28. Appendix C: The python-oracledb and cx_Oracle Drivers

The python-oracledb driver is the renamed, major version successor to cx_Oracle 8.3. As a major release, the python-oracledb driver has new features and some Deprecated and Desupported Features. Also see Upgrading from cx_Oracle 8.3 to python-oracledb.

28.1. Differences between the python-oracledb and cx_Oracle Drivers

The differences between the cx_Oracle 8.3 and python-oracledb drivers are listed here.

28.1.1. Mode differences from cx_Oracle

By default, python-oracledb runs in a ‘Thin’ mode which connects directly to Oracle Database. This mode does not need Oracle Client libraries. However, some additional functionality is available when python-oracledb uses them. Python-oracledb is said to be in ‘Thick’ mode when Oracle Client libraries are used. See Enabling python-oracledb Thick mode. Both modes have comprehensive functionality supporting the Python Database API v2.0 Specification.

cx_Oracle always runs in a Thick mode using Oracle Client libraries. The features in python-oracledb Thick mode and cx_Oracle 8.3 are the same, subject to the new features, some deprecations, and to other changes noted in this section.

28.1.2. Oracle Client Library Loading Differences from cx_Oracle

Oracle Client libraries are now only loaded if oracledb.init_oracle_client() is called in your application. This changes python-oracledb to Thick mode. The init_oracle_client() method must be called before any standalone connection or connection pool is created. If a connection or pool is created first in the default Thin mode, then Thick mode cannot be enabled.

See Enabling python-oracledb Thick mode for more information.

Calling the init_oracle_client() method immediately loads Oracle Client libraries. To emulate the cx_Oracle behavior of deferring library loading until the creation of the first connection (in the case when init_oracle_client() is not called), your application will need to explicitly defer calling init_oracle_client() as appropriate.

In python-oracledb, init_oracle_client() can now be called multiple times in the one Python process as long as its arguments are the same each time.

28.1.2.1. oracledb.clientversion()

The oracledb.clientversion() method shows the version of the Oracle Client libraries being used. There is no Oracle Client used in the python-oracledb Thin mode so this function can only be called in python-oracledb Thick mode. If this function is called before oracledb.init_oracle_client(), an exception is thrown.

28.1.3. Connection Differences from cx_Oracle

28.1.3.1. oracledb.connect() Differences

The oracledb.connect() function in the python-oracledb driver differs from cx_Oracle:

  • Keyword parameters must be used in calls to oracledb.connect(). This change makes the driver compliant with the Python Database API specification PEP 249. See Standalone Connections and Common Connection Errors.

  • New optional keyword arguments can be passed to connect(). For example you can pass the hostname, port and servicename as separate parameters instead of using an Easy Connect connection string. In python-oracledb Thin mode, some of the new arguments replace sqlnet.ora settings.

  • A new optional parameter params of type ConnectParams can be used to encapsulate connection properties. See Using the ConnectParams Builder Class for more information.

  • The following parameters are desupported:

    • encoding and nencoding: The encodings in use are always UTF-8.

    • threaded: Threaded Oracle Call Interface (OCI) is now always enabled in Thick mode. This option is not relevant to the Thin mode.

    See Deprecated and Desupported Features for more information.

The use of the class constructor method oracledb.Connection() to create connections is no longer recommended for creating connections. Use connect() instead.

The oracledb.makedsn() method for creating the dsn value has been deprecated. New code should use oracledb.ConnectParams() or use the new keyword arguments in oracledb.connect().

28.1.3.2. Connection Object Differences

The Connection object differences between the python-oracledb and cx_Oracle drivers are:

  • The attribute Connection.maxBytesPerCharacter is deprecated. This will return a constant value of 4 since encodings are always UTF-8.

  • A new boolean attribute, Connection.thin is available. This attribute is True if the connection was established in the Thin mode. In Thick mode, the value of this attribute is False.

  • The new method signature of Connection.outputtypehandler is handler(cursor, metadata). The old signature handler(cursor, name, default_type, length, precision, scale) was deprecated in python-oracledb 1.4 but will still work and will be removed in a future version.

See Connection Attributes for more information.

28.1.4. Pooling Differences from cx_Oracle

It is recommended to use the new ConnectionPool Object instead of the equivalent SessionPool object, which is deprecated. To create a connection pool, use oracledb.create_pool(), which is equivalent to calling cx_Oracle.SessionPool().

For more information, see Connection Pooling.

28.1.4.1. oracledb.SessionPool() Differences

The python-oracledb oracledb.SessionPool() method (which is an alias of oracledb.create_pool()) differs from cx_Oracle.SessionPool() as follows:

  • Keyword parameters must be used in calls. This change makes the driver compliant with the Python Database API specification PEP 249. See Connection pooling and Common Connection Errors.

  • Passing a value to the dsn parameter that contains the user name and password is now supported in the same way as oracledb.connect(). For example dsn="un/pw@cs" can be used.

  • New optional keyword arguments can be passed to create_pool(). For example you can pass the hostname, port and servicename as separate parameters instead of using an Easy Connect connection string. In python-oracledb Thin mode, some of the new arguments replace sqlnet.ora settings.

  • A new optional parameter params of type PoolParams can be used to encapsulate connection properties. See Using the ConnectParams Builder Class for more information.

  • The default mode is POOL_GETMODE_WAIT instead of POOL_GETMODE_NOWAIT. If the mode POOL_GETMODE_NOWAIT is truly desired, modify any pool creation code to specify this value instead. Note the namespace of constants has been improved. Old names like SPOOL_ATTRVAL_NOWAIT can be used but are now deprecated.

  • The encoding and nenecoding parameters are deprecated and ignored. The encodings in use are always UTF-8.

  • New keyword arguments that are used internally to create a PoolParams object before creating the connection.

The oracledb.makedsn() method for creating the dsn value has been deprecated. New code should use oracledb.ConnectParams() or use the new keyword arguments to oracledb.create_pool().

28.1.4.2. SessionPool Object Differences

The SessionPool object (which is an alias for the ConnectionPool object) differences between the python-oracledb and cx_Oracle drivers are:

  • A Python type() will show the class as oracledb.ConnectionPool instead of cx_Oracle.SessionPool.

  • A new boolean attribute, SessionPool.thin (see ConnectionPool.thin) is available. This attribute is True if the connection was established in the Thin mode. In Thick mode, the value of this attribute is False.

28.1.5. Cursor Object Differences from cx_Oracle

The differences between the Cursor object in python-oracledb and cx_Oracle drivers are:

  • Cursor.fetchmany(): The name of the size argument of fetchmany() is size. This change was done to comply with PEP 249. The previous keyword argument name, numRows is deprecated.

  • Cursor.fetchraw(): This method was previously deprecated in cx_Oracle 8.2 and has been removed in python-oracledb. Instead, use one of the other fetch methods such as Cursor.fetchmany().

  • Cursor.executemanyprepared(): This method was previously deprecated in cx_Oracle 6.4 and has been removed in python-oracledb. Instead, use Cursor.executemany(), by passing None for the statement argument and an integer for the parameters argument.

  • Cursor.bindarraysize: This attribute is desupported and removed in python-oracledb. It is not needed in the application code.

  • Cursor.rowcount: After Cursor.execute() or Cursor.executemany() with PL/SQL statements, Cursor.rowcount will return 0. If the cursor or connection are not open, then the value -1 will be returned as required by the Python Database API.

  • Cursor.description: This attribute was previously a sequence of 7-item sequences in cx_Oracle and python-oracledb. Each of these sequences contained information describing one result column, that is, (name, type, display_size, internal_size, precision, scale, null_ok). In python-oracledb 1.4, this attribute was changed to a sequence of FetchInfo objects. Each FetchInfo object describes one result column and can behave as a 7-tuple like before, but contains additional information that may be helpful when using output type handlers.

  • Cursor.outputtypehandler: The new method signature of this attribute is handler(cursor, metadata). The old signature handler(cursor, name, default_type, length, precision, scale) was deprecated in python-oracledb 1.4 but will still work and will be removed in a future version.

28.1.6. Fetching IS JSON Column Differences from cx_Oracle

In python-oracledb, VARCHAR2 and LOB columns that have the IS JSON constraint enabled are fetched as Python objects. These columns are fetched in the same way that JSON type columns are fetched when using Oracle Database 21c (or later). The returned value varies depending on the JSON data. If the JSON data is an object, then a dictionary is returned. If it is an array, then a list is returned. If it is a scalar value, then that particular scalar value is returned.

In cx_Oracle, VARCHAR2 and LOB columns that have the IS JSON constraint enabled are fetched as strings and LOB objects respectively. To enable this same fetch behavior in python-oracledb, you must use an output type handler as shown below.

def type_handler(cursor, fetch_info):
    if fetch_info.is_json:
        return cursor.var(fetch_info.type_code, cursor.arraysize)

28.1.7. Advanced Queuing (AQ) Differences from cx_Oracle

Use the new Advanced Queuing (AQ) API instead of the older API which was deprecated in cx_Oracle 7.2 and is not available in python-oracledb. Note that AQ is only available in python-oracledb Thick mode.

Replace:

The AQ support in python-oracledb has the following enhancements from cx_Oracle:

  • AQ messages can be enqueued and dequeued as a JSON payload type

  • Recipient lists can be enqueued and dequeued

  • Enqueue options, dequeue options, and message properties can be set

See Oracle Advanced Queuing (AQ).

28.1.8. Error Handling Differences from cx_Oracle

In python-oracledb Thick mode, error messages generated by the Oracle Client libraries and the ODPI-C layer used by cx_Oracle and python-oracledb in Thick mode are mostly returned unchanged from cx_Oracle 8.3. Some exceptions shown below.

Note that the python-oracledb driver error messages can also vary between Thin and Thick modes. See Error Handling in Thin and Thick Modes.

28.1.8.1. ConnectionPool.acquire() Message Differences

ConnectionPool.acquire() ORA errors will be mapped to DPY errors. For example:

DPY-4005: timed out waiting for the connection pool to return a connection

replaces the cx_Oracle 8.3 error:

ORA-24459: OCISessionGet() timed out waiting for pool to create new connections

28.1.8.2. Dead Connection Detection and Timeout Message Differences

Application code which detects connection failures or statement execution timeouts will need to check for new errors, DPY-4011 and DPY-4024 respectively. The error DPY-1001 is returned if an already dead connection is attempted to be used.

The new Error object attribute full_code may be useful for checking the error code.

Example error messages are:

  • Scenario 1: An already closed or dead connection was attempted to be used.

    python-oracledb Thin Error:

    DPY-1001: not connected to database
    

    python-oracledb Thick Error:

    DPY-1001: not connected to database
    

    cx_Oracle Error:

    not connected
    
  • Scenario 2: The database side of the connection was terminated while the connection was being used.

    python-oracledb Thin Error:

    DPY-4011: the database or network closed the connection
    

    python-oracledb Thick Error:

    DPY-4011: the database or network closed the connection
    DPI-1080: connection was closed by ORA-%d
    

    cx_Oracle Error:

    DPI-1080: connection was closed by ORA-%d
    
  • Scenario 3: Statement execution exceeded the connection.call_timeout value.

    python-oracledb Thin Error:

    DPY-4024: call timeout of {timeout} ms exceeded
    

    python-oracledb Thick Error:

    DPY-4024: call timeout of {timeout} ms exceeded
    DPI-1067: call timeout of %u ms exceeded with ORA-%d
    

    cx_Oracle Error:

    DPI-1067: call timeout of %u ms exceeded with ORA-%d
    

28.2. Upgrading from cx_Oracle 8.3 to python-oracledb

This section provides the detailed steps needed to upgrade from cx_Oracle 8.3 to python-oracledb.

28.2.1. Things to Know Before the Upgrade

Below is a list of some useful things to know before upgrading from cx_Oracle to python-oracledb:

  • You can have both cx_Oracle and python-oracledb installed, and can use both in the same application.

  • If you only want to use the python-oracledb driver in Thin mode, then you do not need Oracle Client libraries such as from Oracle Instant Client. You only need to install the driver itself:

    python -m pip install oracledb
    

    See Appendix B: Differences between python-oracledb Thin and Thick Modes.

  • The python-oracledb Thin and Thick modes have the same level of support for the Python Database API specification and can be used to connect to on-premises databases and Oracle Cloud databases. However, the python-oracledb Thin mode does not support some of the advanced Oracle Database features such as Application Continuity (AC), Advanced Queuing (AQ), Continuous Query Notification (CQN), and Sharding. See Features Supported for details.

  • python-oracledb can be used in SQLAlchemy, Django, Pandas, and other frameworks and Object-relational Mappers (ORMs). To use python-oracledb in versions of these libraries that don’t have native support for the new name, you can override the use of cx_Oracle with a few lines of code. See Python Frameworks, SQL Generators, and ORMs.

  • python-oracledb connection and pool creation calls require keyword arguments to conform with the Python Database API specification. For example you must use:

    oracledb.connect(user="scott", password=pw, dsn="localhost/orclpdb")
    

    This no longer works:

    oracledb.connect("scott", pw, "localhost/orclpdb")
    
  • The python-oracledb Thin mode ignores all NLS environment variables. It also ignores the ORA_TZFILE environment variable. Thick mode does use these variables. See Character Sets and Globalization for alternatives.

  • To use a tnsnames.ora file in the python-oracledb Thin mode, you must explicitly set the environment variable TNS_ADMIN to the directory containing the file, or set defaults.config_dir, or set the config_dir parameter when connecting.

    Only python-oracledb Thick mode will read sqlnet.ora files. The Thin mode lets equivalent properties be set in the application when connecting.

    Configuration files in a “default” location such as the Instant Client network/admin/ subdirectory, in $ORACLE_HOME/network/admin/, or in $ORACLE_BASE/homes/XYZ/network/admin/ (in a read-only Oracle Database home) are not automatically loaded in Thin mode. Default locations are only automatically searched in Thick mode.

  • To use the python-oracledb Thin mode in an ORACLE_HOME database installation environment, you must use an explicit connection string since the ORACLE_SID, TWO_TASK, and LOCAL environment variables are not used. They are used in Thick mode.

  • This is a major release so some previously deprecated features are no longer available. See Deprecated and Desupported Features.

28.2.2. Steps to Upgrade to python-oracledb

If you are creating new applications, follow Installing python-oracledb and refer to other sections of the documentation for usage information.

To upgrade existing code from cx_Oracle to python-oracledb, perform the following steps:

  1. Install the new python-oracledb module:

    python -m pip install oracledb
    

    See Installing python-oracledb for more details.

  2. Import the new interface module. This can be done in two ways. You can change:

    import cx_Oracle
    

    to:

    import oracledb as cx_Oracle
    

    Alternatively, you can replace all references to the module cx_Oracle with oracledb. For example, change:

    import cx_Oracle
    c = cx_Oracle.connect(...)
    

    to:

    import oracledb
    c = oracledb.connect(...)
    

    Any new code being introduced during the upgrade should aim to use the latter syntax.

  3. Use keyword parameters in calls to oracledb.connect(), oracledb.Connection(), and oracledb.SessionPool().

    You must replace positional parameters with keyword parameters, unless only one parameter is being passed. Python-oracledb uses keyword parameters exclusively unless a DSN containing the user, password, and connect string combined, for example un/pw@cs, is used. This change makes the driver compliant with the Python Database API specification PEP 249.

    For example, the following code will fail:

    c = oracledb.connect("un", "pw", "cs")
    

    and needs to be changed to:

    c = oracledb.connect(user="un", password="pw", dsn="cs")
    

    The following example will continue to work without change:

    c = oracledb.connect("un/pw@cs")
    
  4. Review obsolete encoding parameters in calls to oracledb.connect(), oracledb.Connection(), and oracledb.SessionPool():

    • encoding and nencoding are ignored by python-oracledb. The python-oracledb driver uses UTF-8 exclusively.

    • threaded is ignored in oracledb.connect() and oracledb.Connection() by python-oracledb. This parameter was already ignored in oracledb.SessionPool() from cx_Oracle 8.2.

  5. Remove all references to Cursor.fetchraw() as this method was deprecated in cx_Oracle 8.2 and has been removed in python-oracledb. Instead, use one of the other fetch methods such as Cursor.fetchmany().

  6. The default value of the oracledb.SessionPool() parameter getmode now waits for an available connection. That is, the default is now POOL_GETMODE_WAIT instead of POOL_GETMODE_NOWAIT. The new default value improves the behavior for most applications. If the pool is in the middle of growing, the new value prevents transient connection creation errors from occurring when using the Thin mode, or when using the Thick mode with recent Oracle Client libraries.

    If the old default value is required, modify any pool creation code to explicitly specify getmode=oracledb.POOL_GETMODE_NOWAIT.

    Note a ConnectionPool class deprecates the equivalent SessionPool class. The method oracledb.create_pool() deprecates the use of oracledb.SessionPool(). New pool parameter constant names such as POOL_GETMODE_NOWAIT and PURITY_SELF are now preferred. The old namespaces still work.

  7. The method signature of the output type handler which can be specified on a connection or on a cursor is handler(cursor, metadata). The old signature handler(cursor, name, default_type, length, precision, scale) was deprecated in python-oracledb 1.4 but will still work and will be removed in a future version.

  8. VARCHAR2 and LOB columns that have the IS JSON constraint enabled are fetched by default as Python objects in python-oracledb. In cx_Oracle, VARCHAR2 and LOB columns that contain JSON data are fetched by default as strings and LOB objects respectively. See Fetching IS JSON Column Differences from cx_Oracle.

  9. Review the following sections to see if your application requirements are satisfied by the python-oracledb Thin mode:

    If your application requirements are not supported by the Thin mode, then use the python-oracledb Thick mode.

  10. Review Differences between the python-oracledb and cx_Oracle Drivers.

If your code base uses an older cx_Oracle version, review the previous release notes for additional changes to modernize the code.

  1. Modernize code as needed or desired. See Deprecated and Desupported Features for the list of deprecations in python-oracledb.

28.2.2.1. Additional Upgrade Steps to use python-oracledb Thin Mode

To use python-oracledb Thin mode, the following changes need to be made in addition to the common Steps to Upgrade to python-oracledb:

  1. Remove calls to init_oracle_client() since this turns on the python-oracledb Thick mode.

  2. If the config_dir parameter of init_oracle_client() had been used, then set the new defaults.config_dir attribute to the desired value or set the config_dir parameter when connecting. For example:

    oracledb.defaults.config_dir = "/opt/oracle/config"
    

    Also, see Oracle Net Services and Client Configuration Files.

  3. If the application is connecting using an Oracle Net service name from a tnsnames.ora file located in a “default” location such as the Instant Client network/admin/ subdirectory, in $ORACLE_HOME/network/admin/, or in $ORACLE_BASE/homes/XYZ/network/admin/ (in a read-only Oracle Database home), then the configuration file directory must now explicitly be set as shown above.

  4. Remove calls to oracledb.clientversion() which is only available in the python-oracledb Thick mode. Oracle Client libraries are not available in Thin mode.

  5. Ensure that any assumptions about when connections are created in the connection pool are eliminated. The python-oracledb Thin mode creates connections in a daemon thread and so the attribute ConnectionPool.opened will change over time and will not be equal to ConnectionPool.min immediately after the pool is created. Note that this behavior is also similar in recent versions of the Oracle Call Interface (OCI) Session Pool used by the Thick mode. Unless the oracledb.SessionPool() function’s parameter getmode is oracledb.POOL_GETMODE_WAIT, then applications should not call ConnectionPool.acquire() until sufficient time has passed for connections in the pool to be created.

  6. Review error handling improvements. See Error Handling in Thin and Thick Modes.

  7. Review locale and globalization usage. See Character Sets and Globalization.

28.2.2.2. Additional Upgrade Steps to use python-oracledb Thick Mode

To use python-oracledb Thick mode, the following changes need to be made in addition to the common Steps to Upgrade to python-oracledb:

  1. The function init_oracle_client() must be called. It can be called anywhere before the first call to connect(), oracledb.Connection(), and oracledb.SessionPool(). This enables the Thick mode. See Enabling python-oracledb Thick mode for more details.

    The requirement to call init_oracle_client() means that Oracle Client library loading is not automatically deferred until the driver is first used, such as when a connection is opened. The application must explicitly manage this, if deferral is required. In python-oracledb, init_oracle_client() can be called multiple times in a Python process as long as arguments are the same.

    Note that on Linux and related operating systems, the init_oracle_client() parameter lib_dir should not be passed. Instead, set the system library search path with ldconfig or LD_LIBRARY_PATH prior to running Python.

  2. Replace all usages of the deprecated Advanced Queuing API with the new AQ API originally introduced in cx_Oracle 7.2, see the cx_Oracle Advanced Queuing (AQ) documentation.

  3. Review error handling improvements. See Error Handling in Thin and Thick Modes.

28.2.3. Code to Aid the Upgrade to python-oracledb

28.2.3.1. Toggling between Drivers

The sample oracledb_upgrade.py shows a way to toggle applications between cx_Oracle and the two python-oracledb modes. Note this script cannot map some functionality such as obsolete cx_Oracle features or error message changes.

An example application showing this module in use is:

# test.py

import oracledb_upgrade as cx_Oracle
import os

un = os.environ.get("PYTHON_USERNAME")
pw = os.environ.get("PYTHON_PASSWORD")
cs = os.environ.get("PYTHON_CONNECTSTRING")

connection = cx_Oracle.connect(user=un, password=pw, dsn=cs)
with connection.cursor() as cursor:
    sql = """SELECT UNIQUE CLIENT_DRIVER
             FROM V$SESSION_CONNECT_INFO
             WHERE SID = SYS_CONTEXT('USERENV', 'SID')"""
    for r, in cursor.execute(sql):
        print(r)

You can then choose what mode is in use by setting the environment variable ORA_PYTHON_DRIVER_TYPE to one of “cx”, “thin”, or “thick”:

export ORA_PYTHON_DRIVER_TYPE=thin
python test.py

Output shows the python-oracledb Thin mode was used:

python-oracledb thn : 1.0.0

You can customize oracledb_upgrade.py to your needs. For example, if your connection and pool creation calls always use keyword parameters, you can remove the shims that map from positional arguments to keyword arguments.

The simplest form is shown in Python Frameworks, SQL Generators, and ORMs.

28.2.3.2. Testing Which Driver is in Use

To know whether the driver is cx_Oracle or python-oracledb, you can use code similar to:

import oracledb as cx_Oracle
# or:
# import cx_Oracle

if cx_Oracle.__name__ == 'cx_Oracle':
       print('cx_Oracle')
else:
       print('oracledb')

Another method that can be used to check which driver is in use is to query V$SESSION_CONNECT_INFO, see Finding the python-oracledb Mode.

28.2.4. Python Frameworks, SQL Generators, and ORMs

Python-oracledb’s Thin and Thick modes cover the feature needs of frameworks that depend upon the Python Database API.

For versions of SQLAlchemy, Django, other frameworks, object-relational mappers (ORMs), and libraries that don’t have native support for python-oracledb, you can add code like this to use python-oracledb in-place of cx_Oracle:

import sys
import oracledb
oracledb.version = "8.3.0"
sys.modules["cx_Oracle"] = oracledb
import cx_Oracle

Note

The import of cx_Oracle occurs last. This code must be run before the library code does its own import of cx_Oracle.

SQLAlchemy 2 and Django 5 have native support for python-oracledb so this code snippet is not needed in those versions.

To use Thick mode, an additional call to init_oracle_client() is needed, see Enabling python-oracledb Thick mode.