Cycles and Re-entry Safety

Visualization boundary: the diagrams show loop semantics only; JSON/YAML export still depends on named handlers and conditions.

TriggerFlow can implement loops, but the recommended form is always a loop with a provable stop condition.

1. Three safe loop patterns

How to read this diagram

• Safety does not come from “being able to loop”; it comes from clear stop conditions and clear control ownership.
• These three modes correspond to internal control, approval-driven control, and externally driven control.

2. Bounded self-loop

The simplest safe pattern is to encode an explicit termination rule:

async def loop_step(data):
    count = int(data.state.get("count", 0) or 0)
    if count >= 3:
        data.set_result({"done": True, "count": count})
        return
    data.state.set("count", count + 1, emit=False)
    await data.async_emit("Loop", count + 1)

3. Pause/resume round loop

A more stable pattern is to pause after each round and let an external system decide whether to continue:

• step -> pause_for()
• external continue_with()
• ResumeLoop -> emit("Loop")

Suitable for:

• human-in-the-loop workflows
• approval flows
• long-running transactions

4. External event re-entry

Often the safest “loop” is no self-loop at all. The execution simply waits for repeated outside re-entry:

execution.emit("Tick", 1)
execution.emit("Tick", 1)
execution.emit("Tick", 1)

Suitable for:

• webhooks
• MQ consumers
• scheduled triggers

5. Key principles

• every round must have an explicit exit condition
• if the loop owns the final result, call set_result() or end the chain explicitly
• never treat infinite spinning as the default control flow