Re-executes an ancestor activity in a new dimensional thread, enabling retry loops and iterative patterns without violating the DAG constraint. The cycle activity targets a specific ancestor (typically a Hook with cycle: true) and sends execution back to that point.

Each cycle iteration runs in a fresh dimensional thread — individual activity state is isolated per iteration, while shared job state (job.maps) accumulates across iterations. This pattern enables retries, polling loops, and iterative processing.

app:
id: myapp
version: '1'
graphs:
- subscribes: retry.start
expire: 300

activities:
t1:
type: trigger

pivot:
type: hook
cycle: true # marks this activity as a cycle target
output:
maps:
retryCount: 0

do_work:
type: worker
topic: work.do
output:
schema:
type: object
properties:
result: { type: string }

retry:
type: cycle
ancestor: pivot # re-execute from this activity
input:
maps:
retryCount: # increment retry counter each cycle
'@pipe':
- ['{pivot.output.data.retryCount}', 1]
- ['{@math.add}']

done:
type: hook

transitions:
t1:
- to: pivot
pivot:
- to: do_work
do_work:
- to: retry
conditions:
code: 500 # cycle on error
- to: done
  • The ancestor field must reference an activity with cycle: true.
  • The cycle activity's input.maps override the ancestor's output data for the next iteration, allowing each cycle to pass different values.
  • Dimensional isolation ensures parallel cycle iterations don't collide.

Cycle is a Category A (Leg 1 only) activity:

  • Maps input data, resolves the re-entry dimensional address, and publishes a stream message addressed to the ancestor activity.
  • The ancestor re-enters via its Leg 2 path in the new dimension.

CycleActivity for the TypeScript interface

Hierarchy (view full)

Constructors

Properties

adjacencyList: StreamData[]
adjacentIndex: number = 0
code: number = 200
context: JobState
guidLedger: number = 0
logger: ILogger
status: StreamStatus = StreamStatus.SUCCESS

Methods