Skip to content

Conversation

@kfirtoledo
Copy link
Collaborator

This PR adds the complete Python backend implementation for the filesystem offloading connector.
Included in this PR:
Adds the Python package modules:

  • factory.py - registers the fs connector spec (SharedStorageOffloadingSpec)
  • manager.py - implements the OffloadingManager logic for the filesystem backend
  • mediums.py - defines the storage medium used by the connector
  • spec.py - defines the filesystem storage backend class used by vLLM
  • worker.py - executes asynchronous file I/O tasks for reading and writing blocks

Adds project configuration:

  • pyproject.toml
  • README.md
  • docs/

Adds initial test coverage:

  • tests/test_fs_backend.py

Key capabilities:
Provides the Python interface for the filesystem connector used by the offloading subsystem.
Implements connector registration, scheduling logic, and storage medium abstraction.
Connects cleanly to the underlying C++ and CUDA extension modules.
Defines SharedStorageOffloadingSpec for integration with vLLM's KV connector framework.
This module completes the high-level Python layer required for end-to-end operation of the filesystem offloading connector.

More details can be found in the README file.

…d of offloading connector (fs_connector)

Signed-off-by: Kfir Toledo <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant