Skip to main content

Fix “composite node loops or stalls”

Stub

This How-to is a stub. JARVIS supports budgets and recovery artifacts, but durable scheduling and loop prevention are still evolving.

Goal​

You will stop composite execution loops/stalls by tuning budgets, tightening recovery, and inspecting artifacts.

When to use this​

  • A run keeps creating more node runs without finishing.
  • A composite node run never reaches a terminal state.

Prerequisites​

  • Budgets/constraints enabled
  • Access to artifacts/events for the run

Steps​

  1. Inspect decomposition output (subtasks) for runaway expansion.
  2. Inspect recovery actions for repeated retries/remaps.
  3. Tighten budgets:
    • max depth,
    • max subtasks,
    • max decomposition rounds,
    • max retries/remaps.
  4. Fail closed when budgets are exceeded.

Verify​

  • The run terminates deterministically (success or safe failure).

Troubleshooting​

  • Planner keeps expanding → enforce hard bounds after parsing.
  • Binding keeps choosing failing nodes → add remap limits and better candidate filtering.
  • Stalls after creating children → scheduling/dispatch issue → inspect coordinator dispatch configuration.

Cleanup / Rollback​

  • Reset budgets to defaults after fixing root cause.

Next steps​