Java JDBC efficiency: How long should a connection be maintained?

Question

I'm still working on the same problem mention here. It seems to work fine especially after creating an AbstractModel class shown below:

public abstract class AbstractModel {

    protected static Connection myConnection = SingletonConnection.instance().establishConnection();
    protected static Statement stmt;
    protected static ResultSet rs;

    protected boolean loginCheck;                   // if userId and userLoginHistoryId are valid - true, else false
    protected boolean userLoggedIn;                 // if user is already logged in - true, else false

    public AbstractModel (int userId, Long userLoginHistoryId){
        createConnection();                                 // establish connection
            loginCheck = false;
        userLoggedIn = false;
        if (userId == 0 && userLoginHistoryId == 0){        // special case for login
            loginCheck = true;                              // 0, 0, false, false
            userLoggedIn = false;                           // set loginCheck to true, userLogged in to false
        } else {
            userLoggedIn = true;
            try{
                String query = "select \"user_login_session_check\"(" + userId + ", " + userLoginHistoryId + ");";
                System.out.println("query: " + query);
                stmt = myConnection.createStatement();
                rs = stmt.executeQuery(query);
                while (rs.next()){
                    loginCheck = rs.getBoolean(1);
                }
            } catch (SQLException e){
                System.out.println("SQL Exception: ");
                e.printStackTrace();
            }
        }

    }
    // close connection
    public void closeConnection(){
        try{
            myConnection.close();
        } catch (SQLException e){
            System.out.println("SQL Exception: ");
            e.printStackTrace();
        }

    }
    // establish connection
    public void createConnection(){
        myConnection = SingletonConnection.instance().establishConnection();
    }

    // login session check
    public boolean expiredLoginCheck (){
        if (loginCheck == false && userLoggedIn == true){
            closeConnection();
            return false;
        } else {
            return true;
        }
    }

}

I've already posted the stored procedures and Singleton Pattern implementation in the link to the earlier question above.

I'm under the impression that I don't need to close the connection to the database after each single data transaction, as it would just slow the application. I'm looking at about 30 users for this system I'm building, so performance and usability is important.

Is it correct to prolong the connection for at least 3-4 data transactions? Eg. Validation checks to user inputs for some form, or, something similar to google's auto-suggest ... These are all separate stored function calls based on user input. Can I use one connection instance, instead of connecting and disconnecting after each data transaction? Which is more efficient?

If my assumptions are correct (more efficient to use one connection instance) then opening and closing of the connection should be handled in the controller, which is why I created the createConnection() and closeConnection() methods.

Thanks.

Answer 1

Here is the Singleton Pattern which I initialize the myConenction field in all my Models to:

public class DatabaseConnection {
    private static final String uname = "*******";
    private static final String pword = "*******";
    private static final String url = "*******************************";

    Connection connection;
    // load jdbc driver
    public DatabaseConnection(){
        try{
            Class.forName("org.postgresql.Driver");
            establishConnection();
        } catch (ClassNotFoundException ce) {
            System.out.println("Could not load jdbc Driver: ");
            ce.printStackTrace();
        }
    }
    public Connection establishConnection() {
        // TODO Auto-generated method stub
        try{
            connection = DriverManager.getConnection(url, uname, pword);
        } catch (SQLException e){
            System.out.println("Could not connect to database: ");
            e.printStackTrace();
        }
        return connection;
    }
}

public class SingletonConnection {

    private static DatabaseConnection con;

    public SingletonConnection(){}

    public static DatabaseConnection instance(){

        assert con == null;
            con = new DatabaseConnection();
        return con;
    }
}

Of course each and every connection to the database from the app goes through a Model.

Answer 2

Like Tomasz said, you should never ever depend on the fact that your application will be used by a small number of clients. The fact that the driver will timeout after a certain amount of time does not guarantee you that you will have enough available connections. Picture this: a lot of databases come pre-configured with a maximum number of connections set to (say) 15 and a timeout of (let's say) 10-15 minutes. If you have 30 clients and each does an operation, somewhere around half-way you'll be stuck short on connections.

You should handle connections, files, streams and other resources the following way:

public void doSomething()
{
    Connection connection = null;
    Statement stmt = null;
    ResultSet rs = null;

    final String sql = "SELECT ....");

    try
    {
        connection = getConnection();
        stmt = connection.createStatement();

        rs = stmt.executeQuery(sql);
        if (rs.next())
        {
            // Do something here...
        }
    }
    catch (SQLException e)
    {
        e.printStackTrace();
    }
    finally
    {
        closeResultSet(rs);
        closeStatement(stmt);
        closeConnection(connection);
    }
}

The try/catch/finally guarantees you that the connection will get closed no matter the outcome. If there is some sort of failure, the finally block will still close the connection, just like it would do, if things were okay.

Similarly, with file and streams you need to do the same thing. Initialize the respective object as null outside your try/catch/finally, then follow the approach above.

This misconception makes a lot of Java applications misbehave under Windows, where people don't close files (streams to files, etc) and these files become locked, forcing you to either kill the JVM, or even restart your machine.

You can also use a connection pool such as for example Apache's DBCP, but even then you should take care of closing your resources, despite the fact that internally, the different connection pool implementations do not necessarily close the connections.

Answer 3

Your code should never depend on the fact, that your application is currently the only client to the database or that you have only 30 users. So you should handle database connections like files, sockets and all other kinds of scarce resources that you may run ouf of.

Thus you should always clean up after yourself. No matter what you do. Open connection, do your stuff (one or SQL statements) and close connection. Always!

In your code you create your connection and save it into a static variable - this connection will last as long as your AbstractModel class lives, probably forever - this is bad. As with all similar cases put you code inside try/finally to make sure the connection gets always closed.

I have seen application servers running ouf of connections because of web applications not closing connections. Or because they closed at logout and somebody said "we will never have more then that much users at the same time" but it just scaled a little to high.

Now as you have your code running and closing the connections properly add connection pooling, like zaske said. This will remedy the performance problem of opening/closing database connection, which truely is costly. On the logical layer (your application) you doesn't want to know when to open/close physical connection, the db layer (db pool) will handle it for you.

Then you can even go and set up a single connection for your whole session model, which is also supported by DBCP - this is no danger, since you can reconfigure the pool afterwards if you need without touching your client code.

Answer 4

You'are right that you don't need to close the connection after each call.
Bare in mind that that modern database implement internal connection pools, but your application still need to connect and retrieve a connection object, and this is what it does now.
You should consider using a database connection pool - there are various Java frameworks to provide you such a solution, and they will define (you will be able to configure of course) when a database connection pool is closed.

In general - you should ask yourself whether your database serves only your application, or does it serve other application as well - if it does not serve other application as well, you might be able to be more "greedy" and keep connections open for a longer time.
I would also recommend that your application will create on start a fixed number of connections (define it in your configuration with a value of "Minimum connections number") and you will let it grow if needed to a maximum connection numbers.

As I previously mentioned - the ideas are suggest are implemented already by all kinds of frameworks, for example - the DBCP project of Apache.