Merge pull request #305 from buildkite/remove-deleted-queues

Delete Prometheus labels for deleted queues

backend/prometheus.go

@@ -23,14 +23,16 @@ var (
 // Note: these metrics are not unique to a cluster / queue, as these labels are added to the
 // value when it is set.
 type Prometheus struct {
-	totals map[string]*prometheus.GaugeVec
-	queues map[string]*prometheus.GaugeVec
+	totals    map[string]*prometheus.GaugeVec
+	queues    map[string]*prometheus.GaugeVec
+	oldQueues map[string]map[string]struct{} // cluster -> set of queues in cluster from last collect
 }
 
 func NewPrometheusBackend() *Prometheus {
 	return &Prometheus{
-		totals: make(map[string]*prometheus.GaugeVec),
-		queues: make(map[string]*prometheus.GaugeVec),
+		totals:    make(map[string]*prometheus.GaugeVec),
+		queues:    make(map[string]*prometheus.GaugeVec),
+		oldQueues: make(map[string]map[string]struct{}),
 	}
 }
 
@@ -46,38 +48,57 @@ func (p *Prometheus) Serve(path, addr string) {
 // Note: This is called once per agent token per interval
 func (p *Prometheus) Collect(r *collector.Result) error {
 	for name, value := range r.Totals {
-		labelNames := []string{"cluster"}
 		gauge, ok := p.totals[name]
 		if !ok { // first time this metric has been seen so create a new gauge
 			gauge = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 				Name: fmt.Sprintf("buildkite_total_%s", camelToUnderscore(name)),
 				Help: fmt.Sprintf("Buildkite Total: %s", name),
-			}, labelNames)
+			}, []string{"cluster"})
 			prometheus.MustRegister(gauge)
 			p.totals[name] = gauge
 		}
 
 		// note that r.Cluster will be empty for unclustered agents, this label will be dropped by prometheus
-		gauge.WithLabelValues(r.Cluster).Set(float64(value))
+		gauge.With(prometheus.Labels{"cluster": r.Cluster}).Set(float64(value))
 	}
 
+	currentQueues := make(map[string]struct{})
+	oldQueues := p.oldQueues[r.Cluster]
 	for queue, counts := range r.Queues {
+		currentQueues[queue] = struct{}{}
+		delete(oldQueues, queue) // still current
+
 		for name, value := range counts {
 			gauge, ok := p.queues[name]
 			if !ok { // first time this metric has been seen so create a new gauge
-				labelNames := []string{"queue", "cluster"}
 				gauge = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 					Name: fmt.Sprintf("buildkite_queues_%s", camelToUnderscore(name)),
 					Help: fmt.Sprintf("Buildkite Queues: %s", name),
-				}, labelNames)
+				}, []string{"queue", "cluster"})
 				prometheus.MustRegister(gauge)
 				p.queues[name] = gauge
 			}
 
 			// note that r.Cluster will be empty for unclustered agents, this label will be dropped by prometheus
-			gauge.WithLabelValues(queue, r.Cluster).Set(float64(value))
+			gauge.With(prometheus.Labels{
+				"cluster": r.Cluster,
+				"queue":   queue,
+			}).Set(float64(value))
+		}
+	}
+
+	// oldQueues contains queues that were in the previous collector result, but
+	// are no longer present.
+	// This is to prevent accumulating label values for deleted queues.
+	for queue := range oldQueues {
+		for _, gauge := range p.queues {
+			gauge.Delete(prometheus.Labels{
+				"cluster": r.Cluster,
+				"queue":   queue,
+			})
 		}
 	}
+	p.oldQueues[r.Cluster] = currentQueues
 
 	return nil
 }