Events
Hook into the job lifecycle for logging, metrics and alerting.
The worker announces what it's doing. Listen in to feed your logs, metrics or pager.
import "github.com/rjp2525/laravel-queue-go/pkg/events"
mgr.On(events.JobProcessed, func(e events.Event) {
metrics.Timing("jobs.duration", e.Duration, "job:"+e.JobName)
})
mgr.On(events.JobFailed, func(e events.Event) {
alert.Send(fmt.Sprintf("%s failed after %d attempts: %v", e.JobName, e.Attempt, e.Error))
})
The events
| Event | Fires when | Laravel's equivalent |
|---|---|---|
events.JobProcessing |
A job is about to run | JobProcessing |
events.JobProcessed |
A handler returned nil |
JobProcessed |
events.JobExceptionOccurred |
A handler returned an error or panicked | JobExceptionOccurred |
events.JobReleasedAfterError |
A failed attempt was put back for a retry | JobReleasedAfterException |
events.JobFailed |
A job is out of attempts and has been failed | JobFailed |
events.WorkerStopping |
Run is about to return |
WorkerStopping |
A failing job that still has retries left produces JobExceptionOccurred then JobReleasedAfterError. On its last attempt it produces JobExceptionOccurred then JobFailed.
What an event carries
type Event struct {
Type EventType
JobUUID string
JobName string // the display name, usually the class
Queue string
Attempt int // 1 on the first run
Duration time.Duration // set on JobProcessed and JobFailed
Error error // set on the failure events
Connection string // reserved; not populated yet
}
JobFailed can arrive without a JobUUID or JobName. That happens when the payload couldn't be parsed at all, so there was nothing to read them from. Queue and Error are always there.
Listeners run inline
A listener is called on the goroutine that's processing the job, before the worker moves on. That keeps ordering simple, and it means two things:
- Keep them quick. A listener that makes a slow network call slows every job. Hand slow work to a channel or a goroutine.
- Make them concurrency-safe. With
Concurrencyabove 1, the same listener is called from several goroutines at once.
WorkerStopping fires once per Run, after in-flight jobs have finished, whatever the reason: Stop(), a cancelled context, MaxJobs, MaxTime or an empty queue with StopOnEmpty. It's the right place to flush metrics.
A Prometheus-shaped example
var (
processed = promauto.NewCounterVec(prometheus.CounterOpts{Name: "queue_jobs_total"}, []string{"job", "result"})
duration = promauto.NewHistogramVec(prometheus.HistogramOpts{Name: "queue_job_seconds"}, []string{"job"})
)
mgr.On(events.JobProcessed, func(e events.Event) {
processed.WithLabelValues(e.JobName, "ok").Inc()
duration.WithLabelValues(e.JobName).Observe(e.Duration.Seconds())
})
mgr.On(events.JobReleasedAfterError, func(e events.Event) {
processed.WithLabelValues(e.JobName, "retried").Inc()
})
mgr.On(events.JobFailed, func(e events.Event) {
processed.WithLabelValues(e.JobName, "failed").Inc()
})
For queue depth, poll the driver: PendingSize, DelayedSize and ReservedSize are described in going to production.