Class BlockExecStats

python/ray/data/block.py:207–234

Execution stats for a single output block produced by a task.

@DeveloperAPI
@dataclass(frozen=True)
class BlockExecStats:
    """Execution stats for a single output block produced by a task."""

    # Index of the task that produced this block, used to attribute rows
    # to individual tasks in per-task statistics.
    task_idx: Optional[int] = None

    # Ray node ID of the worker that produced this block.
    node_id: str = field(
        default_factory=lambda: ray.runtime_context.get_runtime_context().get_node_id()
    )

    # Absolute wall-clock timestamp when block generation started.
    start_time_s: Optional[float] = None
    # Absolute wall-clock timestamp when block generation finished.
    end_time_s: Optional[float] = None
    # Total wall-clock duration of the block generation (computed as end_time_s - start_time_s).
    wall_time_s: Optional[float] = None
    # Time spent inside UDF while generating block.
    udf_time_s: Optional[float] = 0
    # Time spent serializing this block into a Ray object.
    block_ser_time_s: Optional[float] = None
    # Total CPU time consumed by the worker process during the task, across all threads.
    cpu_time_s: Optional[float] = None

    @staticmethod
    def builder() -> "_BlockExecStatsBuilder":
        return _BlockExecStatsBuilder()


class _BlockExecStatsBuilder: