ExecutionOptions#

class ray.data.ExecutionOptions(resource_limits: ExecutionResources | None = None, exclude_resources: ExecutionResources | None = None, preserve_order: bool = False, actor_locality_enabled: bool = True, verbose_progress: bool | None = None, label_selector: Dict[str, str] | None = None)[source]#

Common options for execution.

Some options may not be supported on all executors (e.g., resource limits).

resource_limits#: Set a limit on the logical resources a Dataset can use. Autodetected by default.

exclude_resources#: Deprecated. Use label_selector to constrain Ray Data work to labeled nodes.

preserve_order#: Set this to preserve the ordering between blocks processed by operators. Off by default.

actor_locality_enabled#: Deprecated. Ray Data manages actor locality internally.

verbose_progress#: Whether to report progress individually per operator. By default, only AllToAll operators and global progress is reported. This option is useful for performance debugging. On by default.

label_selector#: A mapping of label key to label value. When set, every task and actor launched by this Dataset (including shuffle, sort, and aggregator actors) carries this label selector in its remote args, constraining placement to nodes whose labels satisfy the selector. Used to scope a Dataset to a labeled subset of the cluster (e.g. {"ray-subcluster": "training"}). Operator-level label_selector entries in ray_remote_args take precedence on key conflicts so existing node-pin selectors are preserved.

DeveloperAPI: This API may change across minor Ray releases.

is_resource_limits_default()[source]#: Returns True if resource_limits is the default value.