ray.data.ExecutionOptions#

class ray.data.ExecutionOptions(resource_limits: ray.data._internal.execution.interfaces.execution_options.ExecutionResources = <factory>, locality_with_output: Union[bool, List[str]] = False, preserve_order: bool = False, actor_locality_enabled: bool = True, verbose_progress: bool = False)[source]#

Bases: object

Common options for execution.

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

resource_limits#

Set a soft limit on the resource usage during execution. This is not supported in bulk execution mode. Autodetected by default.

Type

ray.data._internal.execution.interfaces.execution_options.ExecutionResources

locality_with_output#

Set this to prefer running tasks on the same node as the output node (node driving the execution). It can also be set to a list of node ids to spread the outputs across those nodes. Off by default.

Type

Union[bool, List[str]]

preserve_order#

Set this to preserve the ordering between blocks processed by operators under the streaming executor. The bulk executor always preserves order. Off by default.

Type

bool

actor_locality_enabled#

Whether to enable locality-aware task dispatch to actors (on by default). This applies to both ActorPoolStrategy map and streaming_split operations.

Type

bool

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. Off by default.

Type

bool

DeveloperAPI: This API may change across minor Ray releases.