aion/workflow/run

workflow.run (recorded activity dispatch) + now + random (determinism bindings)

Types

Timestamp

opaque </>

A timestamp supplied by AD’s determinism context.

The inner value is the canonical millisecond timestamp string returned by the engine. Workflow code should treat it as recorded deterministic data, not as a wall-clock reading.

pub opaque type Timestamp

Values

now

</>

pub fn now() -> Result(Timestamp, error.EngineError)

Return AD’s recorded deterministic timestamp.

This is the only time source exposed to workflow code. Workflow authors must not call wall-clock APIs such as Gleam/Erlang clocks from workflow logic, as ambient time would desynchronise replay.

random

</>

pub fn random() -> Result(Float, error.EngineError)

Draw a deterministic floating-point value from AD’s seeded RNG.

The engine keys the RNG seed on WorkflowId + RunId so replay observes the same sequence. Workflow authors must not call ambient entropy sources.

random_int

</>

pub fn random_int(
  min: Int,
  max: Int,
) -> Result(Int, error.EngineError)

Draw a deterministic integer in the engine-defined inclusive range.

Values come from AD’s seeded RNG through the FFI boundary; no wall-clock or ambient entropy binding is exposed by the SDK.

run

</>

pub fn run(
  activity_value: activity.Activity(i, o),
) -> Result(o, error.ActivityError)

Dispatch an activity through the single recorded side-effect boundary.

Plain Gleam workflow code is re-run on replay. The only recorded side-effectful path exposed by this SDK is run (and later concurrency combinators over activities); there is intentionally no generic side_effect(fn) escape hatch. The activity input is encoded with the activity’s input Codec, the engine dispatches and records via AD, and the returned payload is decoded with the output Codec.

timestamp_to_milliseconds

</>

pub fn timestamp_to_milliseconds(timestamp: Timestamp) -> Int

Return the canonical millisecond representation of a deterministic timestamp.