stopPosition property
A worker may split an active map task in two parts, "primary" and "residual", continuing to process the primary part and returning the residual part into the pool of available work.
This event is called a "dynamic split" and is critical to the dynamic work rebalancing feature. The two obtained sub-tasks are called "parts" of the split. The parts, if concatenated, must represent the same input as would be read by the current task if the split did not happen. The exact way in which the original task is decomposed into the two parts is specified either as a position demarcating them (stop_position), or explicitly as two DerivedSources, if this task consumes a user-defined source type (dynamic_source_split). The "current" task is adjusted as a result of the split: after a task with range [A, B) sends a stop_position update at C, its range is considered to be [A, C), e.g.: * Progress should be interpreted relative to the new range, e.g. "75% completed" means "75% of [A, C) completed" * The worker should interpret proposed_stop_position relative to the new range, e.g. "split at 68%" should be interpreted as "split at 68% of [A, C)". * If the worker chooses to split again using stop_position, only stop_positions in [A, C) will be accepted. * Etc. dynamic_source_split has similar semantics: e.g., if a task with source S splits using dynamic_source_split into {P, R} (where P and R must be together equivalent to S), then subsequent progress and proposed_stop_position should be interpreted relative to P, and in a potential subsequent dynamic_source_split into {P', R'}, P' and R' must be together equivalent to P, etc.
Implementation
Position? stopPosition;