Category
Robustness (potential edge-case failure)
Component
AICPU Scheduler
Description
When the orch submit path and scheduler thread 0 build task dependencies, the runtime constructs each task's fanin/fanout relationship. Very large fanin or fanout can make dependency wiring harder to reason about and may indicate an unexpectedly broad dependency shape.
Add a warning diagnostic when a task's fanin or fanout is greater than 16. The warning should identify the task and whether the high-water value is fanin or fanout, so workload authors can spot unusually dense dependency graphs early.
Because this is on the AICPU scheduler/orchestration path, the warning should avoid hot-path log flooding. Prefer logging only when the threshold is crossed for a task, or as a high-water diagnostic, rather than logging unconditionally in an inner loop.
Related: #959
Location
src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cppsrc/a2a3/runtime/tensormap_and_ringbuffer/runtime/scheduler/pto_scheduler.h
Proposed Fix
During dependency construction, track the computed fanin/fanout count for each task. If either count is greater than 16, emit a bounded warning log that includes at least:
- task id / slot id if available
- fanin count
- fanout count
- the dependency-construction phase where the value was observed
Keep the diagnostic bounded to avoid per-edge or per-iteration AICPU logging.
Priority
Medium (minor risk, should fix in next few releases)
Category
Robustness (potential edge-case failure)
Component
AICPU Scheduler
Description
When the orch submit path and scheduler thread 0 build task dependencies, the runtime constructs each task's fanin/fanout relationship. Very large fanin or fanout can make dependency wiring harder to reason about and may indicate an unexpectedly broad dependency shape.
Add a warning diagnostic when a task's fanin or fanout is greater than 16. The warning should identify the task and whether the high-water value is fanin or fanout, so workload authors can spot unusually dense dependency graphs early.
Because this is on the AICPU scheduler/orchestration path, the warning should avoid hot-path log flooding. Prefer logging only when the threshold is crossed for a task, or as a high-water diagnostic, rather than logging unconditionally in an inner loop.
Related: #959
Location
src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cppsrc/a2a3/runtime/tensormap_and_ringbuffer/runtime/scheduler/pto_scheduler.hProposed Fix
During dependency construction, track the computed fanin/fanout count for each task. If either count is greater than 16, emit a bounded warning log that includes at least:
Keep the diagnostic bounded to avoid per-edge or per-iteration AICPU logging.
Priority
Medium (minor risk, should fix in next few releases)