backend#

class Backend(cfg: dict[str, Any] | None = None)#

Bases: ABC

type: str#

abstractmethod classmethod default_config() → dict[str, Any]#: Return a complete default configuration for this backend.

abstract property resource_specs: list[dict]#

abstract property valid_launchers: set[str]#

classmethod matches(arg: str) → bool#

property name: str#

abstractmethod submission_manager() → HPCSubmissionManager#

abstractmethod launcher() → HPCLauncher#

configure(cfg: dict[str, Any] | None = None) → dict[str, Any]#

describe() → str#

supports_subscheduling() → bool#

supports_dependencies() → bool#

validate() → None#

property resource_index: dict[str, list[tuple[dict, str | None]]]#

make_resource_index() → dict[str, list[tuple[dict, str | None]]]#: Map resource type -> list of (resource_spec, parent_type)

resource_types() → list[str]#: Return the types of resources available

count_per_node(rtype: str, default: int | None = None) → int#

count_per_socket(rtype: str, default: int | None = None) → int#

property node_count: int#

property sockets_per_node: int#

nodes_required(**rtypes: int) → int#: Nodes required to run tasks tasks. A task can be thought of as a single MPI rank

canonical_type_name(rtype: str) → str#

resource_view(*, ranks: int | None = None, ranks_per_socket: int | None = None) → dict[str, int]#

Return basic information about how to allocate resources on this machine for a job requiring ranks ranks.

Parameters#

ranksint: The number of ranks to use for a job
ranks_per_socketint: Number of ranks per socket, for performance use

Returns#

view:: view[‘np’] view[‘ranks’] view[‘nodes’] view[‘sockets’] view[‘ranks_per_socket’]

walk_resources(rspec: dict, *, parent_type: str | None = None) → Generator[tuple[dict, str | None], None, None]#