shm

SHM-backed NumPy and pandas helpers.

These types are thin wrappers around NumPy/pandas objects that keep a reference to the shared-memory block that backs the underlying data. They enable brmspy's codecs to avoid extra copies when transporting large numeric payloads between the main process and the worker.

See Also

brmspy._session.codec.builtin.NumpyArrayCodec Encodes/decodes NumPy arrays into shared memory. brmspy._session.codec.builtin.PandasDFCodec Encodes/decodes DataFrames into shared memory. brmspy.types.shm Base shared-memory block and pool types.

Classes

ShmArray

Bases: ndarray

NumPy array view backed by a shared-memory block.

Attributes:

Name	Type	Description
block	ShmRef	Reference to the shared-memory block backing the array data.

Notes

This is a view over SharedMemory.buf. Closing/unlinking the underlying shared memory while the array is still in use will lead to undefined behavior.

Functions

from_block(block, shape, dtype, **kwargs) classmethod

Create an array view backed by an existing shared-memory block.

Parameters:

Name	Type	Description	Default
block	ShmBlock	Attached shared-memory block.	required
shape	tuple[int, ...]	Desired array shape.	required
dtype	dtype	NumPy dtype of the array.	required
**kwargs		Reserved for future compatibility. Currently unused.	{}

Returns:

Type	Description
ShmArray	Array view into the shared-memory buffer.

array_order(a) classmethod

Determine how an array can be reconstructed from a raw buffer.

Returns "C" for C-contiguous arrays, "F" for Fortran-contiguous arrays, otherwise "non-contiguous" (meaning: bytes were obtained by forcing a contiguous copy during encoding).

ShmDataFrameSimple

Bases: DataFrame

pandas DataFrame backed by a single shared-memory block (numeric only).

Attributes:

Name	Type	Description
block	ShmRef	Reference to the shared-memory block backing the DataFrame's values.

Functions

from_block(block, nrows, ncols, columns, index, dtype) classmethod

Construct a DataFrame backed by a single SHM block.

Parameters:

Name	Type	Description	Default
block	ShmBlock	Attached shared-memory block containing a contiguous 2D numeric matrix.	required
nrows	int	DataFrame shape.	required
ncols	int	DataFrame shape.	required
columns	list[Any] or None	Column/index labels.	required
index	list[Any] or None	Column/index labels.	required
dtype	str or dtype	Dtype of the matrix stored in the block.	required

Returns:

Type	Description
ShmDataFrameSimple

ShmDataFrameColumns

Bases: DataFrame

pandas DataFrame backed by per-column shared-memory blocks (numeric only).

Attributes:

Name	Type	Description
_blocks_columns	dict[str, PandasColumnMetadata]	Mapping from column name to data required for its reconstruction

Classes

ShmRef

Bases: TypedDict

Reference to a shared-memory block sent over IPC.

Attributes:

Name	Type	Description
name	str	Shared memory block name (as assigned by SharedMemoryManager).
size	int	Allocated block size in bytes.
content_size	int	Actual used size
temporary	bool	Whether this buffer can be GC-d immediately after use or should it be attached to object its constructed into.

Notes

Codecs may store a logical payload smaller than size. In that case, the codec metadata must include the logical nbytes/length so that decoders can slice the buffer appropriately.

ShmBlock dataclass

Attached shared-memory block (name/size + live SharedMemory handle).

Notes

This object owns a SharedMemory handle and must be closed when no longer needed. In brmspy this is managed by a ShmPool implementation.

ShmPool

Minimal interface for allocating and attaching shared-memory blocks.

The concrete implementation lives in brmspy._session.transport.ShmPool and tracks blocks so they can be closed on teardown.

Functions

__init__(manager)

Create a pool bound to an existing SharedMemoryManager.

Parameters:

Name	Type	Description	Default
manager	SharedMemoryManager	Manager used to allocate blocks.	required

alloc(size, temporary=False)

Allocate a new shared-memory block.

Parameters:

Name	Type	Description	Default
size	int	Size in bytes.	required

Returns:

Type	Description
ShmBlock	Newly allocated block.

attach(ref)

Attach to an existing shared-memory block by name.

Returns:

Type	Description
ShmBlock	Attached block.

close_all()

Close all tracked shared-memory handles owned by this pool.

Returns:

Type	Description
None