An overview of new features in cx_Oracle, the standard Python extension for accessing the Oracle Database
By Anthony Tuininga
Published April 2009
cx_Oracle is a Python extension module that enables querying and updating of Oracle databases using a database API that is common to all database access modules. A number of extensions to the common database API have also been included in order to take advantage of some of the features available exclusively to Oracle.
cx_Oracle has been in development for more than ten years and has covered most of the requirements of those needing access to Oracle in Python. In December 2008, however, a major new release addressed a number of limitations in earlier versions and added support for Python 3.0 and some new features recently made available in Oracle. This article aims to cover the new features in depth and give reasons why you might want to use them in your code.
In earlier versions of cx_Oracle, no help was given to those wishing to use Unicode strings in their code. Code would have to be written something like that shown in Listing 1.Listing 1: Old-style Unicode handling in cx_Oracle 4.x
import cx_Oracle import os os.environ["NLS_LANG"] = ".AL32UTF8" START_VALUE = u"Unicode \u3042 3" END_VALUE = u"Unicode \u3042 6" connection = cx_Oracle.Connection("cx_Oracle/dev@t11g") cursor = connection.cursor() cursor.execute(""" select UnicodeCol from TestUnicodes where UnicodeCol > :startValue and UnicodeCol < :endValue""", startValue = START_VALUE.encode(connection.nencoding), endValue = END_VALUE.encode(connection.nencoding)) for rawValue, in cursor: actualValue = rawValue.decode(connection.nencoding) print "Actual Value:", actualValueNote that any time data was passed to Oracle Database, it would have to be encoded into the client character set; any time data was retrieved from the database it would have to be decoded from the client character set. Although clearly code could be written to handle Unicode strings properly, it was messy and prone to error. With the advent of Python 3.0, in which all strings are Unicode strings, cx_Oracle obviously needed to adapt. Just as clearly, the changes to the Python 3.x C API were significant enough to make introducing support for Python 3.x and Unicode strings at the same time difficult. For this reason, support for Unicode strings was first added to Python 2.x in a form of Unicode mode, and then support for Python 3.x was added later.
In Unicode mode, all strings passed to Oracle are expected to be in Unicode, and all strings returned are also in Unicode. This not only includes bind variables and result sets from queries but also statement text, connection parameters, and attribute values, just as would take place in Python 3.x. Unicode mode is a compile time option that is disabled by default in Python 2.x but can be enabled by setting the environment variable WITH_UNICODE before building the module.
Regardless of which mode is in use, all data represented by the national character set (NCLOB and NVARCHAR2 columns, for example) is now returned in Unicode and expected to be bound in Unicode. This means that the code sample shown in Listing 1 can be simplified by removing all encoding and decoding operations. Listing 2 demonstrates this.Listing 2: Simplified handling of Unicode in cx_Oracle 5.0
import cx_Oracle START_VALUE = u"Unicode \u3042 3" END_VALUE = u"Unicode \u3042 6" connection = cx_Oracle.Connection("cx_Oracle/dev@t11g") cursor = connection.cursor() cursor.execute(""" select UnicodeCol from TestUnicodes where UnicodeCol > :startValue and UnicodeCol < :endValue""", startValue = START_VALUE, endValue = END_VALUE) for actualValue, in cursor: print "Actual Value:", actualValue
A second major feature of cx_Oracle 5.x is the introduction of input and output type handlers. With these in place, a great deal of flexibility and customizability is now possible. For example, if it was considered desirable to have all strings returned as Unicode, the code in Listing 3 could be used. Listing 3: How to return all strings as Unicode strings
import cx_Oracle def OutputTypeHandler(cursor, name, defaultType, size, precision, scale): if defaultType in (cx_Oracle.STRING, cx_Oracle.FIXED_CHAR): return cursor.var(unicode, size, cursor.arraysize) connection = cx_Oracle.Connection("cx_Oracle/dev@t11g") connection.outputtypehandler = OutputTypeHandler cursor = connection.cursor() cursor.execute(""" select * from TestStrings where StringCol > :startValue and StringCol < :endValue""", startValue = u"String 3", endValue = u"String 6") for row in cursor: print "Row:", row
An input type handler method is called whenever a bind variable needs to be created to pass data from Python into Oracle. The cursor, the value being bound, and the number of elements that need to be created are passed to the method. The return value is expected to be a variable object or None, meaning that default processing will take place. The variable object created should be able to bind the Python value. In order to help in this regard, variable objects can now be created with input converters. These methods are called with the Python value being bound and expect in return a Python value that cx_Oracle already knows how to bind directly. One final point: calling PL/SQL procedures and functions or anonymous PL/SQL blocks introduces the possibility of input/output or even output-only bind variables. These are still covered by input type handlers, even though data can be returned to Python from Oracle.
An output type handler method is called whenever a bind variable needs to be created to return data to Python from Oracle after a query has been executed. In this case, the cursor; the name of the column being returned; the default variable type; and the size, precision, and scale of the column being returned are passed to the method. As with input type handlers, the return value is expected to be a variable object or None, meaning that default processing will take place. If further manipulation of the data returned to Python is desired, an output converter can be set on the variable. These methods are called with the value that would otherwise be returned to Python and return the value that will actually be returned to Python.
A couple more examples should make clear how powerful this enhancement can be in controlling the interface between Python and Oracle. Listing 4 demonstrates how to bind an arbitrary object to a cursor. Listing 5 demonstrates how to return numbers as decimal.Decimal objects or as strings in order to manage numbers that are much larger than floating precision numbers can handle.
Suppose you need to upload a couple of binary large objects (BLOBs) to the database and save them to the blob_tab table that you might have created in a custom database schema, as follows:Listing 4: Binding an arbitrary Python object to a cursor
import cx_Oracle class ArbitraryObject(object): def __init__(self, intValue, someOtherData): self.intValue = intValue self.someOtherData = someOtherData def BindValue(self): return self.intValue def InputTypeHandler(cursor, value, numElements): if isinstance(value, ArbitraryObject): return cursor.var(int, arraysize = numElements, inconverter = ArbitraryObject.BindValue) connection = cx_Oracle.Connection("cx_Oracle/dev@t11g") connection.inputtypehandler = InputTypeHandler cursor = connection.cursor() cursor.execute(""" select * from TestNumbers where IntCol = :obj""", obj = ArbitraryObject(1, "arbitrary")) for row in cursor: print "Row:", row
import cx_Oracle import decimal def NumbersAsDecimal(cursor, name, defaultType, size, precision, scale): if defaultType == cx_Oracle.NUMBER: return cursor.var(str, 100, cursor.arraysize, outconverter = decimal.Decimal) def NumbersAsString(cursor, name, defaultType, size, precision, scale): if defaultType == cx_Oracle.NUMBER: return cursor.var(str, 100, cursor.arraysize) connection = cx_Oracle.Connection("cx_Oracle/dev@t11g") connection.outputtypehandler = NumbersAsDecimal cursor = connection.cursor() cursor.execute(""" select * from TestNumbers where IntCol < 3""") for row in cursor: print "Row (as decimal):", row cursor = connection.cursor() cursor.outputtypehandler = NumbersAsString cursor.execute(""" select * from TestNumbers where IntCol between 6 and 8""") for row in cursor: print "Row (as string):", row
In Oracle Database 11g Release 1, Oracle introduced Database Resident Connection Pooling (DRCP, see the Technical White Paper). Unlike session pooling, where connections are pooled in the client and are only effective within a single threaded process, DRCP pools connections on the server, enabling pooling across multiple client processes. This enhancement is particularly useful with Web servers such as Apache that are frequently set up to spawn a process for each hit. The performance increase gained and the resource requirements reduced by using DRCP in this situation can be quite dramatic. Support for this feature was added to PHP in 2008; with version 5, cx_Oracle adds support for this feature to Python as well.
Once the DRCP pool has been started on the database and a tnsnames.ora entry augmented with SERVER=POOLED, use of this feature within Python is very straightforward. The only difference between creating a normal connection and creating a connection using DRCP is the use of the parameter cclass. This parameter is a string that specifies the connection class, a logical grouping of servers. Connections made using the same connection class may be shared, whereas connections made using different connection classes will never be shared. One other parameter available in the connection constructor that influences whether or not an existing connection in the pool can be used is the purity parameter. If this value is set to cx_Oracle.ATTR_PURITY_SELF (which is also the default value), then existing sessions in the pool will be used if available. If this value is set to cx_Oracle.ATTR_PURITY_NEW, however, then a new session will be created.
Once a connection has been created, it can be used in exactly the same way as any other connection. When the connection is closed (or the connection object is garbage collected), the session is returned to the pool for possible reuse. Listing 6 shows a simple example of creating connections using DRCP.Listing 6: Creating connections with DRCP
import cx_Oracle o connection = cx_Oracle.Connection("cx_Oracle/dev@t11g_drcp", cclass = "TEST1.CLASS") connection2 = cx_Oracle.Connection("cx_Oracle/dev@t11g_drcp", cclass = "TEST2.CLASS", purity = cx_Oracle.ATTR_PURITY_NEW)
Another feature of Oracle Database, initially introduced in Oracle Database 10g Release 2 and enhanced further in Oracle Database 11g Release 1, is Database Change Notification. This feature enables client applications to register queries with the database and receive notification of data manipulation language (DML) or data definition language (DDL) changes to the tables underlying those queries when the transaction making those changes is committed. Notification can be done via e-mail, HTTP URL PL/SQL procedure, or callback. cx_Oracle currently only supports the callback method but the other methods will be supported in future releases.
In order to get started with database change notification, a subscription object must be created. This object describes the conditions under which notification takes place and the action to take when those conditions are met. A subscription object is created by calling the subscribe method on the connection. The parameters to this method control the behavior of the subscription and cannot be changed once the subscription has been created. Any number of subscription objects can be created, however, each with different characteristics. The values of these characteristics are available as read-only attributes on the created subscription object for convenience.
The only parameter that is required for any useful work to be done is the callback parameter. It specifies a callable object that accepts a single argument: a message object that is generated during the callback and describes the event that triggered the notification (more on that later). All of the other parameters are optional, and their default values are reasonable for most uses.
In order to filter the operations on which notifications will be sent, use the operations parameter which is constructed from a bitwise-include OR of flags from the table below. The following table describes the various options and their meanings. The values can be "or’ed" together in order to cover multiple operations at the same time.
The table has been altered.
Rows have been deleted from the table.
The table has been dropped.
Rows have been inserted into the table.
Rows have been updated in the table.
The final parameter that can have a meaningful value is called rowids and is a Boolean value specifying whether or not the rowids of the affected rows should be included in the notification messages that are sent. The other two parameters (namespace and protocol) are reserved for future expansion and should be left at their default values.
Once a subscription object has been created, the final step is to register queries, each of which will result in notifications sent for each table underlying those queries that is changed. A single message is sent for each transaction that changes the registered tables. Queries are registered by calling the registerquery method, which behaves similarly to the execute method on a cursor. It accepts the SQL for the query and optionally a list of arguments. In the current version, the arguments have no bearing on the messages that are sent. In a future version, cx_Oracle will add support for query change notification (available in Oracle Database 11g Release 1) and then the arguments can be important.
Once queries have been registered on the subscription object, callbacks will take place every time one of the requested operations takes place on a registered table. This takes place in a separate thread, so care should be taken to avoid using the same connection in two threads at the same time. The message object that is passed to the callback has three attributes describing the event that took place. The first is called dbname and gives the name of the database that generated the notification. The second is type and describes the type of event that took place, but its value in this version is always the same: cx_Oracle.EVENT_OBJCHANGE. The third parameter is called tables and is a list of message table objects.
Message table objects have these attributes: name (the name of the table affected), operation (the operation that took place on the table), and rows. The rows attribute is a list of message row objects if the value of the rowids parameter when creating the subscription object was true. The message row object has two attributes: operation (the operation that took place on that row) and rowid (contains the rowid of the row that was affected).
Listing 7 is an example of how to use database change notification in cx_Oracle. Once the code is running, insert, update, or delete rows in the table cx_Oracle.TestExecuteMany to see the notifications taking place.Listing 7: Notification Demonstration
import cx_Oracle def OperationToString(operation): operations =  if operation & cx_Oracle.OPCODE_INSERT: operations.append("insert") if operation & cx_Oracle.OPCODE_DELETE: operations.append("delete") if operation & cx_Oracle.OPCODE_UPDATE: operations.append("update") if operation & cx_Oracle.OPCODE_ALTER: operations.append("alter") if operation & cx_Oracle.OPCODE_DROP: operations.append("drop") if operation & cx_Oracle.OPCODE_ALLOPS: operations.append("all operations") return ", ".join(operations) def OnChanges(message): print "Message received" print " Database Name:", message.dbname print " Tables:" for table in message.tables: print " Name:", table.name, print " Operations:", print OperationToString(table.operation) if table.rows is None \ or table.operation & cx_Oracle.OPCODE_ALLROWS: print " Rows: all rows" else: print " Rows:" for row in table.rows: print " Rowid:", row.rowid print " Operation:", print OperationToString(row.operation) connection = cx_Oracle.Connection("cx_Oracle/dev@t11g", events = True) sql = "select * from TestExecuteMany" subscriptionAll = connection.subscribe(callback = OnChanges) subscriptionAll.registerquery(sql) subscriptionInsertUpdate = \ connection.subscribe(callback = OnChanges, operations = cx_Oracle.OPCODE_INSERT | \ cx_Oracle.OPCODE_UPDATE, rowids = True) subscriptionInsertUpdate.registerquery(sql) raw_input("Hit enter to terminate...\n")
With the release of version 5, cx_Oracle has kept pace with the ongoing development of Python, enabled greater flexibility and customization for those requiring it, and added support for a number of exciting new features in Oracle Database.
Anthony Tuininga is an Oracle DBA and developer based in Edmonton, Alberta. He is the author of cx_Oracle, the defacto standard for accessing Oracle databases with Python.