The Shift Around Batch: PEND Tasks Still Accumulate

Hidden behind recent updates, aid 8.38.0 suffers a subtle but stubborn flaw: PEND tasks pile up after concurrency kicks in. Even with four tasks running at once, three stay stuck, stuck in PEND for over 8 minutes - like a digital traffic jam. The problem isn’t that tasks fail, but that they’re delayed by what feels like an invisible backlog.nnHere’s the breakdown:

• When --max-concurrent 4 runs five tasks, the system queues three to wait.
• Those waiting tasks don’t jump into RUN immediately - they linger.
• This delay isn’t a bug in execution, but in scheduling logic. nnWhy does this matter? In fast-paced workflows - design sprints, customer support queues, or content moderation - delays stack, eroding trust and momentum. Users expect near-instant start-up; instead, tasks sit idle while the system ‘thinks’ - a disconnect shaped by modern expectations of speed.nnBut there’s a blind spot: many users assume PEND means failure, not a wait state. Misconceptions fuel frustration. Meanwhile, safety codes in the backend fail to trigger faster task handoff, even when slots open. This isn’t a security issue, but a usability gap - one that undermines the fluidity promised by new concurrency features.nnFor now, users should treat PEND not as failure, but as a wait state - monitoring progress closely. Future fixes may require smarter task prioritization and clearer status signaling, turning PEND from a black hole into a bridge. When will our systems stop letting tasks wait longer than they should? That’s the real question behind the code.nnThe bottom line: PEND delays in aid 8.38.0 aren’t just technical - they’re cultural. Users demand speed; systems still play catch-up. How do we fix that without breaking trust?