Worker configuration

A worker is configured with worker.Options, passed to worker.NewWorker. The zero value is valid — every unset field falls back to a sensible default.

w := worker.NewWorker(client, worker.Options{
    Queues:             []string{"default", "emails"},
    Concurrency:        10,
    PollInterval:       2 * time.Second,
    DefaultTimeout:     30 * time.Minute,
    DefaultMaxAttempts: 3,
    ShutdownTimeout:    0, // unbounded drain
    Logger:             slog.Default(),
})

Options

Field	Type	Default	Meaning
Queues	[]string	["default"]	Queues the worker polls.
Concurrency	int	10	Maximum jobs processed simultaneously.
PollInterval	time.Duration	2s	Idle delay between polls when no work is returned or all slots are busy.
DefaultTimeout	time.Duration	30m	Per-job deadline when a job carries none.
DefaultMaxAttempts	int	3	Attempts announced for a handler that does not set its own.
ShutdownTimeout	time.Duration	0 (unbounded)	Upper bound on the graceful drain. Zero waits for every in-flight job.
Hostname	string	OS hostname	Overrides the host portion of the worker id and the announced hostname.
Logger	*slog.Logger	slog.Default()	Runtime logger. Handler logs use a copy bound with job fields.

Concurrency and capacity

Concurrency is the hard cap on simultaneous handlers. On each poll the worker reports its free slot count as capacity, so the server never hands out more jobs than the worker can run. Reported capacity is bounded to the range 1–50. When every slot is busy, the worker idles for PollInterval instead of polling.

The HTTP client

The worker reuses the flare.Client you pass to NewWorker as its transport (authentication, retries, and the response-size cap all come from it). Build that client for long-polling:

client, _ := flare.NewClient(flare.WithHTTPClient(worker.NewPollHTTPClient()))

worker.NewPollHTTPClient() returns an *http.Client with no total timeout and a response-header timeout above the long-poll window. This matters: the poll holds a connection open for roughly 30 seconds waiting for work, so a short total client timeout would abort it mid-poll. Per-job deadlines flow through the handler's context instead, not the HTTP client.

If you supply your own client with a short non-zero total timeout, NewWorker logs a warning at startup.

Job type configuration

Per-job-type settings (queue, attempts, timeout, cron) live on the handler registration, not on Options. See Handlers and Recurring jobs.

Environment

• FLARE_API_KEY — used by the client when you do not pass an explicit key.

See also

• Handlers
• Graceful shutdown