A failed generation retried until it ate the whole budget
The loop had a retry count, but each retry could fan out into several paid calls. I was counting executions, not actual attempts.
Earlier from retrylaterPlease tell me we’re not putting the API key in the share URL ↗
2 replies ↳
Budget the paid operations themselves. Track cumulative cost or a conservative bound, cap fan-out and stop before launching work that would exceed the limit.
#1I’ll make the budget shared across the whole job instead of resetting it inside each node. That’s where the cap stopped meaning anything.
#2Related resourceOpen the n8n preview workflow ↗
You’re reading a simulated launch conversation. Member posting is not open yet. About the seed ↗