datafusion.context#

SessionContext — entry point for running DataFusion queries.

A SessionContext holds registered tables, catalogs, and configuration for the current session. It is the first object most programs create: from it you register data, run SQL strings (SessionContext.sql()), read files (SessionContext.read_csv(), SessionContext.read_parquet(), …), and construct DataFrame objects in memory (SessionContext.from_pydict(), SessionContext.from_arrow()).

Session behavior (memory limits, batch size, configured optimizer passes, …) is controlled by SessionConfig and RuntimeEnvBuilder; SQL dialect limits are controlled by SQLOptions.

Examples

>>> ctx = dfn.SessionContext()
>>> df = ctx.from_pydict({"a": [1, 2, 3]})
>>> ctx.sql("SELECT 1 AS n").to_pydict()
{'n': [1]}

See Concepts in the online documentation for the broader execution model.

Classes#

`ArrowArrayExportable`	Type hint for object exporting Arrow C Array via Arrow PyCapsule Interface.
`ArrowStreamExportable`	Type hint for object exporting Arrow C Stream via Arrow PyCapsule Interface.
`PhysicalOptimizerRuleExportable`	Type hint for object that has __datafusion_physical_optimizer_rule__ PyCapsule.
`RuntimeEnvBuilder`	Runtime configuration options.
`SQLOptions`	Options to be used when performing SQL queries.
`SessionConfig`	Session configuration options.
`SessionContext`	This is the main interface for executing queries and creating DataFrames.
`TableProviderExportable`	Type hint for object that has __datafusion_table_provider__ PyCapsule.

Module Contents#

class datafusion.context.ArrowArrayExportable#

Bases: Protocol

Type hint for object exporting Arrow C Array via Arrow PyCapsule Interface.

https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html

__arrow_c_array__(requested_schema: object | None = None) → tuple[object, object]#

class datafusion.context.ArrowStreamExportable#

Bases: Protocol

Type hint for object exporting Arrow C Stream via Arrow PyCapsule Interface.

https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html

__arrow_c_stream__(requested_schema: object | None = None) → object#

class datafusion.context.PhysicalOptimizerRuleExportable#

Bases: Protocol

Type hint for object that has __datafusion_physical_optimizer_rule__ PyCapsule.

The method returns a PyCapsule wrapping an FFI_PhysicalOptimizerRule, typically produced by a separate compiled extension.

__datafusion_physical_optimizer_rule__() → object#

class datafusion.context.RuntimeEnvBuilder#

Runtime configuration options.

Create a new RuntimeEnvBuilder with default values.

with_disk_manager_disabled() → RuntimeEnvBuilder#

Disable the disk manager, attempts to create temporary files will error.

Returns:: A new RuntimeEnvBuilder object with the updated setting.

with_disk_manager_os() → RuntimeEnvBuilder#

Use the operating system’s temporary directory for disk manager.

Returns:: A new RuntimeEnvBuilder object with the updated setting.

with_disk_manager_specified(*paths: str | pathlib.Path) → RuntimeEnvBuilder#

Use the specified paths for the disk manager’s temporary files.

Parameters:: paths – Paths to use for the disk manager’s temporary files.
Returns:: A new RuntimeEnvBuilder object with the updated setting.

with_fair_spill_pool(size: int) → RuntimeEnvBuilder#

Use a fair spill pool with the specified size.

This pool works best when you know beforehand the query has multiple spillable operators that will likely all need to spill. Sometimes it will cause spills even when there was sufficient memory (reserved for other operators) to avoid doing so:

┌───────────────────────z──────────────────────z───────────────┐
│                       z                      z               │
│                       z                      z               │
│       Spillable       z       Unspillable    z     Free      │
│        Memory         z        Memory        z    Memory     │
│                       z                      z               │
│                       z                      z               │
└───────────────────────z──────────────────────z───────────────┘

Parameters:: size – Size of the memory pool in bytes.
Returns:: A new RuntimeEnvBuilder object with the updated setting.

Examples

>>> config = dfn.RuntimeEnvBuilder().with_fair_spill_pool(1024)

with_greedy_memory_pool(size: int) → RuntimeEnvBuilder#

Use a greedy memory pool with the specified size.

This pool works well for queries that do not need to spill or have a single spillable operator. See with_fair_spill_pool() if there are multiple spillable operators that all will spill.

Parameters:: size – Size of the memory pool in bytes.
Returns:: A new RuntimeEnvBuilder object with the updated setting.

Examples

>>> config = dfn.RuntimeEnvBuilder().with_greedy_memory_pool(1024)

with_temp_file_path(path: str | pathlib.Path) → RuntimeEnvBuilder#

Use the specified path to create any needed temporary files.

Parameters:: path – Path to use for temporary files.
Returns:: A new RuntimeEnvBuilder object with the updated setting.

Examples

>>> config = dfn.RuntimeEnvBuilder().with_temp_file_path("/tmp")

with_unbounded_memory_pool() → RuntimeEnvBuilder#

Use an unbounded memory pool.

Returns:: A new RuntimeEnvBuilder object with the updated setting.

config_internal#

class datafusion.context.SQLOptions#

Options to be used when performing SQL queries.

Create a new SQLOptions with default values.

The default values are: - DDL commands are allowed - DML commands are allowed - Statements are allowed

with_allow_ddl(allow: bool = True) → SQLOptions#

Should DDL (Data Definition Language) commands be run?

Examples of DDL commands include CREATE TABLE and DROP TABLE.

Parameters:: allow – Allow DDL commands to be run.
Returns:: A new SQLOptions object with the updated setting.

Examples

>>> options = dfn.SQLOptions().with_allow_ddl(True)

with_allow_dml(allow: bool = True) → SQLOptions#

Should DML (Data Manipulation Language) commands be run?

Examples of DML commands include INSERT INTO and DELETE.

Parameters:: allow – Allow DML commands to be run.
Returns:: A new SQLOptions object with the updated setting.

Examples

>>> options = dfn.SQLOptions().with_allow_dml(True)

with_allow_statements(allow: bool = True) → SQLOptions#

Should statements such as SET VARIABLE and BEGIN TRANSACTION be run?

Parameters:: allow – Allow statements to be run.
Returns:: py:class:SQLOptions` object with the updated setting.
Return type:: A new

Examples

>>> options = dfn.SQLOptions().with_allow_statements(True)

options_internal#

class datafusion.context.SessionConfig(config_options: dict[str, str] | None = None)#

Session configuration options.

Create a new SessionConfig with the given configuration options.

Parameters:: config_options – Configuration options.

set(key: str, value: str) → SessionConfig#

Set a configuration option.

Args: key: Option key. value: Option value.

Returns:: A new SessionConfig object with the updated setting.

with_batch_size(batch_size: int) → SessionConfig#

Customize batch size.

Parameters:: batch_size – Batch size.
Returns:: A new SessionConfig object with the updated setting.

with_create_default_catalog_and_schema(enabled: bool = True) → SessionConfig#

Control if the default catalog and schema will be automatically created.

Parameters:: enabled – Whether the default catalog and schema will be automatically created.
Returns:: A new SessionConfig object with the updated setting.

with_default_catalog_and_schema(catalog: str, schema: str) → SessionConfig#

Select a name for the default catalog and schema.

Parameters:

catalog – Catalog name.
schema – Schema name.

Returns:

A new SessionConfig object with the updated setting.

with_extension(extension: Any) → SessionConfig#

Create a new configuration using an extension.

Parameters:

extension – A custom configuration extension object. These are
library. (shared from another DataFusion extension)

Returns:

A new SessionConfig object with the updated setting.

with_information_schema(enabled: bool = True) → SessionConfig#

Enable or disable the inclusion of information_schema virtual tables.

Parameters:: enabled – Whether to include information_schema virtual tables.
Returns:: A new SessionConfig object with the updated setting.

with_parquet_pruning(enabled: bool = True) → SessionConfig#

Enable or disable the use of pruning predicate for parquet readers.

Pruning predicates will enable the reader to skip row groups.

Parameters:: enabled – Whether to use pruning predicate for parquet readers.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_aggregations(enabled: bool = True) → SessionConfig#

Enable or disable the use of repartitioning for aggregations.

Enabling this improves parallelism.

Parameters:: enabled – Whether to use repartitioning for aggregations.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_file_min_size(size: int) → SessionConfig#

Set minimum file range size for repartitioning scans.

Parameters:: size – Minimum file range size.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_file_scans(enabled: bool = True) → SessionConfig#

Enable or disable the use of repartitioning for file scans.

Parameters:: enabled – Whether to use repartitioning for file scans.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_joins(enabled: bool = True) → SessionConfig#

Enable or disable the use of repartitioning for joins to improve parallelism.

Parameters:: enabled – Whether to use repartitioning for joins.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_sorts(enabled: bool = True) → SessionConfig#

Enable or disable the use of repartitioning for window functions.

This may improve parallelism.

Parameters:: enabled – Whether to use repartitioning for window functions.
Returns:: A new SessionConfig object with the updated setting.

with_repartition_windows(enabled: bool = True) → SessionConfig#

Enable or disable the use of repartitioning for window functions.

This may improve parallelism.

Parameters:: enabled – Whether to use repartitioning for window functions.
Returns:: A new SessionConfig object with the updated setting.

with_target_partitions(target_partitions: int) → SessionConfig#

Customize the number of target partitions for query execution.

Increasing partitions can increase concurrency.

Parameters:: target_partitions – Number of target partitions.
Returns:: A new SessionConfig object with the updated setting.

config_internal#

class datafusion.context.SessionContext(config: SessionConfig | None = None, runtime: RuntimeEnvBuilder | None = None)#

This is the main interface for executing queries and creating DataFrames.

See Concepts in the online documentation for more information.

Main interface for executing queries with DataFusion.

Maintains the state of the connection between a user and an instance of the connection between a user and an instance of the DataFusion engine.

Parameters:

config – Session configuration options.
runtime – Runtime configuration options.

Example usage:

The following example demonstrates how to use the context to execute a query against a CSV data source using the DataFrame API:

from datafusion import SessionContext

ctx = SessionContext()
df = ctx.read_csv("data.csv")

__datafusion_logical_extension_codec__() → Any#: Access the PyCapsule FFI_LogicalExtensionCodec.

__datafusion_physical_extension_codec__() → Any#: Access the PyCapsule FFI_PhysicalExtensionCodec.

__datafusion_task_context_provider__() → Any#: Access the PyCapsule FFI_TaskContextProvider.

__repr__() → str#: Print a string representation of the Session Context.

static _convert_file_sort_order(file_sort_order: collections.abc.Sequence[collections.abc.Sequence[datafusion.expr.SortKey]] | None) → list[list[datafusion._internal.expr.SortExpr]] | None#

Convert nested SortKey sequences into raw sort expressions.

Each SortKey can be a column name string, an Expr, or a SortExpr and will be converted using datafusion.expr.sort_list_to_raw_sort_list().

static _convert_table_partition_cols(table_partition_cols: list[tuple[str, str | pyarrow.DataType]]) → list[tuple[str, pyarrow.DataType]]#

add_physical_optimizer_rule(rule: PhysicalOptimizerRuleExportable) → None#

Append a user-defined physical optimizer rule to the session.

The rule is imported via its __datafusion_physical_optimizer_rule__ PyCapsule, typically produced by a separate compiled extension. The underlying SessionState is rebuilt from its current state with the new rule appended, so previously registered tables, UDFs, and catalogs are preserved.

Parameters:: rule – Object exposing __datafusion_physical_optimizer_rule__, a PhysicalOptimizerRuleExportable.

Examples

>>> from datafusion import SessionContext
>>> ctx = SessionContext()
>>> from my_extension import MyPhysicalOptimizerRule  
>>> rule = MyPhysicalOptimizerRule()  
>>> ctx.add_physical_optimizer_rule(rule)  

catalog(name: str = 'datafusion') → datafusion.catalog.Catalog#: Retrieve a catalog by name.

catalog_names() → set[str]#: Returns the list of catalogs in this context.

copied_config() → SessionConfig#

Return a copy of the active SessionConfig.

Mutating the returned config does not affect this context; use the result when you need a starting point for a new context or want to inspect the current settings independent of further changes here.

Examples

>>> ctx = SessionContext(SessionConfig().with_batch_size(1024))
>>> isinstance(ctx.copied_config(), SessionConfig)
True

create_dataframe(partitions: list[list[pyarrow.RecordBatch]], name: str | None = None, schema: pyarrow.Schema | None = None) → datafusion.dataframe.DataFrame#

Create and return a dataframe using the provided partitions.

Parameters:

partitions – pa.RecordBatch partitions to register.
name – Resultant dataframe name.
schema – Schema for the partitions.

Returns:

DataFrame representation of the SQL query.

create_dataframe_from_logical_plan(plan: datafusion.plan.LogicalPlan) → datafusion.dataframe.DataFrame#

Create a DataFrame from an existing plan.

Parameters:: plan – Logical plan.
Returns:: DataFrame representation of the logical plan.

deregister_object_store(schema: str, host: str | None = None) → None#

Remove an object store from the session.

Parameters:

schema – The data source schema (e.g. "s3://").
host – URL for the host (e.g. bucket name).

deregister_table(name: str) → None#: Remove a table from the session.

deregister_udaf(name: str) → None#

Remove a user-defined aggregate function from the session.

Parameters:: name – Name of the UDAF to deregister.

deregister_udf(name: str) → None#

Remove a user-defined scalar function from the session.

Parameters:: name – Name of the UDF to deregister.

deregister_udtf(name: str) → None#

Remove a user-defined table function from the session.

Parameters:: name – Name of the UDTF to deregister.

deregister_udwf(name: str) → None#

Remove a user-defined window function from the session.

Parameters:: name – Name of the UDWF to deregister.

empty_table() → datafusion.dataframe.DataFrame#: Create an empty DataFrame.

enable_ident_normalization() → bool#

Return whether identifier normalization (lowercasing) is enabled.

Examples

>>> ctx = SessionContext()
>>> ctx.enable_ident_normalization()
True

enable_spark_functions() → None#

Registers every UDF/UDAF/UDWF from the datafusion-spark crate, overriding any DataFusion built-ins of the same name with their Spark-semantics version (e.g. substring becomes 1-indexed, concat propagates NULL, round uses HALF_UP rounding).

For DataFrame use, import the typed wrappers from datafusion.functions.spark directly; this method is only needed for SQL queries.

Examples

>>> ctx = dfn.SessionContext()
>>> ctx.enable_spark_functions()
>>> ctx.sql(
...     "SELECT sha2('hello', 256) AS h"
... ).collect_column("h")[0].as_py()
'2cf24dba5fb0a30e26e83b2ac5b9e29e1b161e5c1fa7425e73043362938b9824'

enable_url_table() → SessionContext#

Control if local files can be queried as tables.

Returns:: A new SessionContext object with url table enabled.

execute(plan: datafusion.plan.ExecutionPlan, partitions: int) → datafusion.record_batch.RecordBatchStream#: Execute the plan and return the results.

execute_logical_plan(plan: datafusion.plan.LogicalPlan) → datafusion.dataframe.DataFrame#

Execute a LogicalPlan and return a DataFrame.

Parameters:: plan – Logical plan to execute.
Returns:: DataFrame resulting from the execution.

Examples

>>> ctx = SessionContext()
>>> df = ctx.from_pydict({"a": [1, 2, 3]})
>>> plan = df.logical_plan()
>>> df2 = ctx.execute_logical_plan(plan)
>>> df2.collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3
]

from_arrow(data: ArrowStreamExportable | ArrowArrayExportable, name: str | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame from an Arrow source.

The Arrow data source can be any object that implements either __arrow_c_stream__ or __arrow_c_array__. For the latter, it must return a struct array.

Arrow data can be Polars, Pandas, Pyarrow etc.

Parameters:

data – Arrow data source.
name – Name of the DataFrame.

Returns:

DataFrame representation of the Arrow table.

from_pandas(data: pandas.DataFrame, name: str | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame from a Pandas DataFrame.

Parameters:

data – Pandas DataFrame.
name – Name of the DataFrame.

Returns:

DataFrame representation of the Pandas DataFrame.

from_polars(data: polars.DataFrame, name: str | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame from a Polars DataFrame.

Parameters:

data – Polars DataFrame.
name – Name of the DataFrame.

Returns:

DataFrame representation of the Polars DataFrame.

from_pydict(data: dict[str, list[Any]], name: str | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame from a dictionary.

Parameters:

data – Dictionary of lists.
name – Name of the DataFrame.

Returns:

DataFrame representation of the dictionary of lists.

from_pylist(data: list[dict[str, Any]], name: str | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame from a list.

Parameters:

data – List of dictionaries.
name – Name of the DataFrame.

Returns:

DataFrame representation of the list of dictionaries.

classmethod global_ctx() → SessionContext#

Retrieve the global context as a SessionContext wrapper.

Returns:: A SessionContext object that wraps the global SessionContextInternal.

static parse_capacity_limit(config_name: str, limit: str) → int#

Parse a size string into a byte count.

Accepts strings like "100M", "1.5G", or "512K". "0" is accepted and returns 0. config_name is used purely for error messages and identifies which configuration setting the limit belongs to. Use this helper when constructing a RuntimeEnvBuilder from a human-friendly size string.

Examples

>>> SessionContext.parse_capacity_limit(
...     "datafusion.runtime.memory_limit", "1M"
... )
1048576
>>> SessionContext.parse_capacity_limit(
...     "datafusion.runtime.memory_limit", "0"
... )
0

parse_sql_expr(sql: str, schema: datafusion.common.DFSchema) → datafusion.expr.Expr#

Parse a SQL expression string into a logical expression.

Parameters:

sql – SQL expression string.
schema – Schema to use for resolving column references.

Returns:

Parsed expression.

Examples

>>> from datafusion.common import DFSchema
>>> ctx = SessionContext()
>>> schema = DFSchema.empty()
>>> ctx.parse_sql_expr("1 + 2", schema=schema)
Expr(Int64(1) + Int64(2))

read_arrow(path: str | pathlib.Path, schema: pyarrow.Schema | None = None, file_extension: str = '.arrow', file_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None) → datafusion.dataframe.DataFrame#

Create a DataFrame for reading an Arrow IPC data source.

Parameters:

path – Path to the Arrow IPC file.
schema – The data source schema.
file_extension – File extension to select.
file_partition_cols – Partition columns.

Returns:

DataFrame representation of the read Arrow IPC file.

Examples

>>> import tempfile, os
>>> ctx = dfn.SessionContext()
>>> table = pa.table({"a": [1, 2, 3]})
>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.arrow")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     df = ctx.read_arrow(path)
...     df.collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3
]

Provide an explicit schema to override schema inference:

>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.arrow")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     df = ctx.read_arrow(path, schema=pa.schema([("a", pa.int64())]))
...     df.collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3
]

Use file_extension to read files with a non-default extension:

>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.ipc")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     df = ctx.read_arrow(path, file_extension=".ipc")
...     df.collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3
]

read_avro(path: str | pathlib.Path, schema: pyarrow.Schema | None = None, file_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, file_extension: str = '.avro') → datafusion.dataframe.DataFrame#

Create a DataFrame for reading Avro data source.

Parameters:

path – Path to the Avro file.
schema – The data source schema.
file_partition_cols – Partition columns.
file_extension – File extension to select.

Returns:

DataFrame representation of the read Avro file

read_batch(batch: pyarrow.RecordBatch) → datafusion.dataframe.DataFrame#

Return a DataFrame reading a single batch.

Convenience wrapper around read_batches() for the single-batch case. Unlike register_batch(), this does not register the batch as a named table; it returns an anonymous DataFrame directly.

Parameters:: batch – Record batch to wrap as a DataFrame.

Examples

>>> ctx = dfn.SessionContext()
>>> batch = pa.RecordBatch.from_pydict({"a": [1, 2, 3]})
>>> ctx.read_batch(batch).to_pydict()
{'a': [1, 2, 3]}

read_batches(batches: collections.abc.Iterable[pyarrow.RecordBatch]) → datafusion.dataframe.DataFrame#

Return a DataFrame reading the given batches.

All batches must share the same schema. Any iterable of pa.RecordBatch is accepted (list, tuple, generator); it is materialized into a list before being handed to the underlying Rust binding. Unlike register_record_batches(), this does not register the batches as a named table; it returns an anonymous DataFrame directly.

Parameters:: batches – Record batches to wrap as a DataFrame.

Examples

>>> ctx = dfn.SessionContext()
>>> b1 = pa.RecordBatch.from_pydict({"a": [1, 2]})
>>> b2 = pa.RecordBatch.from_pydict({"a": [3, 4]})
>>> ctx.read_batches([b1, b2]).to_pydict()
{'a': [1, 2, 3, 4]}

A generator works too:

>>> ctx.read_batches(b for b in [b1, b2]).to_pydict()
{'a': [1, 2, 3, 4]}

read_csv(path: str | pathlib.Path | list[str] | list[pathlib.Path], schema: pyarrow.Schema | None = None, has_header: bool = True, delimiter: str = ',', schema_infer_max_records: int = DEFAULT_MAX_INFER_SCHEMA, file_extension: str = '.csv', table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, file_compression_type: str | None = None, options: datafusion.options.CsvReadOptions | None = None) → datafusion.dataframe.DataFrame#

Read a CSV data source.

Parameters:

path – Path to the CSV file
schema – An optional schema representing the CSV files. If None, the CSV reader will try to infer it based on data in file.
has_header – Whether the CSV file have a header. If schema inference is run on a file with no headers, default column names are created.
delimiter – An optional column delimiter.
schema_infer_max_records – Maximum number of rows to read from CSV files for schema inference if needed.
file_extension – File extension; only files with this extension are selected for data input.
table_partition_cols – Partition columns.
file_compression_type – File compression type.
options – Set advanced options for CSV reading. This cannot be combined with any of the other options in this method.

Returns:

DataFrame representation of the read CSV files

read_empty() → datafusion.dataframe.DataFrame#: Create an empty DataFrame with no columns or rows.

See also

This is an alias for empty_table().

read_json(path: str | pathlib.Path, schema: pyarrow.Schema | None = None, schema_infer_max_records: int = 1000, file_extension: str = '.json', table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, file_compression_type: str | None = None) → datafusion.dataframe.DataFrame#

Read a line-delimited JSON data source.

Parameters:

path – Path to the JSON file.
schema – The data source schema.
schema_infer_max_records – Maximum number of rows to read from JSON files for schema inference if needed.
file_extension – File extension; only files with this extension are selected for data input.
table_partition_cols – Partition columns.
file_compression_type – File compression type.

Returns:

DataFrame representation of the read JSON files.

read_parquet(path: str | pathlib.Path, table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, parquet_pruning: bool = True, file_extension: str = '.parquet', skip_metadata: bool = True, schema: pyarrow.Schema | None = None, file_sort_order: collections.abc.Sequence[collections.abc.Sequence[datafusion.expr.SortKey]] | None = None) → datafusion.dataframe.DataFrame#

Read a Parquet source into a Dataframe.

Parameters:

path – Path to the Parquet file.
table_partition_cols – Partition columns.
parquet_pruning – Whether the parquet reader should use the predicate to prune row groups.
file_extension – File extension; only files with this extension are selected for data input.
skip_metadata – Whether the parquet reader should skip any metadata that may be in the file schema. This can help avoid schema conflicts due to metadata.
schema – An optional schema representing the parquet files. If None, the parquet reader will try to infer it based on data in the file.
file_sort_order – Sort order for the file. Each sort key can be specified as a column name (str), an expression (Expr), or a SortExpr.

Returns:

DataFrame representation of the read Parquet files

read_table(table: datafusion.catalog.Table | TableProviderExportable | datafusion.dataframe.DataFrame | pyarrow.dataset.Dataset) → datafusion.dataframe.DataFrame#: Creates a DataFrame from a table.

refresh_catalogs() → None#

Refresh catalog metadata.

Examples

>>> ctx = SessionContext()
>>> ctx.refresh_catalogs()

register_arrow(name: str, path: str | pathlib.Path, schema: pyarrow.Schema | None = None, file_extension: str = '.arrow', table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None) → None#

The registered table can be referenced from SQL statements executed against this context.

Parameters:

name – Name of the table to register.
path – Path to the Arrow IPC file.
schema – The data source schema.
file_extension – File extension to select.
table_partition_cols – Partition columns.

Examples

>>> import tempfile, os
>>> ctx = dfn.SessionContext()
>>> table = pa.table({"x": [10, 20, 30]})
>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.arrow")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     ctx.register_arrow("arrow_tbl", path)
...     ctx.sql("SELECT * FROM arrow_tbl").collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  10,
  20,
  30
]

Provide an explicit schema to override schema inference:

>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.arrow")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     ctx.register_arrow(
...         "arrow_schema",
...         path,
...         schema=pa.schema([("x", pa.int64())]),
...     )
...     ctx.sql("SELECT * FROM arrow_schema").collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  10,
  20,
  30
]

Use file_extension to read files with a non-default extension:

>>> with tempfile.TemporaryDirectory() as tmpdir:
...     path = os.path.join(tmpdir, "data.ipc")
...     with pa.ipc.new_file(path, table.schema) as writer:
...         writer.write_table(table)
...     ctx.register_arrow(
...         "arrow_ipc", path, file_extension=".ipc"
...     )
...     ctx.sql("SELECT * FROM arrow_ipc").collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  10,
  20,
  30
]

register_avro(name: str, path: str | pathlib.Path, schema: pyarrow.Schema | None = None, file_extension: str = '.avro', table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None) → None#

The registered table can be referenced from SQL statement executed against this context.

Parameters:

name – Name of the table to register.
path – Path to the Avro file.
schema – The data source schema.
file_extension – File extension to select.
table_partition_cols – Partition columns.

register_batch(name: str, batch: pyarrow.RecordBatch) → None#

Parameters:

name – Name of the resultant table.
batch – Record batch to register as a table.

Examples

>>> ctx = dfn.SessionContext()
>>> batch = pa.RecordBatch.from_pydict({"a": [1, 2, 3]})
>>> ctx.register_batch("batch_tbl", batch)
>>> ctx.sql("SELECT * FROM batch_tbl").collect()[0].column(0)
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3
]

register_catalog_provider(name: str, provider: datafusion.catalog.CatalogProviderExportable | datafusion.catalog.CatalogProvider | datafusion.catalog.Catalog) → None#: Register a catalog provider.

register_catalog_provider_list(provider: datafusion.catalog.CatalogProviderListExportable | datafusion.catalog.CatalogProviderList | datafusion.catalog.CatalogList) → None#: Register a catalog provider list.

register_csv(name: str, path: str | pathlib.Path | list[str | pathlib.Path], schema: pyarrow.Schema | None = None, has_header: bool = True, delimiter: str = ',', schema_infer_max_records: int = DEFAULT_MAX_INFER_SCHEMA, file_extension: str = '.csv', file_compression_type: str | None = None, options: datafusion.options.CsvReadOptions | None = None) → None#

The registered table can be referenced from SQL statement executed against.

Parameters:

name – Name of the table to register.
path – Path to the CSV file. It also accepts a list of Paths.
schema – An optional schema representing the CSV file. If None, the CSV reader will try to infer it based on data in file.
has_header – Whether the CSV file have a header. If schema inference is run on a file with no headers, default column names are created.
delimiter – An optional column delimiter.
schema_infer_max_records – Maximum number of rows to read from CSV files for schema inference if needed.
file_extension – File extension; only files with this extension are selected for data input.
file_compression_type – File compression type.
options – Set advanced options for CSV reading. This cannot be combined with any of the other options in this method.

register_dataset(name: str, dataset: pyarrow.dataset.Dataset) → None#

Parameters:

name – Name of the table to register.
dataset – PyArrow dataset.

register_json(name: str, path: str | pathlib.Path, schema: pyarrow.Schema | None = None, schema_infer_max_records: int = 1000, file_extension: str = '.json', table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, file_compression_type: str | None = None) → None#

The registered table can be referenced from SQL statement executed against this context.

Parameters:

name – Name of the table to register.
path – Path to the JSON file.
schema – The data source schema.
schema_infer_max_records – Maximum number of rows to read from JSON files for schema inference if needed.
file_extension – File extension; only files with this extension are selected for data input.
table_partition_cols – Partition columns.
file_compression_type – File compression type.

register_listing_table(name: str, path: str | pathlib.Path, table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, file_extension: str = '.parquet', schema: pyarrow.Schema | None = None, file_sort_order: collections.abc.Sequence[collections.abc.Sequence[datafusion.expr.SortKey]] | None = None) → None#

Registers a Table that can assemble multiple files from locations in an ObjectStore instance.

Parameters:

name – Name of the resultant table.
path – Path to the file to register.
table_partition_cols – Partition columns.
file_extension – File extension of the provided table.
schema – The data source schema.
file_sort_order – Sort order for the file. Each sort key can be specified as a column name (str), an expression (Expr), or a SortExpr.

register_object_store(schema: str, store: Any, host: str | None = None) → None#

Add a new object store into the session.

Parameters:

schema – The data source schema.
store – The ObjectStore to register.
host – URL for the host.

register_parquet(name: str, path: str | pathlib.Path, table_partition_cols: list[tuple[str, str | pyarrow.DataType]] | None = None, parquet_pruning: bool = True, file_extension: str = '.parquet', skip_metadata: bool = True, schema: pyarrow.Schema | None = None, file_sort_order: collections.abc.Sequence[collections.abc.Sequence[datafusion.expr.SortKey]] | None = None) → None#

The registered table can be referenced from SQL statement executed against this context.

Parameters:

name – Name of the table to register.
path – Path to the Parquet file.
table_partition_cols – Partition columns.
parquet_pruning – Whether the parquet reader should use the predicate to prune row groups.
file_extension – File extension; only files with this extension are selected for data input.
skip_metadata – Whether the parquet reader should skip any metadata that may be in the file schema. This can help avoid schema conflicts due to metadata.
schema – The data source schema.
file_sort_order – Sort order for the file. Each sort key can be specified as a column name (str), an expression (Expr), or a SortExpr.

register_record_batches(name: str, partitions: list[list[pyarrow.RecordBatch]]) → None#

This function will convert the provided partitions into a table and register it into the session using the given name.

Parameters:

name – Name of the resultant table.
partitions – Record batches to register as a table.

register_table(name: str, table: datafusion.catalog.Table | TableProviderExportable | datafusion.dataframe.DataFrame | pyarrow.dataset.Dataset) → None#

The registered table can be referenced from SQL statements executed against this context.

Parameters:

name – Name of the resultant table.
table – Any object that can be converted into a Table.

register_table_factory(format: str, factory: datafusion.catalog.TableProviderFactory | datafusion.catalog.TableProviderFactoryExportable) → None#

The registered factory can be referenced from SQL DDL statements executed against this context.

Parameters:

format – The value to be used in STORED AS ${format} clause.
factory – A PyCapsule that implements TableProviderFactoryExportable

register_table_provider(name: str, provider: datafusion.catalog.Table | TableProviderExportable | datafusion.dataframe.DataFrame | pyarrow.dataset.Dataset) → None#

Deprecated: use register_table() instead.

register_udaf(udaf: datafusion.user_defined.AggregateUDF) → None#: Register a user-defined aggregation function (UDAF) with the context.

register_udf(udf: datafusion.user_defined.ScalarUDF) → None#: Register a user-defined function (UDF) with the context.

register_udtf(func: datafusion.user_defined.TableFunction) → None#: Register a user defined table function.

register_udwf(udwf: datafusion.user_defined.WindowUDF) → None#: Register a user-defined window function (UDWF) with the context.

register_view(name: str, df: datafusion.dataframe.DataFrame) → None#

Parameters:

name (str) – The name to register the view under.
df (DataFrame) – The DataFrame to be converted into a view and registered.

remove_optimizer_rule(name: str) → bool#

Remove an optimizer rule by name.

Parameters:: name – Name of the optimizer rule to remove.
Returns:: True if a rule with the given name was found and removed.

Examples

>>> ctx = SessionContext()
>>> ctx.remove_optimizer_rule("nonexistent_rule")
False

session_id() → str#: Return an id that uniquely identifies this SessionContext.

session_start_time() → str#

Return the session start time as an RFC 3339 formatted string.

Examples

>>> ctx = SessionContext()
>>> ctx.session_start_time()  
'2026-01-01T12:34:56.123456789+00:00'

sql(query: str, options: SQLOptions | None = None, param_values: dict[str, Any] | None = None, **named_params: Any) → datafusion.dataframe.DataFrame#

Create a DataFrame from SQL query text.

See the online documentation for a description of how to perform parameterized substitution via either the param_values option or passing in named_params.

Note: This API implements DDL statements such as CREATE TABLE and CREATE VIEW and DML statements such as INSERT INTO with in-memory default implementation.See sql_with_options().

Parameters:

query – SQL query text.
options – If provided, the query will be validated against these options.
param_values – Provides substitution of scalar values in the query after parsing.
named_params – Provides string or DataFrame substitution in the query string.

Returns:

DataFrame representation of the SQL query.

sql_with_options(query: str, options: SQLOptions, param_values: dict[str, Any] | None = None, **named_params: Any) → datafusion.dataframe.DataFrame#

Create a DataFrame from SQL query text.

This function will first validate that the query is allowed by the provided options.

Parameters:

query – SQL query text.
options – SQL options.
param_values – Provides substitution of scalar values in the query after parsing.
named_params – Provides string or DataFrame substitution in the query string.

Returns:

DataFrame representation of the SQL query.

table(name: str) → datafusion.dataframe.DataFrame#: Retrieve a previously registered table by name.

table_exist(name: str) → bool#: Return whether a table with the given name exists.

table_provider(name: str) → datafusion.catalog.Table#

Return the Table for the given table name.

Parameters:: name – Name of the table.
Returns:: The table provider.
Raises:: KeyError – If the table is not found.

Examples

>>> import pyarrow as pa
>>> ctx = SessionContext()
>>> batch = pa.RecordBatch.from_pydict({"x": [1, 2]})
>>> ctx.register_record_batches("my_table", [[batch]])
>>> tbl = ctx.table_provider("my_table")
>>> tbl.schema
x: int64

udaf(name: str) → datafusion.user_defined.AggregateUDF#

Look up a registered aggregate UDF by name.

Returns the same AggregateUDF wrapper that register_udaf() accepts. Built-in aggregate functions such as sum or avg are also discoverable through this lookup. See udf() for a worked late-binding example; the pattern is identical for aggregates.

Parameters:: name – Name of the registered aggregate UDF.
Raises:: KeyError – If no aggregate UDF is registered under name.

Examples

Look up a built-in aggregate by name and use it in aggregate():

>>> ctx = dfn.SessionContext()
>>> sum_fn = ctx.udaf("sum")
>>> df = ctx.from_pydict({"a": [1, 2, 3]})
>>> df.aggregate([], [sum_fn(col("a")).alias("total")]).to_pydict()
{'total': [6]}

udafs() → list[str]#

Return the sorted names of all registered aggregate UDFs.

Examples

>>> ctx = dfn.SessionContext()
>>> "sum" in ctx.udafs()
True

udf(name: str) → datafusion.user_defined.ScalarUDF#

Look up a registered scalar UDF by name.

Returns the same ScalarUDF wrapper that register_udf() accepts, so it can be invoked as an expression in the DataFrame API or re-registered into a different SessionContext. Built-in scalar functions from the session’s function registry are also looked up.

Parameters:: name – Name of the registered scalar UDF.
Raises:: KeyError – If no scalar UDF is registered under name.

Examples

>>> ctx = dfn.SessionContext()
>>> nullcheck = dfn.udf(
...     lambda x: x.is_null(),
...     [pa.int64()],
...     pa.bool_(),
...     volatility="immutable",
...     name="nullcheck",
... )
>>> ctx.register_udf(nullcheck)
>>> fn = ctx.udf("nullcheck")
>>> df = ctx.from_pydict({"a": [1, None, 3]})
>>> df.select(fn(col("a")).alias("is_null")).to_pydict()
{'is_null': [False, True, False]}

Late-binding: the function name can come from configuration rather than an imported symbol, which is useful when the set of UDFs is plugin-driven or chosen at runtime:

>>> config = {"null_check": "nullcheck"}
>>> fn = ctx.udf(config["null_check"])
>>> df.select(fn(col("a")).alias("is_null")).to_pydict()
{'is_null': [False, True, False]}

udfs() → list[str]#

Return the sorted names of all registered scalar UDFs.

Includes both user-registered and built-in scalar functions. Pair with udf() to drive discovery, validation, or config-based dispatch.

Examples

>>> ctx = dfn.SessionContext()
>>> "abs" in ctx.udfs()
True

udwf(name: str) → datafusion.user_defined.WindowUDF#

Look up a registered window UDF by name.

Returns the same WindowUDF wrapper that register_udwf() accepts. Built-in window functions such as row_number or rank are also discoverable through this lookup. See udf() for a worked late-binding example; the pattern is identical for window functions.

Parameters:: name – Name of the registered window UDF.
Raises:: KeyError – If no window UDF is registered under name.

Examples

Look up a built-in window function by name and use it in select:

>>> ctx = dfn.SessionContext()
>>> rn = ctx.udwf("row_number")
>>> df = ctx.from_pydict({"a": [10, 20, 30]})
>>> df.select(col("a"), rn().alias("rn")).to_pydict()
{'a': [10, 20, 30], 'rn': [1, 2, 3]}

udwfs() → list[str]#

Return the sorted names of all registered window UDFs.

Examples

>>> ctx = dfn.SessionContext()
>>> "row_number" in ctx.udwfs()
True

with_logical_extension_codec(codec: datafusion.user_defined.LogicalExtensionCodecExportable | _typeshed.CapsuleType) → SessionContext#

Create a new session context with specified codec.

Only FFI codecs are supported. Pass any object implementing __datafusion_logical_extension_codec__ (see LogicalExtensionCodecExportable).

with_physical_extension_codec(codec: datafusion.user_defined.PhysicalExtensionCodecExportable | _typeshed.CapsuleType) → SessionContext#

Create a new session context with the specified physical codec.

Only FFI codecs are supported. Pass any object implementing __datafusion_physical_extension_codec__ (see PhysicalExtensionCodecExportable).

with_python_udf_inlining(*, enabled: bool) → SessionContext#

Control whether Python UDFs are embedded in serialized expressions.

enabled is keyword-only and required: callers must pick a mode explicitly. Fresh sessions inline UDFs (enabled=True behavior) until this method overrides the toggle.

With enabled=True, serialized expressions carry the Python code for any scalar, aggregate, or window UDFs they reference. The receiver rebuilds the UDFs from those bytes and does not need to register them first.

With enabled=False, serialized expressions store only the UDF names. This has two uses:

Cross-language portability. The bytes can be decoded by a non-Python receiver, which must already have UDFs registered under matching names.
Safer deserialization. Expr.from_bytes() will refuse to rebuild Python UDFs rather than call cloudpickle.loads on untrusted input.

The setting affects Expr.to_bytes() and Expr.from_bytes() whenever this session is passed as the ctx argument. pickle.dumps() and pickle.loads() do not pass a context, so to apply the setting through pickle, register this session with datafusion.ipc.set_sender_ctx() on the sender and datafusion.ipc.set_worker_ctx() on the receiver.

Warning

Security This setting narrows only Expr.from_bytes(). Calling pickle.loads() on untrusted bytes remains unsafe regardless of the toggle.

Returns a new SessionContext with the toggle applied; the original session is unchanged.

Examples

>>> import pyarrow as pa
>>> from datafusion import SessionContext, Expr, col, udf
>>> ctx = SessionContext()
>>> identity = udf(lambda a: a, [pa.int64()], pa.int64(),
...                volatility="immutable", name="identity_demo")
>>> ctx.register_udf(identity)
>>> blob = identity(col("x")).to_bytes(ctx)
>>> strict = SessionContext().with_python_udf_inlining(enabled=False)
>>> try:
...     Expr.from_bytes(blob, strict)
... except Exception as e:
...     print("Refusing to deserialize" in str(e))
True

ctx#

class datafusion.context.TableProviderExportable#

Bases: Protocol

Type hint for object that has __datafusion_table_provider__ PyCapsule.

https://datafusion.apache.org/python/user-guide/io/table_provider.html

__datafusion_table_provider__(session: Any) → object#