Metrics

Every pool keeps atomic counters. Snapshot returns a copy of them:

snap := pool.Snapshot()

fmt.Printf("submitted=%d completed=%d failed=%d\n",
    snap.JobsSubmitted, snap.JobsCompleted, snap.JobsFailed)
fmt.Printf("avg=%v success=%.1f%% throughput=%.1f/s\n",
    snap.AverageJobDuration(), snap.SuccessRate()*100, snap.Throughput())

Fields

Field	Meaning
JobsSubmitted	jobs accepted into the queue
JobsCompleted	jobs that finished without error
JobsFailed	jobs that finished with an error, after all retries
JobsRetried	retry attempts that ran
JobsPanicked	attempts that panicked
ActiveWorkers	workers running a job right now
QueueLength	jobs waiting in the queue right now
TotalDuration	summed duration of finished jobs, including retries and delays
Elapsed	time since the pool was created

JobsCompleted and JobsFailed count jobs, not attempts: a job that fails twice and then succeeds adds 2 to JobsRetried and 1 to JobsCompleted. A panicking job counts in both JobsPanicked and JobsFailed.

Derived values

Method	Returns
AverageJobDuration()	TotalDuration divided by finished jobs; 0 if none
SuccessRate()	completed ÷ finished, from 0.0 to 1.0; 1.0 if none have finished
Throughput()	completed jobs per second since the pool was created

Throughput averages over the pool's whole life. For a current rate, take two snapshots and divide the difference by the time between them.

Consistency

Each counter is read atomically, but a snapshot is not taken under one lock, so under load the fields can be very slightly out of step with each other — a job may count as completed in a snapshot that does not yet count it as submitted. That is fine for dashboards and alerts; do not build exact accounting on it.

Exporting to Prometheus

go-worker has no dependencies, so it ships no Prometheus integration. Exporting a snapshot takes a few lines with prometheus.NewCounterFunc and prometheus.NewGaugeFunc:

reg.MustRegister(
    prometheus.NewCounterFunc(prometheus.CounterOpts{
        Name: "worker_jobs_completed_total",
        Help: "Jobs finished without error.",
    }, func() float64 { return float64(pool.Snapshot().JobsCompleted) }),

    prometheus.NewCounterFunc(prometheus.CounterOpts{
        Name: "worker_jobs_failed_total",
        Help: "Jobs finished with an error after all retries.",
    }, func() float64 { return float64(pool.Snapshot().JobsFailed) }),

    prometheus.NewGaugeFunc(prometheus.GaugeOpts{
        Name: "worker_queue_length",
        Help: "Jobs waiting in the queue.",
    }, func() float64 { return float64(pool.QueueLength()) }),

    prometheus.NewGaugeFunc(prometheus.GaugeOpts{
        Name: "worker_active_workers",
        Help: "Workers running a job.",
    }, func() float64 { return float64(pool.ActiveWorkers()) }),
)

For per-job latency histograms, observe durations in OnJobComplete and OnJobFailed hooks.

Scheduler metrics

Scheduled tasks have their own run counts, last error, and success rate — see Inspecting tasks.