grind/queue

Tunes how a node runs one queue. A queue needs no configuration to run: grind.with_worker derives the queues from the workers, with the defaults below. Add a Queue with grind.with_queue only to change them.

grind.new(pool)
|> grind.with_worker(mailer())
|> grind.with_queue(queue.new("mailers") |> queue.with_concurrency(20))

Concurrency is per node: each node that runs the queue claims up to this many jobs at once. A claimed attempt holds a lease in the database, renewed every third of its length by a reserved connection; an attempt whose lease expires is recovered by its worker’s abandonment policy. The lease must be at least four times the statement deadline. On shutdown a queue stops claiming and waits up to its grace for running jobs. grind.start and grind.supervised check these values and report a grind.InvalidQueue error.

Every bound in Grind is a gleam/time/duration.Duration, not a bare Int: the queue’s settings, the worker’s timeout and retry delays, the facade’s deadlines and grind.await(within:). One type keeps the unit in the call (duration.milliseconds(250), duration.seconds(30)), so a lease in seconds cannot be passed where milliseconds are read; the lease rule above compares two of them. An application holding milliseconds converts once with duration.milliseconds(ms), which needs gleam_time as a direct dependency.

pub type Queue =
  @internal Queue

pub fn name(queue: Queue) -> String

pub fn new(name: String) -> Queue

pub fn with_concurrency(queue: Queue, concurrency: Int) -> Queue

pub fn with_lease(
  queue: Queue,
  lease: duration.Duration,
) -> Queue

pub fn with_poll_interval(
  queue: Queue,
  interval: duration.Duration,
) -> Queue

pub fn with_shutdown_grace(
  queue: Queue,
  grace: duration.Duration,
) -> Queue