dragon.infrastructure.node_desc

A class that uniquely defines properties of a managed hardware node for infrastructure communication.

This is a stub at the moment, and should be extended for dynamic hardware allocations/elasticity aka The Cloud (TM) and more complex node properties (accelerators, networks, encryption).

Classes

NodeDescriptor

Globally available description of a 'managed node'.
class NodeDescriptor

Bases: object

Globally available description of a ‘managed node’.

class State

Bases: IntEnum

An enumeration.

NULL = 1
DISCOVERABLE = 2
PENDING = 3
ACTIVE = 4
IDLE = 5
ERROR = 6
DOWN = 7
__init__(state: State = State.NULL, h_uid: int | None = None, name: str = '', ip_addrs: list[str] | None = None, fabric_ep_addrs_available: bool = False, fabric_ep_addrs: list[str] | None = None, fabric_ep_addr_lens: list[int] | None = None, port: int | None = None, num_cpus: int = 0, physical_mem: int = 0, is_primary: bool = False, host_id: int | None = None, shep_cd: str = '', overlay_cd: str = '', host_name: str = '', cpu_devices: list[int] | None = None, accelerators: AcceleratorDescriptor | None = None)
classmethod get_localservices_node_conf(name: str | None = None, host_name: str | None = None, host_id: int | None = None, is_primary: bool = False, ip_addrs: list[str] | None = None, shep_cd: str | None = None, cpu_devices: list[int] | None = None, accelerators: AcceleratorDescriptor | None = None)

Return a NodeDescriptor object for Local Services to pass into its SHChannelsUp message

Populates the values in a NodeDescriptor object that Local Services needs to provide to the launcher frontend as part of infrastructure bring-up

Parameters:

  • name (Optional[str], optional) – Name for node. Often resorts to hostname, defaults to None

  • host_name (Optional[str], optional) – Hostname for the node, defaults to gethostname()

  • host_id (Optional[int], optional) – unique host ID of this node, defaults to get_host_id()

  • is_primary (bool, optional) – denote if this is the primary node running GS, defaults to False

  • ip_addrs (Optional[list[str]], optional) – IP addresses used for backend messaging by transport agents, defaults to [“127.0.0.1”]

  • shep_cd (Optional[str], optional) – Channel descriptor for this node’s Local Services, defaults to None

  • cpu_devices (Optional[list[int]], optional) – List of CPUs and IDs on this node, defaults to list(os.sched_getaffinity(0))

  • accelerators (Optional[AcceleratorDescriptor], optional) – List of any accelerators available on this node, defaults to find_accelerators()

classmethod get_local_node_network_conf(network_prefix: str = '^(hsn|ipogif|ib)\\d+$', port_range: Tuple[int, int] | int = (6565, 7565))

Return NodeDescriptor with IP for given network prefix, hostname, and host ID

Parameters:

  • network_prefix (str, optional) – network prefix used to find IP address of this node. Defaults to DEFAULT_TRANSPORT_NETIF

  • port_range (tuple[int, int] | int, optional) – Port and port range to use for communication. Defaults to (DEFAULT_OVERLAY_NETWORK_PORT, DEFAULT_OVERLAY_NETWORK_PORT+DEFAULT_PORT_RANGE) If just the port is passed as an int, the range will be assumed to be DEFAULT_PORT_RANGE

Returns:

Filled with local network info for node of execution

Return type:

NodeDescriptor

property sdesc
get_sdict()
classmethod from_sdict(sdict)