# JythonBook / sandbox / database.rst

Document this as well for some major databases: PostgreSQL driver connection - http://jdbc.postgresql.org/documentation/head/connect.html Oracle - MySQL DB2? SQLServer?

That should suffice... there may even be a repository out there that lists standard URLs. Certainly it's a function of good containers to do this...

Meta - Stuff to cover:

transactions - never use autocommit of course (but perhaps there's an exception for doing so); more importantly things like the isolation of the transaction, also perhaps its durability

preparing a statement existence query - return one row paged queries scrolling bulk inserts/updates

exception mgmt - especially diagnosing various errors (maybe enhance too?)

what improvements can we make to zxJDBC in less than one month? decimal, datetime, etc. should all translate we need to have a protocol level minimum - support jdbc 3, ideally optionally jdbc 4

Add this caveat to the chapter introducing zxJDBC at the beginning of ch 12:

Finally we will look at zxJDBC package, which is a standard part of Jython since version 2.1 and complies with the Python 2.0 DBI standard. zxJDBC can be an appropriate choice for simple one-off scripts where database portabibility is not a concern. In addition, it's (generally) necessary to use zxJDBC when writing a new dialect for SQLAlchemy or Django. [But not strictly true: you can use pg8000, a pure Python DBI driver, and of course write your own DBI drivers. But please don't do that.] So knowing how zxJDBC works can useful when working with these packages. However, it's too low level for us to recommend for more general usage. Use SQLAlchemy or Django if at all possible.

Finally, JDBC itself is also directly accessible, like any other Java package from Jython. Simply use the java.sql package. In practice this should be rarely necessary. [XXX But perhaps something about how we can still look at the underlying JDBC connection and result sets, if necessary? and if so, how?] [XXX there was also an article on doing this on IBM developer works if I'm not mistaken]

## zxJDBC – Using Python’s DB API via JDBC

### Introduction to zxJDBC

The zxJDBC package provides an easy-to-use Python wrapper around JDBC. zxJDBC bridges two standards:

• JDBC is the standard platform for database access in Java.
• DBI is the standard database API for Python apps.

zxJDBC, part of Jython, provides a DBI 2.0 standard compliant interface to JDBC. Over 200 drivers are available for JDBC [XXX http://developers.sun.com/product/jdbc/drivers], and they all work with zxJDBC. High performance drivers are available for all major relational databases, including DB2, Derby, MySQL, Oracle, PostgreSQL, SQLite, SQLServer, and Sybase. And drivers are also available for non-relational and specialized databases too.

However, unlike JDBC, zxJDBC when used in the simplest way possible, blocks SQL injection attacks, minimizes overhead, and avoids resource exhaustion. In addition, zxJDBC defaults to using a transactional model (when available), instead of autocommit.

First we will look at connections and cursors, which are the key resources in working with zxJDBC, just like any other DBI package. Then we will look at what you can do them with them, in terms of typical queries and data manipulating transactions.

Note: examples in this section are for Jython 2.5.1 and later. Jython 2.5.1 introduced some simplifications for working with connections and cursors. In addition, we assume PostgreSQL for most examples, using the world sample database (also available for MySQL). [XXX link that in]

XXX setup, including the specific jar and world database

We also look at Oracle, because there has been some additional support for some of the functionality supported by Oracle, including reference cursors.

XXX setup, including the specific jar and hr database

XXX setups for other systems?

### Connections

A database connection is simply a resource object that manages access to the database system. Because database resources are generally expensive objects to allocate, and can be readily exhausted, it is important to close them as soon as you're finished using them. Here's the best way to create a database connection outside of a managed container:

from __future__ import with_statement
from com.ziclix.python.sql import zxJDBC

# for example
jdbc_url =  "jdbc:postgresql:test"
driver = "org.postgresql.Driver"

do_something(conn)


If you're just doing from the command line, you still to connect as before, but we will forego ensuring the connection is closed:

XXX >>> code


Inside a container, like an app server, you should use JDNI to allocate the resource. Generally the connection will be managed by a connection pool:

XXX code

(We will assume these two imports in the remaining code examples in this section.)

In both cases the with statement ensures that the connection is immediately closed. The alternative is to use finally:

XXX code


### More details

XXX maybe move to the end of the section? yes, i think that will break up the flow much less. So let's do that.

There are two ways to create database connections:

• Direct creation. Standalone code, such as a script, will directly create a connection.
• JNDI. Code managed by a container should use JNDI for connection creation. Such containers include GlassFish, JBoss, Tomcat, WebLogic, and WebSphere. Normally connections are pooled when run in this context and are associated with a given security context.

As with regular Java JDBC, t

Let's look at these in turn.

zxJDBC.connect

zxJDBC.lookup

In a managed container, you would use zxJDBC.locate instead of zxJDBC.connect. If you have code that needs to run both inside and outside containers, we recommend you use a factory to abstact this.

XXX sample code for factory

This lookup process does not require knowing the JDBC URL or the driver factory class. These aspects, as well as possibly the user name and password, are configured by the adminstrator of the container using tools specific to that container.

XXX - say something about the naming convention here, specifically java:/comp/env/jdbc/NAME; in a container the jndi name is jdbc/NAME, where jdbc is a convention

Transaction isolation levels

### Cursors

Once you have a connection, you probably want to do something with it. Because you can do multiple things within a transaction - query one table, update another - you need one more resource, which is a cursor. Creating a cursor is easy:

with conn.cursor() as c:
do_some_work(c)


Like connections, you want to ensure the resource is appropriately closed. Of course if you are using Jython from the shell, there's generally no need to worry about resource allocations. So you can just do this to follow the shorter examples we will look at:

Cursors represent a unit of work. The easiest way is to use the execute method:

>>> XXX code select from the world database


If the statement is a database query - uses the select statement - then it will return a result set. The best way to work with that result set is to iterate over it:

XXX code


Binding variables

XXX cursors also support prepare

The unit of work is a cursor. Let's now look at how to create and then use cursors.

Create a cursor with the cursor method on the connection:

(zxJDBC maps a cursor to a JDBC ResultSet.)

#### Resource Management

You should always close connections. This is not only good practice but absolutely essential in a managed container so as to avoid exhausting the corresponding connection pool, which needs the connections returned as soon as they are no longer in use. The with statement makes it easy:

from __future__ import with_statement from itertools import islice from com.ziclix.python.sql import zxJDBC

# externalize jdbc_url = "jdbc:oracle:thin:@host:port:sid" [XXX generalize] username = "world" password = "world" driver = "oracle.jdbc.driver.OracleDriver"

with conn.cursor() as c:

c.execute("select * from emp") for row in islice(c, 20):

print row # let's redo this w/ namedtuple momentarily...

(Jython 2.5.0 requires using contextlib.closing; here's how that looks)

The older alternative is available. It's more verbose, and similar to the Java code that would normally have to be written to ensure that the resource is closed. Note that for backwards compatibility, certain defaults have changed.

## Creating and Executing Queries

Here's how to write a query using zxJDBC:

XXX code


Usually you will want to use binding variables in query statements:

XXX code


Note that the qmark is always ? in zxJDBC:

>>> XXX


What else?

Statement Templates

In the past, building one-off SQL query template engines was common in Python. It might still be popular, but rarely necessary - it's much better to use tools like SQLAlchemy or, if it's specifically for object-relational usage, Django. We will discuss both later.

due to unescaped usage.

## Iteration

Older code often used fetchall. In addition, so-called static cursors were used by default. This practice is not recommended. Iteration will generally perform much better, and it does not produce fragile code when used with large data sets.

c.description

Compare with cx_Oracle or pgsql - it would be nice if the datatype that was provided (the constructor), instead of the numeric type - make it so

name, type, display_size, internal_size, precision, scale, null_ok).

Also make certain isNullable is a boolean

## Data Modification Language

### Bulk Inserts, Updates, and Merges

(what is best practice here)

Unicode Numbers Dates and Times XML BLOBs

Reference Cursors

Databases like Oracle support reference cursors [XXX we may only support Oracle, check zxJDBC source - of course we can also use customization to support]. Such reference cursors can be created a number of ways, but perhaps most easily with cursor expressions:

## Extensions

Metadata, lots of metadata! See PyExtendedCursor (so how do I make one of these from jython source?)

Scrollable cursors

Cursors can support a more complex interaction than simple iteration, through scrolling, and this is exposed by zxJDBC. For web applications, this functionality is not at all useful. In others, it is rarely seen. One possible use case is for modestly-scaled applications with a Swing GUI.

[XXX describe more]

Customization

preExecute/postExecute getRowId

And even more; see http://www.jython.org/archive/21/docs/zxjdbc.html which documents the range of this quite extensively.

zxJDBC was contributed by Brian Zimmer, one-time lead committer for Jython.