foreman

The per-queue task supervisor — Elixir Oban’s Task.Supervisor foreman (queue/supervisor.ex starts one per queue; producer.ex spawns every job through it). gleam_otp has no task supervisor, so this is a factory_supervisor whose template runs thunks: the template executes in the factory supervisor’s process, so process.spawn links the job to the supervisor — Task.Supervisor.async_nolink semantics once the producer adds its own monitor. Children are Temporary: the supervisor never restarts a job process — retries are the engine’s business.

Types

Foreman

</>

The opaque factory handle plus its pid — the pid is kept alongside because supervisor:terminate_child/2 (pkill) needs it and it cannot be extracted from the opaque Supervisor type.

pub type Foreman {
  Foreman(
    supervisor: factory_supervisor.Supervisor(fn() -> Nil, Nil),
    pid: process.Pid,
  )
}

Constructors

Foreman(
  supervisor: factory_supervisor.Supervisor(fn() -> Nil, Nil),
  pid: process.Pid,
)

Values

release_task

</>

pub fn release_task(task task: process.Pid) -> Nil

Let a start_gated_task task begin running.

start

</>

pub fn start() -> Result(actor.Started(Foreman), actor.StartError)

start_gated_task

</>

pub fn start_gated_task(
  foreman foreman: Foreman,
  run run: fn() -> Nil,
) -> Result(process.Pid, actor.StartError)

Like start_task, but the task blocks until release_task — the Task.async_nolink owner handshake. A monitored task must go through this: monitor the returned pid, THEN release it, so a fast task can never exit before the monitor exists (a completed job’s noproc DOWN would be misread as a crash and re-acked).

start_task

</>

pub fn start_task(
  foreman foreman: Foreman,
  run run: fn() -> Nil,
) -> Result(process.Pid, actor.StartError)

Run run in a fresh process linked to the foreman. The caller monitors the returned pid itself (async_nolink: linked to the supervisor, only monitored by the producer).

terminate_task

</>

pub fn terminate_task(
  foreman foreman: Foreman,
  pid pid: process.Pid,
) -> Nil

Task.Supervisor.terminate_child/2: shut the task down through its supervisor (an exit shutdown, not a brutal kill). A task that already finished is ignored, like Elixir’s {:error, :not_found}.