It's a very common procedure in JDBC to create a Connection object (or obtain an existing one) and use it to create aStatement. The Statement, being fed an SQL SELECT, returns a ResultSet. The ResultSet is then fed through a while loop (not unlike an Iterator) until the ResultSet says it's empty, with the body of the loop extracting one column at a time in left-to-right order.
This whole operation is so common that is has become almost sacred: it's done that way simply because that's the way it's done. Alas, it's completely unnecessary.
Introducing the scrollable ResultSet
Many developers are unaware of the fact that JDBC has been considerably enhanced over the years, even though those enhancements are reflected in new version numbers and releases. The first major enhancement, JDBC 2.0, occurred around the time of JDK 1.2. As of this writing, JDBC stands at version 4.0.
One of the interesting (though frequently ignored) enhancements to JDBC 2.0 is the ability to "scroll" through the ResultSet, meaning we can go forward or backward, or even both, as need dictates. Doing so requires a bit of forward-thinking, however — the JDBC call must indicate that it wants a scrollable ResultSet at the time the Statement is created.
Verifying ResultSet type
If you suspect a driver may not actually support scrollable ResultSets, despite what it says in theDatabaseMetaData, you can verify the ResultSet type by calling getType(). Of course, if you're that paranoid, you might not trust the return value of getType(), either. Suffice it to say, if getType() lies about theResultSet returned, they really are out to get you.
If the underlying JDBC driver supports scrolling, a scrollableResultSet will be returned from that Statement, but it's best to figure out if the driver supports scrollability before asking for it. You can ask about scrolling via the DatabaseMetaData object, which can be obtained from any Connection, as described previously.
Once you have a DatabaseMetaData object, a call togetJDBCMajorVersion() will determine whether the driver supports at least the JDBC 2.0 specification. Of course, a driver could lie about its level of support for a given specification, so to play it particularly safe, call the supportsResultSetType() method with the desired ResultSet type. (It's a constant on the ResultSetclass; we'll talk about the values of each in just a second.)
| int JDBCVersion = dbmd.getJDBCMajorVersion(); boolean srs = dbmd.supportsResultSetType(ResultSet.TYPE_SCROLL_INSENSITIVE); if (JDBCVersion > 2 || srs == true) { // scroll, baby, scroll! } |
Requesting a scrollable ResultSet
Assuming your driver says yes (if it doesn't, you need a new driver or database), you can request a scrollable ResultSet by passing two parameters to the Connection.createStatement() call, shown in Listing 2:
| Statement stmt = con.createStatement( ResultSet.TYPE_SCROLL_INSENSITIVE, ResultSet.CONCUR_READ_ONLY); ResultSet scrollingRS = stmt.executeQuery("SELECT * FROM whatever"); |
You have to be particularly careful when calling createStatement() because its first and second parameters are both ints. (Curse the fact that we didn't get enumerated types until Java 5!) Any int value (including the wrong constant value) will work withcreateStatement().
The first parameter, indicating the "scrollability" desired in the ResultSet, can be one of three accepted values:
- ResultSet.TYPE_FORWARD_ONLY: This is the default, firehose-style cursor that we know and love.
- ResultSet.TYPE_SCROLL_INSENSITIVE: This ResultSet enables backward iteration as well as forward, but if the data in the database changes, the ResultSet won't reflect it. This scrollable ResultSet is probably the most commonly desired type.
- ResultSet.TYPE_SCROLL_SENSITIVE: The ResultSet created will not only allow for bidirectional iteration, but will also give a "live" view of the data in the database as it changes.
The second parameter is discussed in the next tip, so hang on.
Once you've obtained a ResultSet from the Statement, scrolling backward through it is just a matter of calling previous(), which goes backward a row instead of forward, as next() would. Or you could call first() to go back to the beginning of theResultSet, or call last() to go to the end of the ResultSet, or ... well, you get the idea.
The relative() and absolute() methods can also be helpful: the first moves the specified number of rows (forward if the value is positive, backward if the value is negative), and the latter moves to the specified row in the ResultSet regardless of where the cursor is. Of course, the current row number is available via getRow().
If you plan on doing a lot of scrolling in a particular direction, you can help the ResultSet by specifying that direction, by callingsetFetchDirection(). (A ResultSet will work regardless of its scrolling direction but knowing beforehand allows it to optimize its data retrieval.)
JDBC doesn't just support bidirectional ResultSets, it also supports in-place updates to ResultSets. This means that rather than create a new SQL statement to change the values currently stored in the database, you can just modify the value held inside the ResultSet, and it will be automatically sent to the database for that column of that row.
Asking for an updateable ResultSet is similar to the process involved in asking for a scrollable ResultSet. In fact, it's where you'll use the second parameter to createStatement(). Instead of specifying ResultSet.CONCUR_READ_ONLY for the second parameter, send ResultSet.CONCUR_UPDATEABLE, as shown in Listing 3:
Listing 3. I'd like an updateable ResultSet, please
| Statement stmt = con.createStatement( ResultSet.TYPE_SCROLL_INSENSITIVE, ResultSet.CONCUR_UPDATEABLE); ResultSet scrollingRS = stmt.executeQuery("SELECT * FROM whatever"); |
Assuming your driver supports updateable cursors (that's another feature of the JDBC 2.0 specification, which most "real-world" databases will support), you can update any given value in a ResultSet by navigating to that row and calling one of theupdate...() methods on it (shown in Listing 6). Like the get...() methods on ResultSet, update...() is overloaded for the actual column type in the ResultSet. So to change the floating-point column named "PRICE", call updateFloat("PRICE"). Doing so only updates the value in the ResultSet, however. To push the value to the database backing it, call updateRow(). If the user changes his or her mind about changing the price, a call to cancelRowUpdates() will kill all pending updates.
| Statement stmt = con.createStatement( ResultSet.TYPE_SCROLL_INSENSITIVE, ResultSet.CONCUR_UPDATEABLE); ResultSet scrollingRS = stmt.executeQuery("SELECT * FROM lineitem WHERE id=1"); scrollingRS.first(); scrollingRS.udpateFloat("PRICE", 121.45f); // ... if (userSaidOK) scrollingRS.updateRow(); else scrollingRS.cancelRowUpdates(); |
JDBC 2.0 supports more than just updates. If the user wants to add a completely new row, rather than create a new Statementand execute an INSERT, just call moveToInsertRow(), call update...() for each column, then call insertRow() to complete the work. If a column value isn't specified, it's assumed to be an SQL NULL (which might trigger an SQLException if the database schema doesn't allow NULLs for that column).
Naturally, if the ResultSet supports updating a row, it must also support deleting one, via deleteRow().
Oh, and before I forget, all of this scrollability and updateability applies equally to PreparedStatement (by passing those parameters to the prepareStatement() method), which is infinitely preferable to a regular Statement due to the constant danger of SQL injection attacks.
If all this functionality has been in JDBC for the better part of a decade, why are most developers still stuck on forward-scrollingResultSets and disconnected access?
The main culprit is scalability. Keeping database connections to a minimum is key to supporting the massive numbers of users that the Internet can bring to a company's web site. Because scrolling and/or updating ResultSets usually requires an open network connection, many developers will not (or cannot) use them.
Fortunately, JDBC 3.0 introduced an alternative that lets you do many of the same things you would with a ResultSet, without necessarily needing to keep the database connection open.
In concept, a Rowset is essentially a ResultSet, but one which allows for either a connected or disconnected model. All you need to do is create a Rowset, point it at a ResultSet, and when it's done populating itself, use it as you would a ResultSet, shown in Listing 5:
Listing 5. Rowset replaces ResultSet
| Statement stmt = con.createStatement( ResultSet.TYPE_SCROLL_INSENSITIVE, ResultSet.CONCUR_UPDATEABLE); ResultSet scrollingRS = stmt.executeQuery("SELECT * FROM whatever"); if (wantsConnected) JdbcRowSet rs = new JdbcRowSet(scrollingRS); // connected else CachedRowSet crs = new CachedRowSet(scrollingRS); disconnected |
JDBC comes with five "implementations" (meaning extended interfaces) of the Rowset interface. JdbcRowSet is a connectedRowset implementation; the remaining four are disconnected:
- CachedRowSet is just a disconnected Rowset.
- WebRowSet is a subclass of CachedRowSet that knows how to transform its results to XML and back again.
- JoinRowSet is a WebRowSet that also knows how to form the equivalent of an SQL JOIN without having to connect back to the database.
- FilteredRowSet is a WebRowSet that also knows how to further filter the data handed back without having to connect back to the database.
Rowsets are full JavaBeans, meaning they support listener-style events, so any modifications to the Rowset can be caught, examined, and acted upon, if desired. In fact, Rowset can even manage the complete act against the database if it has itsUsername, Password, URL, and DatasourceName properties set (which means it will create a connection usingDriverManager.getConnection()) or its Datasource property set (which was probably obtained via JNDI). You would then specify the SQL to execute in the Command property, call execute(), and start working with the results — no further work required.
Rowset implementations are generally provided by the JDBC driver, so the actual name and/or package will depend on what JDBC driver you use. Rowset implementations have been a part of the standard distribution since Java 5, so you should be able to just create a ...RowsetImpl() and go. (In the unlikely event that your driver doesn't provide one, Sun offers a reference implementation; see Resources for the link.)
Despite their usefulness, Rowsets sometimes just don't meet all your needs, and you may need to fall back to writing straight SQL statements. In those situations, particularly when you're facing a slew of work, you might appreciate the ability to do batch updates, executing more than one SQL statement against the database as part of one network round-trip.
To determine whether the JDBC driver supports batch updates, a quick call to theDatabaseMetaData.supportsBatchUpdates() yields a boolean telling the story. Assuming batch updates are supported (indicated by anything non-SELECT), queue one up and release it in a blast, like in Listing 6:
Listing 6. Let the database have it!
| conn.setAutoCommit(false); PreparedStatement pstmt = conn.prepareStatement("INSERT INTO lineitems VALUES(?,?,?,?)"); pstmt.setInt(1, 1); pstmt.setString(2, "52919-49278"); pstmt.setFloat(3, 49.99); pstmt.setBoolean(4, true); pstmt.addBatch(); // rinse, lather, repeat int[] updateCount = pstmt.executeBatch(); conn.commit(); conn.setAutoCommit(true); |
The call to setAutoCommit() is necessary because by default, the driver will try to commit every statement that it is fed. Other than that, the rest of the code is pretty straightforward: do the usual SQL thing with the Statement or PreparedStatement, but instead of calling execute(), invoke executeBatch(), which queues the call instead of sending it right away.
When the whole mess of statements is ready to go, fire them all at the database with executeBatch(), which returns an array of integer values, each of which holds the same result as if executeUpdate() had been used.
In the event that a statement in the batch fails, if the driver doesn't support batch updates, or if a statement in the batch returns aResultSet, the driver will throw a BatchUpdateException. In some cases, the driver may have tried to continue executing statements after an exception was thrown. The JDBC specification doesn't mandate particular behavior, so you are advised to experiment with your driver beforehand, so that you know exactly how it behaves. (But of course you'll be running unit tests, so you'll discover the error long before it becomes a problem, right?)
No comments:
Post a Comment