execution stats may mismatch real changes caused by custom stage listener code

[vitalz]

Execution stats may mismatch real changes caused by custom stage listener code.

For example, there is dry delete stage listener code:

public class DryDeleteStageListener<T extends BaseDataObject> {
    private static final Logger log = LoggerFactory.getLogger(DryDeleteStageListener.class);
    private final Consumer<T> consumer;

    public DryDeleteStageListener(Consumer<T> consumer) {
        this.consumer = consumer;
    }

    @AfterMissingTargetsFiltered
    public void afterMissingTargetsFiltered(Execution exec, DeleteSegment<T> segment) {
        segment.getMissingTargets().forEach(rowProxy -> consumer.accept((T) rowProxy.get(DeleteSegment.TARGET_COLUMN)));
        log.debug("Setting an empty list for the missing targets.");
        segment.setMissingTargets(DataFrameBuilder.builder().empty());
    }

}

And a LinkMove task like:

        LmTask inactivateTask = lmService.getTaskService()
                .delete(Record.class)
                .sourceMatchExtractor("record.xml")
                .matchById()
                .stageListener(new DryDeleteStageListener<Subject>(r -> r.setActive(false)))
                .batchSize(10000)
                .task();

Then its execution report will look like:

Deleted=N

Though really there are no records have been deleted.

For example, the cause of the issue for delete task stats is current implementation in com.nhl.link.move.runtime.task.delete.DeleteStatsListener for @AfterMissingTargetsFiltered event. It increments delete stats value as:

e.getStats().incrementDeleted(segment.getMissingTargets().height());

[andrus]

Fixed in 3.0.M1 as we switched to collecting statistics using @AfterTargetsCommitted event