5. Executing SQL

Executing SQL statements is the primary way in which a Python application communicates with Oracle Database. Statements include queries, Data Manipulation Language (DML), and Data Definition Language (DDL). A few other specialty statements can also be executed. Statements are executed using one of these methods Cursor.execute(), Cursor.executemany(), Connection.fetch_df_all(), Connection.fetch_df_batches(), AsyncCursor.execute(), AsyncCursor.executemany(), AsyncConnection.execute(), AsyncConnection.executemany(), AsyncConnection.fetch_df_all(), AsyncConnection.fetch_df_batches(), or AsyncConnection.run_pipeline().

This chapter discusses python-oracledb’s synchronous methods. The asynchronous methods and pipelining functionality are discussed in detail in Concurrent Programming with asyncio and Pipelining.

PL/SQL statements are discussed in Executing PL/SQL. Other chapters contain information on specific data types and features. See Executing Batch Statements and Bulk Loading, Using CLOB, BLOB, NCLOB, and BFILE Data, Using JSON Data, and Using XMLTYPE Data.

Python-oracledb can be used to execute individual statements, one at a time. Once a statement has finished execution, only then will the next statement execute. If you try to execute statements concurrently in a single connection, the statements are queued and run consecutively in the order they are executed in the application code.

Python-oracledb does not read SQL*Plus “.sql” files. To read SQL files, use a technique like the one in run_sql_script() in samples/sample_env.py.

SQL statements should not contain a trailing semicolon (“;”) or forward slash (“/”). This will fail:

cursor.execute("select * from MyTable;")   # fails due to semicolon

This is correct:

cursor.execute("select * from MyTable")

5.1. SQL Queries

Queries (statements beginning with SELECT or WITH) can be executed using the method Cursor.execute(). Rows can then be iterated over, or can be fetched using one of the methods Cursor.fetchone(), Cursor.fetchmany() or Cursor.fetchall(). There is a default type mapping to Python types that can be optionally overridden.

Important

Interpolating or concatenating user data with SQL statements, for example cursor.execute("SELECT * FROM mytab WHERE mycol = '" + myvar + "'") is a security risk and impacts performance. Use bind variables instead, for example cursor.execute("SELECT * FROM mytab WHERE mycol = :mybv", mybv=myvar).

5.1.1. Fetch Methods

Rows can be fetched in various ways.

• After Cursor.execute(), the cursor is returned as a convenience. This allows code to iterate over rows like:

  cursor = connection.cursor()
  for row in cursor.execute("select * from MyTable"):
      print(row)
  

• Rows can also be fetched one at a time using the method Cursor.fetchone():

  cursor = connection.cursor()
  cursor.execute("select * from MyTable")
  while True:
      row = cursor.fetchone()
      if row is None:
          break
      print(row)
  

• If rows need to be processed in batches, the method Cursor.fetchmany() can be used. The size of the batch is controlled by the size parameter, which defaults to the value of Cursor.arraysize.

  cursor = connection.cursor()
  cursor.execute("select * from MyTable")
  num_rows = 10
  while True:
      rows = cursor.fetchmany(size=num_rows)
      if not rows:
          break
      for row in rows:
          print(row)
  

  Note the size parameter only affects the number of rows returned to the application, not to the internal buffer size used for tuning fetch performance. That internal buffer size is controlled only by changing Cursor.arraysize, see Tuning Fetch Performance.

• If all of the rows need to be fetched and can be contained in memory, the method Cursor.fetchall() can be used.

  cursor = connection.cursor()
  cursor.execute("select * from MyTable")
  rows = cursor.fetchall()
  for row in rows:
      print(row)
  

  The fetch methods return data as tuples. To return results as dictionaries, see Changing Query Results with Rowfactories.

• Data can also be fetched in Arrow data format, see Fetching Data Frames.

5.1.2. Closing Cursors

Once cursors are no longer needed, they should be closed in order to reclaim resources in the database. Note cursors may be used to execute multiple statements before being closed.

Cursors can be closed in various ways:

• A cursor will be closed automatically when the variable referencing it goes out of scope (and no further references are retained). A with context manager block is a convenient and preferred way to ensure this. For example:

  with connection.cursor() as cursor:
      for row in cursor.execute("select * from MyTable"):
          print(row)
  

  This code ensures that once the block is completed, the cursor is closed and database resources can be reclaimed. In addition, any attempt to use the variable cursor outside of the block will fail.

• Cursors can be explicitly closed by calling Cursor.close():

  cursor = connection.cursor()
  
  ...
  
  cursor.close()
  

5.1.3. Query Column Metadata

After executing a query, the column metadata such as column names and data types can be obtained using Cursor.description:

with connection.cursor() as cursor:
    cursor.execute("select * from MyTable")
    for column in cursor.description:
        print(column)

This could result in metadata like:

('ID', <class 'oracledb.DB_TYPE_NUMBER'>, 39, None, 38, 0, 0)
('NAME', <class 'oracledb.DB_TYPE_VARCHAR'>, 20, 20, None, None, 1)

To extract the column names from a query you can use code like:

with connection.cursor() as cursor:
    cursor.execute("select * from locations")
    columns = [col.name for col in cursor.description]
    print(columns)
    for r in cursor:
        print(r)

This will print:

['LOCATION_ID', 'STREET_ADDRESS', 'POSTAL_CODE', 'CITY', 'STATE_PROVINCE', 'COUNTRY_ID']
(1000, '1297 Via Cola di Rie', '00989', 'Roma', None, 'IT')
(1100, '93091 Calle della Testa', '10934', 'Venice', None, 'IT')
. . .

Changing Column Names to Lowercase

To change all column names to lowercase you could do:

cursor.execute("select * from locations where location_id = 1000")

columns = [col.name.lower() for col in cursor.description]
print(columns)

The output is:

['location_id', 'street_address', 'postal_code', 'city', 'state_province',
'country_id']

5.1.4. Fetch Data Types

The following table provides a list of all of the data types that python-oracledb knows how to fetch. The middle column gives the type that is returned in the query metadata. The last column gives the type of Python object that is returned by default. Python types can be changed with Output Type Handlers.

Oracle Database Type	oracledb Database Type	Default Python type
BFILE	oracledb.DB_TYPE_BFILE	oracledb.LOB
BINARY_DOUBLE	oracledb.DB_TYPE_BINARY_DOUBLE	float
BINARY_FLOAT	oracledb.DB_TYPE_BINARY_FLOAT	float
BLOB	oracledb.DB_TYPE_BLOB	oracledb.LOB
CHAR	oracledb.DB_TYPE_CHAR	str
CLOB	oracledb.DB_TYPE_CLOB	oracledb.LOB
CURSOR	oracledb.DB_TYPE_CURSOR	oracledb.Cursor
DATE	oracledb.DB_TYPE_DATE	datetime.datetime
INTERVAL DAY TO SECOND	oracledb.DB_TYPE_INTERVAL_DS	datetime.timedelta
INTERVAL YEAR TO MONTH	oracledb.DB_TYPE_INTERVAL_YM	oracledb.IntervalYM
JSON	oracledb.DB_TYPE_JSON	dict, list or a scalar value [4]
LONG	oracledb.DB_TYPE_LONG	str
LONG RAW	oracledb.DB_TYPE_LONG_RAW	bytes
NCHAR	oracledb.DB_TYPE_NCHAR	str
NCLOB	oracledb.DB_TYPE_NCLOB	oracledb.LOB
NUMBER	oracledb.DB_TYPE_NUMBER	float or int [1]
NVARCHAR2	oracledb.DB_TYPE_NVARCHAR	str
OBJECT [3]	oracledb.DB_TYPE_OBJECT	oracledb.Object
RAW	oracledb.DB_TYPE_RAW	bytes
ROWID	oracledb.DB_TYPE_ROWID	str
TIMESTAMP	oracledb.DB_TYPE_TIMESTAMP	datetime.datetime
TIMESTAMP WITH LOCAL TIME ZONE	oracledb.DB_TYPE_TIMESTAMP_LTZ	datetime.datetime [2]
TIMESTAMP WITH TIME ZONE	oracledb.DB_TYPE_TIMESTAMP_TZ	datetime.datetime [2]
UROWID	oracledb.DB_TYPE_ROWID, oracledb.DB_TYPE_UROWID	str
VARCHAR2	oracledb.DB_TYPE_VARCHAR	str

[1]

If the precision and scale obtained from query column metadata indicate that the value can be expressed as an integer, the value will be returned as an int. If the column is unconstrained (no precision and scale specified), the value will be returned as a float or an int depending on whether the value itself is an integer. In all other cases the value is returned as a float.

[2] (1,2)

The timestamps returned are naive timestamps without any time zone information present.

[3]

These include all user-defined types such as VARRAY, NESTED TABLE, etc.

[4]

If the JSON is an object, then a dict 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.

5.1.5. Changing Fetched Data

Data returned by python-oracledb queries can be changed by using output type handlers, by using “outconverters”, or by using row factories.

5.1.5.1. Changing Fetched Data Types with Output Type Handlers

Sometimes the default conversion from an Oracle Database type to a Python type must be changed in order to prevent data loss or to fit the purposes of the Python application. In such cases, an output type handler can be specified for queries. This asks the database to do a conversion from the column type to a different type before the data is returned from the database to python-oracledb. If the database does not support such a mapping, an error will be returned. Output type handlers only affect query output and do not affect values returned from Cursor.callfunc() or Cursor.callproc().

Output type handlers can be specified on a connection or on a cursor. If specified on a cursor, fetch type handling is only changed on that particular cursor. If specified on a connection, all cursors created by that connection will have their fetch type handling changed.

The output type handler is expected to be a function with the following signature:

handler(cursor, metadata)

The metadata parameter is a FetchInfo object, which is the same value found in Cursor.description.

The function is called once for each column that is going to be fetched. The function is expected to return a variable object (generally by a call to Cursor.var()) or the value None. The value None indicates that the default type should be used.

For example:

def output_type_handler(cursor, metadata):
    if metadata.type_code is oracledb.DB_TYPE_NUMBER:
        return cursor.var(oracledb.DB_TYPE_VARCHAR, arraysize=cursor.arraysize)

This output type handler is called once for each column in the SELECT query. For each numeric column, the database will now return a string representation of each row’s value. Using it in a query:

cursor.outputtypehandler = output_type_handler

cursor.execute("select 123 from dual")
r = cursor.fetchone()
print(r)

prints ('123',) showing the number was converted to a string. Without the type handler, the output would have been (123,).

When creating variables using Cursor.var() in a handler, the arraysize parameter should be the same as the Cursor.arraysize of the query cursor. In python-oracledb Thick mode, the query (and var()) arraysize multiplied by the byte size of the particular column must be less than INT_MAX.

To unset an output type handler, set it to None. For example if you had previously set a type handler on a cursor, you can remove it with:

cursor.outputtypehandler = None

Other examples of output handlers are shown in Fetched Number Precision, Fetching LOBs as Strings and Bytes, and Fetching Raw Data. Also see samples such as samples/type_handlers_json_strings.py.

5.1.5.2. Changing Query Results with Outconverters

Python-oracledb “outconverters” can be used with output type handlers to change returned data.

For example, to convert numbers to strings:

def output_type_handler(cursor, metadata):

    def out_converter(d):
        if isinstance(d, str):
            return f"{d} was a string"
        else:
            return f"{d} was not a string"

    if metadata.type_code is oracledb.DB_TYPE_NUMBER:
        return cursor.var(oracledb.DB_TYPE_VARCHAR,
             arraysize=cursor.arraysize, outconverter=out_converter)

The output type handler is called once for each column in the SELECT query. For each numeric column, the database will now return a string representation of each row’s value, and the outconverter will be called for each of those values.

Using it in a query:

cursor.outputtypehandler = output_type_handler

cursor.execute("select 123 as col1, 'abc' as col2 from dual")
for r in cursor.fetchall():
    print(r)

prints:

('123 was a string', 'abc')

This shows that the number was first converted to a string by the database, as requested in the output type handler. The out_converter function then appended “was a string” to the data before the value was returned to the application.

Note outconverters are not called for NULL data values unless the Cursor.var() parameter convert_nulls is True.

Another example of an outconverter is shown in fetching VECTORs as lists.

5.1.5.3. Changing Query Results with Rowfactories

Python-oracledb “rowfactories” are methods called for each row retrieved from the database. The Cursor.rowfactory() method is called with the tuple fetched from the database before it is returned to the application. The method can convert the tuple to a different value.

Fetching Rows as Dictionaries

For example, to fetch each row of a query as a dictionary:

cursor.execute("select * from locations where location_id = 1000")

columns = [col.name for col in cursor.description]
cursor.rowfactory = lambda *args: dict(zip(columns, args))
data = cursor.fetchone()
print(data)

The output is:

{'LOCATION_ID': 1000, 'STREET_ADDRESS': '1297 Via Cola di Rie',
'POSTAL_CODE': '00989', 'CITY': 'Roma', 'STATE_PROVINCE': None,
'COUNTRY_ID': 'IT'}

Also see how JSON_OBJECT is used in Using JSON Data.

If you join tables where the same column name occurs in both tables with different meanings or values, then use a column alias in the query. Otherwise, only one of the similarly named columns will be included in the dictionary:

select
    cat_name,
    cats.color as cat_color,
    dog_name,
    dogs.color
from cats, dogs

Example with an Output Type Handler, Outconverter, and Row Factory

An example showing an output type handler, an outconverter, and a row factory is:

def output_type_handler(cursor, metadata):

    def out_converter(d):
        if type(d) is str:
            return f"{d} was a string"
        else:
            return f"{d} was not a string"

    if metadata.type_code is oracledb.DB_TYPE_NUMBER:
        return cursor.var(oracledb.DB_TYPE_VARCHAR,
            arraysize=cursor.arraysize, outconverter=out_converter)

cursor.outputtypehandler = output_type_handler

cursor.execute("select 123 as col1, 'abc' as col2 from dual")

columns = [col.name.lower() for col in cursor.description]
cursor.rowfactory = lambda *args: dict(zip(columns, args))
for r in cursor.fetchall():
    print(r)

The database converts the number to a string before it is returned to python-oracledb. The outconverter appends “was a string” to this value. The column names are converted to lowercase. Finally, the row factory changes the complete row to a dictionary. The output is:

{'col1': '123 was a string', 'col2': 'abc'}

5.1.6. Fetched Number Precision

Oracle Database uses decimal numbers and these cannot be converted seamlessly to binary number representations like Python floats. In addition, the range of Oracle numbers exceeds that of floating point numbers. Python has decimal objects which do not have these limitations. In python-oracledb you can set defaults.fetch_decimals so that Decimals are returned to the application, ensuring that numeric precision is not lost when fetching certain numbers.

The following code sample demonstrates the issue:

cursor.execute("create table test_float (X number(5, 3))")
cursor.execute("insert into test_float values (7.1)")

cursor.execute("select * from test_float")
val, = cursor.fetchone()
print(val, "* 3 =", val * 3)

This displays 7.1 * 3 = 21.299999999999997

Using Python decimal objects, however, there is no loss of precision:

oracledb.defaults.fetch_decimals = True

cursor.execute("select * from test_float")
val, = cursor.fetchone()
print(val, "* 3 =", val * 3)

This displays 7.1 * 3 = 21.3

See samples/return_numbers_as_decimals.py

An equivalent, longer, older coding idiom to defaults.fetch_decimals is to use an output type handler do the conversion.

import decimal

def number_to_decimal(cursor, metadata):
    if metadata.type_code is oracledb.DB_TYPE_NUMBER:
        return cursor.var(decimal.Decimal, arraysize=cursor.arraysize)

cursor.outputtypehandler = number_to_decimal

cursor.execute("select * from test_float")
val, = cursor.fetchone()
print(val, "* 3 =", val * 3)

This displays 7.1 * 3 = 21.3

The Python decimal.Decimal converter gets called with the string representation of the Oracle number. The output from decimal.Decimal is returned in the output tuple.

5.1.7. Scrollable Cursors

Scrollable cursors enable applications to move backwards, forwards, to skip rows, and to move to a particular row in a query result set. The result set is cached on the database server until the cursor is closed. In contrast, regular cursors are restricted to moving forward.

A scrollable cursor is created by setting the parameter scrollable=True when creating the cursor. The method Cursor.scroll() is used to move to different locations in the result set.

Examples are:

cursor = connection.cursor(scrollable=True)
cursor.execute("select * from ChildTable order by ChildId")

cursor.scroll(mode="last")
print("LAST ROW:", cursor.fetchone())

cursor.scroll(mode="first")
print("FIRST ROW:", cursor.fetchone())

cursor.scroll(8, mode="absolute")
print("ROW 8:", cursor.fetchone())

cursor.scroll(6)
print("SKIP 6 ROWS:", cursor.fetchone())

cursor.scroll(-4)
print("SKIP BACK 4 ROWS:", cursor.fetchone())

See samples/scrollable_cursors.py for a runnable example.

5.1.8. Fetching Oracle Database Objects and Collections

Oracle Database named object types and user-defined types can be fetched directly in queries. Each item is represented as a Python object corresponding to the Oracle Database object. This Python object can be traversed to access its elements. Attributes including DbObjectType.name and DbObjectType.iscollection, and methods including DbObject.aslist() and DbObject.asdict() are available.

For example, if a table MYGEOMETRYTAB contains a column GEOMETRY of Oracle’s predefined Spatial object type SDO_GEOMETRY, then it can be queried and printed:

cursor.execute("select geometry from mygeometrytab")
for obj, in cursor:
    dumpobject(obj)

Where dumpobject() is defined as:

def dumpobject(obj, prefix = ""):
    if obj.type.iscollection:
        print(prefix, "[")
        for value in obj.aslist():
            if isinstance(value, oracledb.Object):
                dumpobject(value, prefix + "  ")
            else:
                print(prefix + "  ", repr(value))
        print(prefix, "]")
    else:
        print(prefix, "{")
        for attr in obj.type.attributes:
            value = getattr(obj, attr.name)
            if isinstance(value, oracledb.Object):
                print(prefix + "   " + attr.name + ":")
                dumpobject(value, prefix + "  ")
            else:
                print(prefix + "   " + attr.name + ":", repr(value))
        print(prefix, "}")

This might produce output like:

{
  SDO_GTYPE: 2003
  SDO_SRID: None
  SDO_POINT:
  {
    X: 1
    Y: 2
    Z: 3
  }
  SDO_ELEM_INFO:
  [
    1
    1003
    3
  ]
  SDO_ORDINATES:
  [
    1
    1
    5
    7
  ]
}

Other information on using Oracle objects is in Using Bind Variables.

Performance-sensitive applications should consider using scalar types instead of objects. If you do use objects, avoid calling Connection.gettype() unnecessarily, and avoid objects with large numbers of attributes.

5.1.9. Fetching Data Frames

Python-oracledb can fetch directly to data frames that expose an Apache Arrow PyCapsule Interface. This can reduce application memory requirements and allow zero-copy data interchanges between Python data frame libraries. It is an efficient way to work with data using Python libraries such as Apache PyArrow, Pandas, Polars, NumPy, PyTorch, or to write files in Apache Parquet format.

Note

The data frame support in python-oracledb 3.1 is a pre-release and may change in a future version.

The method Connection.fetch_df_all() fetches all rows from a query. The method Connection.fetch_df_batches() implements an iterator for fetching batches of rows. The methods return OracleDataFrame objects.

For example, to fetch all rows from a query and print some information about the results:

sql = "select * from departments"
# Adjust arraysize to tune the query fetch performance
odf = connection.fetch_df_all(statement=sql, arraysize=100)

print(odf.column_names())
print(f"{odf.num_columns()} columns")
print(f"{odf.num_rows()} rows")

With Oracle Database’s standard DEPARTMENTS table, this would display:

['DEPARTMENT_ID', 'DEPARTMENT_NAME', 'MANAGER_ID', 'LOCATION_ID']
4 columns
27 rows

Summary of Converting OracleDataFrame to Other Data Frames

To do more extensive operations, OracleDataFrames can be converted to your chosen library data frame, and then methods of that library can be used. This section has an overview of how best to do conversions. Some examples are shown in subsequent sections.

To convert OracleDataFrame to a PyArrow Table, use pyarrow.Table.from_arrays() which leverages the Arrow PyCapsule interface.

To convert OracleDataFrame to a Pandas DataFrame, use pyarrow.Table.to_pandas().

If you want to use a data frame library other than Pandas or PyArrow, use the library’s from_arrow() method to convert a PyArrow Table to the applicable data frame, if your library supports this. For example, with Polars use polars.from_arrow().

Lastly, if your data frame library does not support from_arrow(), then use from_dataframe() if the library supports it. This can be slower, depending on the implementation.

The general recommendation is to use Apache Arrow as much as possible but if there are no options, then use from_dataframe().

Data Frame Type Mapping

Internally, python-oracledb’s OracleDataFrame support makes use of Apache nanoarrow libraries to build data frames.

The following data type mapping occurs from Oracle Database types to the Arrow types used in OracleDataFrame objects. Querying any other data types from Oracle Database will result in an exception.

Oracle Database Type	Arrow Data Type
DB_TYPE_NUMBER	DECIMAL128, INT64, or DOUBLE
DB_TYPE_CHAR	STRING
DB_TYPE_VARCHAR	STRING
DB_TYPE_BINARY_FLOAT	FLOAT
DB_TYPE_BINARY_DOUBLE	DOUBLE
DB_TYPE_BOOLEAN	BOOLEAN
DB_TYPE_DATE	TIMESTAMP
DB_TYPE_TIMESTAMP	TIMESTAMP
DB_TYPE_TIMESTAMP_LTZ	TIMESTAMP
DB_TYPE_TIMESTAMP_TZ	TIMESTAMP
DB_TYPE_CLOB	LARGE_STRING
DB_TYPE_BLOB	LARGE_BINARY
DB_TYPE_RAW	BINARY

When converting Oracle Database NUMBERs:

• If the column has been created without a precision and scale, then the Arrow data type will be DOUBLE.

• If defaults.fetch_decimals is set to True, then the Arrow data type is DECIMAL128.

• If the column has been created with a scale of 0, and a precision value that is less than or equal to 18, then the Arrow data type is INT64.

• In all other cases, the Arrow data type is DOUBLE.

When converting Oracle Database CLOBs and BLOBs:

• The LOBs must be no more than 1 GB in length.

When converting Oracle Database DATEs and TIMESTAMPs:

• For Oracle Database DATE types, the Arrow TIMESTAMP will have a time unit of “seconds”.

• For Oracle Database TIMESTAMP types, the Arrow TIMESTAMP time unit depends on the Oracle type’s fractional precision as shown in the table below:

  Oracle Database TIMESTAMP fractional second precision range	Arrow TIMESTAMP time unit
  0	seconds
  1 - 3	milliseconds
  4 - 6	microconds
  7 - 9	nanoseconds

Arrow TIMESTAMPs will not have timezone data.

Inserting OracleDataFrames into Oracle Database

To insert data currently in OracleDataFrame format into Oracle Database requires it to be converted. For example, you could convert it into a Pandas DataFrame for insert with the Pandas method to_sql(). Or convert into a Python list via the PyArrow Table.to_pylist() method and then use standard python-oracledb functionality to execute a SQL INSERT statement.

5.1.9.1. Creating PyArrow Tables

An example that creates and uses a PyArrow Table is:

# Get an OracleDataFrame
# Adjust arraysize to tune the query fetch performance
sql = "select id, name from SampleQueryTab order by id"
odf = connection.fetch_df_all(statement=sql, arraysize=100)

# Create a PyArrow table
pyarrow_table = pyarrow.Table.from_arrays(
    arrays=odf.column_arrays(), names=odf.column_names()
)

print("\nNumber of rows and columns:")
(r, c) = pyarrow_table.shape
print(f"{r} rows, {c} columns")

This makes use of OracleDataFrame.column_arrays() which returns a list of OracleArrowArray Objects.

Internally pyarrow.Table.from_arrays() leverages the Apache Arrow PyCapsule interface that OracleDataFrame exposes.

See samples/dataframe_pyarrow.py for a runnable example.

5.1.9.2. Creating Pandas DataFrames

An example that creates and uses a Pandas DataFrame is:

import pandas
import pyarrow

# Get an OracleDataFrame
# Adjust arraysize to tune the query fetch performance
sql = "select * from mytable where id = :1"
myid = 12345  # the bind variable value
odf = connection.fetch_df_all(statement=sql, parameters=[myid], arraysize=1000)

# Get a Pandas DataFrame from the data.
df = pyarrow.Table.from_arrays(
    odf.column_arrays(), names=odf.column_names()
).to_pandas()

# Perform various Pandas operations on the DataFrame
print(df.T)        # transform
print(df.tail(3))  # last three rows

The to_pandas() method supports arguments like types_mapper=pandas.ArrowDtype and deduplicate_objects=False, which may be useful for some data sets.

See samples/dataframe_pandas.py for a runnable example.

5.1.9.3. Creating Polars DataFrames

An example that creates and uses a Polars DataFrame is:

import pyarrow
import polars

# Get an OracleDataFrame
# Adjust arraysize to tune the query fetch performance
sql = "select id from SampleQueryTab order by id"
odf = connection.fetch_df_all(statement=sql, arraysize=100)

# Convert to a Polars DataFrame
pyarrow_table = pyarrow.Table.from_arrays(
    odf.column_arrays(), names=odf.column_names()
)
df = polars.from_arrow(pyarrow_table)

# Perform various Polars operations on the DataFrame
r, c = df.shape
print(f"{r} rows, {c} columns")
print(p.sum())

See samples/dataframe_polars.py for a runnable example.

5.1.9.4. Writing Apache Parquet Files

To write output in Apache Parquet file format, you can use data frames as an efficient intermediary. Use the Connection.fetch_df_batches() iterator and convert to a PyArrow Table that can be written by the PyArrow library.

import pyarrow
import pyarrow.parquet as pq

FILE_NAME = "sample.parquet"

# Tune the fetch batch size for your query
BATCH_SIZE = 10000

sql = "select * from mytable"
pqwriter = None
for odf in connection.fetch_df_batches(statement=sql, size=BATCH_SIZE):

    # Get a PyArrow table from the query results
    pyarrow_table = pyarrow.Table.from_arrays(
        arrays=odf.column_arrays(), names=odf.column_names()
    )

    if not pqwriter:
        pqwriter = pq.ParquetWriter(FILE_NAME, pyarrow_table.schema)

    pqwriter.write_table(pyarrow_table)

pqwriter.close()

See samples/dataframe_parquet_write.py for a runnable example.

5.1.9.5. The DLPack Protocol

The DataFrame format facilitates working with query results as tensors. Conversion can be done using the standard DLPack Protocol implemented by PyArrow.

Using NumPy Arrays

For example, to convert to NumPy ndarray format:

import pyarrow
import numpy

SQL = "select id from SampleQueryTab order by id"

# Get an OracleDataFrame
# Adjust arraysize to tune the query fetch performance
odf = connection.fetch_df_all(statement=SQL, arraysize=100)

# Convert to an ndarray via the Python DLPack specification
pyarrow_array = pyarrow.array(odf.get_column_by_name("ID"))
np = numpy.from_dlpack(pyarrow_array)

# Perform various numpy operations on the ndarray

print(numpy.sum(np))
print(numpy.log10(np))

See samples/dataframe_numpy.py for a runnable example.

Using Torch

An example of working with data as a Torch tensor is:

import pyarrow
import torch

SQL = "select id from SampleQueryTab order by id"

# Get an OracleDataFrame
# Adjust arraysize to tune the query fetch performance
odf = connection.fetch_df_all(statement=SQL, arraysize=100)

# Convert to a Torch tensor via the Python DLPack specification
pyarrow_array = pyarrow.array(odf.get_column_by_name("ID"))
tt = torch.from_dlpack(pyarrow_array)

# Perform various Torch operations on the tensor

print(torch.sum(tt))
print(torch.log10(tt))

See samples/dataframe_torch.py for a runnable example.

5.1.10. Limiting Rows

Query data is commonly broken into one or more sets:

• To give an upper bound on the number of rows that a query has to process, which can help improve database scalability.

• To perform ‘Web pagination’ that allows moving from one set of rows to a next, or previous, set on demand.

• For fetching of all data in consecutive small sets for batch processing. This happens because the number of records is too large for Python to handle at one time.

The latter can be handled by calling Cursor.fetchmany() with one execution of the SQL query.

‘Web pagination’ and limiting the maximum number of rows are detailed in this section. For each ‘page’ of results, a SQL query is executed to get the appropriate set of rows from a table. Since the query may be executed more than once, ensure to use bind variables for row numbers and row limits.

Oracle Database 12c SQL introduced an OFFSET / FETCH clause which is similar to the LIMIT keyword of MySQL. In Python, you can fetch a set of rows using:

myoffset = 0       # do not skip any rows (start at row 1)
mymaxnumrows = 20  # get 20 rows

sql =
  """SELECT last_name
     FROM employees
     ORDER BY last_name
     OFFSET :offset ROWS FETCH NEXT :maxnumrows ROWS ONLY"""

with connection.cursor() as cursor:

    cursor.prefetchrows = mymaxnumrows + 1
    cursor.arraysize = mymaxnumrows

    for row in cursor.execute(sql, offset=myoffset, maxnumrows=mymaxnumrows):
        print(row)

In applications where the SQL query is not known in advance, this method sometimes involves appending the OFFSET clause to the ‘real’ user query. Be very careful to avoid SQL injection security issues.

For Oracle Database 11g and earlier there are several alternative ways to limit the number of rows returned. The old, canonical paging query is:

SELECT *
FROM (SELECT a.*, ROWNUM AS rnum
      FROM (YOUR_QUERY_GOES_HERE -- including the order by) a
      WHERE ROWNUM <= MAX_ROW)
WHERE rnum >= MIN_ROW

Here, MIN_ROW is the row number of first row and MAX_ROW is the row number of the last row to return. For example:

SELECT *
FROM (SELECT a.*, ROWNUM AS rnum
      FROM (SELECT last_name FROM employees ORDER BY last_name) a
      WHERE ROWNUM <= 20)
WHERE rnum >= 1

This always has an ‘extra’ column, here called RNUM.

An alternative and preferred query syntax for Oracle Database 11g uses the analytic ROW_NUMBER() function. For example, to get the 1st to 20th names the query is:

SELECT last_name FROM
(SELECT last_name,
        ROW_NUMBER() OVER (ORDER BY last_name) AS myr
        FROM employees)
WHERE myr BETWEEN 1 and 20

Ensure to use bind variables for the upper and lower limit values.

5.1.11. Fetching Data in Parallel

The performance benefit of selecting table data in parallel from Oracle Database compared with a executing a single query depends on many factors. Partitioning the table and reading one partition per connection is usually the most efficient database-side solution. However, even if a parallel solution appears to be faster, it could be inefficient, thereby impacting, or eventually being limited by, everyone else. Only benchmarking in your environment will determine whether to use this technique.

A naive example using multiple threads is:

# A naive example for fetching data in parallel.
# Many factors affect whether this is beneficial

# The degree of parallelism / number of connections to open
NUM_THREADS = 10

# How many rows to fetch in each thread
BATCH_SIZE = 1000

# Internal buffer size: Tune for performance
oracledb.defaults.arraysize = 1000

# Note OFFSET/FETCH is not particularly efficient.
# It would be better to use a partitioned table
SQL = """
    select data
    from demo
    order by id
    offset :rowoffset rows fetch next :maxrows rows only
    """

def do_query(tn):
    with pool.acquire() as connection:
        with connection.cursor() as cursor:
            cursor.execute(SQL, rowoffset=(tn*BATCH_SIZE), maxrows=BATCH_SIZE)
            while True:
                rows = cursor.fetchmany()
                if not rows:
                    break
                print(f'Thread {tn}', rows)


pool = oracledb.create_pool(user="hr", password=userpwd, dsn="dbhost.example.com/orclpdb",
                            min=NUM_THREADS, max=NUM_THREADS)

thread = []
for i in range(NUM_THREADS):
    t = threading.Thread(target=do_query, args=(i,))
    t.start()
    thread.append(t)

for i in range(NUM_THREADS):
    thread[i].join()

When considering to parallelize queries from a table, some of the many factors include:

• Each connection to Oracle Database can only execute one statement at a time, so to parallelize queries requires using multiple connections.

• Python’s threading behavior and impact of the Python’s Global Interpreter Lock (GIL) may have an impact. You may need to spread work over multiple processes.

• What level of parallelism is most efficient?

• How many rows to fetch in each batch?

• What is your application doing with the data - can the receiving end efficiently process it, or write it to a disk?

• The OFFSET FETCH syntax will still cause database table blocks to be scanned even though not all data is returned to the application. Can the table be partitioned instead?

• There will be extra load on the database, both from the additional connections, and the work they are performing.

• Do your queries use up all of the database’s parallel servers?

• Is the data in the database spread across multiple disk spindles or is it the one disk which continually has to seek?

• Are Oracle Database zone maps being used?

• Is Oracle Exadata with storage indexes being used?

• Do you have function based indexes that are being invoked for every row?

5.1.12. Fetching Raw Data

Sometimes python-oracledb may have problems converting data stored in the database to Python strings. This can occur if the data stored in the database does not match the character set defined by the database. The encoding_errors parameter to Cursor.var() permits the data to be returned with some invalid data replaced, but for additional control the parameter bypass_decode can be set to True and python-oracledb will bypass the decode step and return bytes instead of str for data stored in the database as strings. The data can then be examined and corrected as required. This approach should only be used for troubleshooting and correcting invalid data, not for general use!

The following sample demonstrates how to use this feature:

# define output type handler
def return_strings_as_bytes(cursor, metadata):
    if metadata.type_code is oracledb.DB_TYPE_VARCHAR:
        return cursor.var(str, arraysize=cursor.arraysize,
                          bypass_decode=True)

# set output type handler on cursor before fetching data
with connection.cursor() as cursor:
    cursor.outputtypehandler = return_strings_as_bytes
    cursor.execute("select content, charset from SomeTable")
    data = cursor.fetchall()

This will produce output as:

[(b'Fianc\xc3\xa9', b'UTF-8')]

Note that last \xc3\xa9 is é in UTF-8. Since this is valid UTF-8 you can then perform a decode on the data (the part that was bypassed):

value = data[0][0].decode("UTF-8")

This will return the value “Fiancé”.

If you want to save b'Fianc\xc3\xa9' into the database directly without using a Python string, you will need to create a variable using Cursor.var() that specifies the type as oracledb.DB_TYPE_VARCHAR (otherwise the value will be treated as oracledb.DB_TYPE_RAW). The following sample demonstrates this:

with oracledb.connect(user="hr", password=userpwd,
                       dsn="dbhost.example.com/orclpdb") as conn:
    with conn.cursor() cursor:
        var = cursor.var(oracledb.DB_TYPE_VARCHAR)
        var.setvalue(0, b"Fianc\xc4\x9b")
        cursor.execute("""
            update SomeTable set
                SomeColumn = :param
            where id = 1""",
            param=var)

Warning

The database will assume that the bytes provided are in the character set expected by the database so only use this for troubleshooting or as directed.

5.1.13. Querying Corrupt Data

If queries fail with the error “codec can’t decode byte” when you select data, then:

• Check if your character set is correct. Review the database character sets. Check Fetching Raw Data. Note that the encoding used for all character data in python-oracledb is “UTF-8”.

• Check for corrupt data in the database.

If data really is corrupt, you can pass options to the internal decode() used by python-oracledb to allow it to be selected and prevent the whole query failing. Do this by creating an outputtypehandler and setting encoding_errors. For example, to replace corrupt characters in character columns:

def output_type_handler(cursor, metadata):
    if metadata.type_code is oracledb.DB_TYPE_VARCHAR:
        return cursor.var(metadata.type_code, size,
                          arraysize=cursor.arraysize,
                          encoding_errors="replace")

cursor.outputtypehandler = output_type_handler

cursor.execute("select column1, column2 from SomeTableWithBadData")

Other codec behaviors can be chosen for encoding_errors, see Error Handlers.

5.2. INSERT and UPDATE Statements

SQL Data Manipulation Language statements (DML) such as INSERT and UPDATE can easily be executed with python-oracledb. For example:

with connection.cursor() as cursor:
  cursor.execute("insert into MyTable values (:idbv, :nmbv)", [1, "Fredico"])

Do not concatenate or interpolate user data into SQL statements. See Using Bind Variables instead.

When handling multiple data values, use Cursor.executemany() for performance. See Executing Batch Statements and Bulk Loading

By default data is not committed to the database and other users will not be able to see your changes until your connection commits them by calling Connection.commit(). You can optionally rollback changes by calling Connection.rollback(). An implicit rollback will occur if your application finishes and does not explicitly commit any work.

To commit your changes, call:

connection.commit()

Note that the commit occurs on the connection, not the cursor.

If the attribute Connection.autocommit is True, then each statement executed is automatically committed without the need to call Connection.commit(). However overuse of the attribute causes extra database load, and can destroy transactional consistency.

See Managing Transactions for best practices on committing and rolling back data changes.

5.2.1. Inserting NULLs

Oracle Database requires a type, even for null values. When you pass the value None, then python-oracledb assumes its type is a string. If this is not the desired type, you can explicitly set it. For example, to insert a null Oracle Spatial SDO_GEOMETRY object:

type_obj = connection.gettype("SDO_GEOMETRY")
cursor = connection.cursor()
cursor.setinputsizes(type_obj)
cursor.execute("insert into sometable values (:1)", [None])