worker

Documentation

Overview

Package worker runs the consumer side of dispatch: a loop that leases tasks from a queue, executes them on a node, and reports results. The control plane runs workers as goroutines for local scaling; `dispatch work` runs the same loop in its own process, which is what Kubernetes deployments and serverless containers replicate to scale out.

Index

• type Worker
  • func (w *Worker) Run(ctx context.Context) error

Types

type Worker

type Worker struct {
	// Deployment names the deployment, for metric labels.
	Deployment string
	// Queue supplies tasks; competing workers share it.
	Queue queue.Queue
	// Results receives each task's outcome.
	Results queue.Results
	// Node executes the tasks.
	Node node.Node
	// Recorder receives task metrics. Nil disables recording.
	Recorder metrics.Recorder
	// Backoff is the pause after a transient dequeue failure (e.g. the
	// control plane briefly unreachable). Zero means one second.
	Backoff time.Duration
}

Worker consumes tasks for one deployment on one node.

func (*Worker) Run

func (w *Worker) Run(ctx context.Context) error

Run consumes tasks until ctx is done, which is the only error it returns. Task failures are reported as results, not returned: a bad task must never take the consumer down with it.