Skip to main content

Handling the Abort Flag

The operator's cost and time limits on a workflow are enforced through a simple, elegant mechanism: every time your workflow logs a step, the platform tells it how much it has spent so far — and whether it's time to stop. That signal is the abort flag.

The log response​

Whenever a Circus node calls the platform's log endpoint — which the Agent and Log nodes do on every step — the response carries the run's running tally:

{
  "max_time": 300,
  "time_consumed": 47,
  "max_cost": 5.00,
  "cost_consumed": 0.2156,
  "abort": false
}
  • max_time / time_consumed — the workflow's time limit and how many seconds have elapsed.
  • max_cost / cost_consumed — the cost limit and how much has been spent (the platform computed this from your reported consumption).
  • abort — the whole point: true when either limit has been reached.

Note the boundary is inclusive — reaching a limit exactly trips the flag (the check is "at or over," not "strictly over").

What the nodes do with it​

You don't have to inspect this yourself. The Agent and Log nodes check abort automatically after each log call. If it's true, the node terminates the run — it tells the platform to mark the execution terminated and stops the workflow — with the reason that a cost or time threshold was exceeded. This is layer one of enforcement: the workflow stopping itself, promptly, the moment it goes over budget.

Why it isn't the only layer​

Self-termination is prompt but not guaranteed — a workflow might fail to check the flag, or the terminate call might not land. So the platform doesn't rely on your workflow cooperating: a background reconciler independently watches running executions and remotely stops any that have breached their limits. Enforcement holds even if a node's self-termination misfires.

The practical upshot for you is reassuring: place your Agent and Log nodes, and the limits enforce themselves. You get prompt self-termination for free, with a platform-side backstop behind it — no manual budget-tracking logic in your workflow at all.