Workflow Lifecycle
From authoring through execution, retries, proofs, failures, and completion—for Cascades runs.
Understanding the lifecycle clarifies dashboards, SLA expectations, when proofs emit, and how operators recover degraded runs.
Creation & enqueue
Operators author definitions in the composer or programmatically persist definitions. Runs are queued against an immutable workflow snapshot at submission time—the live graph definition can drift without rewriting historical truth.
Execution may be queued via workers or inline in development configurations—see queue workers.
Execution waves
DAG evaluation walks predecessors to determine ready tasks. Deterministic orchestration emits attempts, logs transitions, and records completion metadata suitable for dashboards and proofs.
Retries & failures
Retry policies backoff transient faults; exhaustion surfaces task-level failure and may flip run state—see retries & timeouts and failure recovery.
Proofs & completion
High-trust environments rely on hashed proof payloads anchored to enqueue-time hashes. Terminal success states align with observable completion and optional transparency anchoring—see Execution proofs.
Operational recovery
Stale locks or poison messages may require DLQ / recovery procedures while preserving audit lineage.