be closed and
garbage collected. In such situations, the database appears slow and
unresponsive. In most of these circumstances, reinitializing the
connections is perhaps the only way to resolve the problem. When using
the JDBC API prior to version 4.0, there is no way to distinguish
between a stale connection and a closed connection. The new API adds an
isValid()
method to the Connection
interface to query if the connection is still valid.
Also, database
connections are often shared among clients, and sometimes some clients
tend to use more resources than others, which can lead to
starvation-like situations. The Connection
interface defines a setClientInfo()
method to define client-specific properties, which could be utilized to
analyze and monitor resource utilization by the clients.
RowId
The RowId
in many databases is a unique way to identify a row in a table. Queries using RowId
in the search criteria are often the fastest way to retrieve data,
especially true in the case of the Oracle and DB2 databases. Since java.sql.RowId
is now a built-in type in Java, you could utilize the performance benefits associated with its usage. RowId
s
are most useful in identifying unique and specific rows when duplicate
data exists and some rows are identical. However, it is important to
understand that RowId
s often are unique only for a table and not for the entire database; they may change and are not supported by all databases. RowId
s are typically not portable across datasources and thus should be used with caution when working with multiple datasources.
A RowId
is valid for the lifetime defined by the datasource and as long as the row is not deleted. The DatabaseMetadata.getRowIdLifetime()
method is called to determine the RowId
's lifetime. The return type is an enumeration type as summarized in the table below.
RowIdLifetime enum type
|
Definition
|
ROWID_UNSUPPORTED
|
Datasource does not support RowId type |
ROWID_VALID_OTHER
|
Implementation-dependent lifetime |
ROWID_VALID_TRANSACTION
|
Lifetime is at least the containing transaction |
ROWID_VALID_SESSION
|
Lifetime is at least the containing session |
ROWID_VALID_FOREVER
|
Unlimited lifetime |
The ROWID_VALID_TRANSACTION
, ROWID_VALID_SESSION
, and ROWID_VALID_FOREVER
definitions are true as long as the row is not deleted. It is important to understand that a new RowId
is assigned if a row is deleted and reinserted, which sometimes could
happen transparently at the datasource. As an example, in Oracle, if
the "enable row movement" clause is set on a partitioned table and an
update of the partition key causes the row to move from one partition
to another, the RowId
will change. Even without the "enable row movement" flag with the "alter table table_name" move, the RowId
could change.
Both the ResultSet
and CallableStatement
interfaces have been updated to include a method called getRowID()
, which returns a javax.sql.RowId type
.
Listing 3 shows how RowId
could be retrieved from a ResultSet
and from a CallableStatement
.
Listing 3. Get RowId
//The method signatures to retrieve RowId from a ResultSet is as follows:
RowId getRowId (int columnIndex)
RowId getRowId (String columnName)
...
Statement stmt = con.createStatement ();
ResultSet rs = stmt. ExecuteQuery (…);
while (rs.next ()) {
...
java.sql.RowId rid = rs.getRowId (1);
...
}
//The method signatures to retrieve RowId from a CallableStatement is as follows:
RowId getRowId (int parameterIndex)
RowId getRowId (String parameterName)
Connection con;
...
CallableStatement cstmt = con.prepareCall (…);
...
cstmt.registerOutParameter (2, Types.ROWID);
...
cstmt.executeUpdate ();
...
java.sql.RowId rid = cstmt.getRowId (2);
The RowId
can be used to refer uniquely to a row and thus can be used to retrieve
the rows or update the row data. When fetching or updating using RowId
references, it is important to know the validity of the RowId
's
lifetime to assure consistent results. It is also advisable to
simultaneously use another reference, such as the primary key, to avoid
inconsistent results in circumstances where the RowId
could change transparently.
The RowId
values can also be set or updated. In the case of an updateable ResultSet
, the updateRowId()
method could be used to update the RowId
for a particular row in a table.
Both the PreparedStatement
and the CallableStatement
interfaces support a setRowId()
method, with different signatures, to set the RowId
as a parameter value. This value could be used to refer to data rows or to update the RowId
value for a particular row in a table.
The facility to set or update the RowId
provides the flexibility to control the unique row identifiers and
could be used to make such identifiers unique across the tables used.
Perhaps, portability of RowId
across supporting
datasources could also be achieved by explicitly setting consistent
values across them. However, because system-generated RowId
s are often efficient, and transparent tasks could alter RowIds
, they are best used by an application as a read-only attribute.
Leverage nonstandard vendor implemented resources
The new JDBC API defines a java.sql.Wrapper
interface. This interface provides the ability to access
datasource-vendor-specific resources by retrieving the delegate
instance using the corresponding wrapped proxy instance.
This Wrapper
interface has 17 sub-interfaces as per the current specification and includes Connection
, ResultSet
, Statement
, CallableStatement
, PreparedStatement
, DataSource
, DatabaseMetaData
, and ResultSetMetaData
,
among others in the list. This is an excellent design as it facilitates
datasource-vedor-specific resource implementation at almost all stages
of the query-creation and result-set-retrieval lifecycles.
The unwrap()
method returns the object that implements the given interface to allow access to vendor-specific methods. The isWrapperFor()
method returns a Boolean value. It returns true if it implements the
interface, or if it directly or indirectly is a wrapper for the object.
As an example, when
using Oracle, Oracle JDBC drivers provide update batching extensions
that are better performing and more efficient as compared to the
standard JDBC batch-updating mechanisms. For earlier JDBC versions,
this implies using the Oracle-specific definitions, such as OraclePreparedStatement
,
in the code. This compromises code portability. With the new API, many
such efficient implementations can be wrapped and exposed within the
standard JDBC definitions.
Service provider mechanism for driver loading
In a nonmanaged or standalone program scenario, prior to JDBC 4.0, you
would have to load the JDBC driver class explicitly by invoking the Class.forName
method, as shown in Listing 4:
Listing 4. Class.forName
Class.forName ("com.driverprovider.jdbc.jdbcDriverImpl");
With JDBC 4.0, if the
JDBC driver vendors package their drivers as services, defined under
the server provider mechanism definitions as per the JAR specification,
the DriverManager
code would implicitly load the driver
by searching for it in the classpath. The benefit of this mechanism is
that the developer does not need to know about the specific driver
class and can write a little less code when using JDBC. Also, since the
driver class name is no longer in the code, a name change would not
require recompilations. If multiple drivers are specified in the
classpath, then DriverManger
will try and establish a connection using the first driver it encounters in the classpath and iterate further if required.
Conclusion
In this article, I have discussed some of the new and improved features
of JDBC 4.0. Many of these new features enhance a developer's
productivity and facilitate development. Also, the specification does
not eradicate the possible use of extra JDBC frameworks to provide
templating facilities and advanced exception-handling capabilities.
However, there is some criticism as well. Some believe that annotations
effectively lead to hard-coding in code, which causes problems in code
maintainability.