Skip to content

yggdrasil.enums.nodetype

nodetype

Centralized Databricks compute node-type identifiers.

Databricks accepts a free-form node_type_id string ("rd-fleet.xlarge", "m5.4xlarge", "Standard_D8ds_v5", "n2-standard-8", …). Every caller used to spell its own default, which made size / cloud changes a search-and-replace exercise and made it easy to typo a SKU.

:class:NodeType pins the convention to one place:

  • canonical members for the SKUs Yggdrasil's compute services prefer (Fleet sizes for AWS Databricks, common Azure/GCP families);
  • :attr:NodeType.DEFAULT — the workspace-cloud-agnostic Fleet default;
  • semantic aliases (:attr:SMALL / :attr:MEDIUM / :attr:LARGE / :attr:XLARGE) that map to Fleet sizes;
  • :meth:from_ and :meth:to_id for forgiving coercion at API boundaries — unknown SKUs pass through as plain strings rather than rejecting valid cloud-specific identifiers callers may use.

Members subclass :class:str, so a :class:NodeType member is interchangeable with the raw node_type_id string everywhere the Databricks SDK / REST API expects one.

Usage::

from yggdrasil.enums import NodeType

# Direct use — the member is a str.
spec = {"node_type_id": NodeType.DEFAULT}

# Forgiving coercion (returns the original string for unknown SKUs).
node_type_id = NodeType.to_id(user_provided_value)

# Membership / known-SKU check.
if NodeType.is_known(value):
    ...

NodeType

Bases: str, Enum

Canonical Databricks node_type_id values.

Member values are the exact strings the Databricks SDK expects. Use :attr:NodeType.DEFAULT for the codebase-wide default node type, the semantic-size aliases (:attr:SMALL / :attr:MEDIUM / :attr:LARGE / :attr:XLARGE) for intent-driven sizing, or the explicit cloud-specific members when you need a precise SKU.

The enum is intentionally not exhaustive — Databricks publishes hundreds of SKUs per cloud and most callers want a small, opinionated set. Coerce caller-provided strings through :meth:to_id / :meth:from_ so unknown SKUs round-trip as plain strings instead of raising.

from_ classmethod 

from_(value: Any, *, default: Any = ...) -> 'NodeType'

Coerce value into a :class:NodeType member.

Accepts:

  • :class:NodeType (returned as-is);
  • any string in :data:_NODETYPE_ALIASES, case-insensitive;
  • a bare member name ("FLEET_XLARGE");
  • None — returns default if supplied, else :attr:NodeType.DEFAULT.

Raises :class:ValueError for unknown strings unless default is supplied. To accept arbitrary SKU strings (without rejecting valid cloud-specific identifiers), call :meth:to_id instead.

to_id classmethod 

to_id(
    value: Union[str, "NodeType", None],
    *,
    default: Optional[Union[str, "NodeType"]] = None
) -> str

Coerce value to a Databricks node_type_id string.

Unlike :meth:from_, this is forgiving: arbitrary SKU strings round-trip unchanged so callers can pass cloud-specific identifiers the enum does not enumerate ("r5d.metal", custom marketplace images, …). Aliases are resolved when present.

Resolution order:

  1. NodeType member → its string value.
  2. Alias / member name / value match → the canonical string.
  3. Bare string → trimmed and returned as-is.
  4. None → default if supplied, else :attr:NodeType.DEFAULT.

is_known classmethod 

is_known(value: Any) -> bool

True when :meth:from_ would succeed for value.

NodeSpec dataclass 

NodeSpec(
    cpu_cores: int,
    ram_gib: float,
    gpu_count: int = 0,
    local_disk_gib: float = 0.0,
    cloud: str = "",
)

Hardware specs for a :class:NodeType member.

The numbers reflect the vendor's published per-VM characteristics for a single worker (or driver) — not the cluster-wide totals. ram_gib is binary GiB (1024 ** 3 bytes), matching every cloud's own documentation convention.

Attributes

cpu_cores Virtual CPU count exposed to Spark. ram_gib Memory available to Spark, in IEC GiB. gpu_count Number of GPUs (0 for non-GPU SKUs). local_disk_gib Local SSD/NVMe storage attached to the instance. 0 when the SKU has no local disk (relies on remote/elastic storage). cloud Cloud family identifier ("fleet" / "aws" / "azure" / "gcp") used by :meth:NodeType.from_cpu_and_ram when prefer is set.