Merge pull request #37 from filecoin-saturn/metrics/pool

add a metric of pool health

metrics.go

@@ -17,6 +17,11 @@ var (
 		Help: "Number of active caboose peers",
 	})
 
+	poolHealthMetric = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: prometheus.BuildFQName("ipfs", "caboose", "pool_health"),
+		Help: "Health of the caboose pool",
+	}, []string{"weight"})
+
 	fetchResponseMetric = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Name: prometheus.BuildFQName("ipfs", "caboose", "fetch_errors"),
 		Help: "Errors fetching from Caboose Peers",
@@ -42,6 +47,7 @@ var (
 func init() {
 	CabooseMetrics.MustRegister(poolErrorMetric)
 	CabooseMetrics.MustRegister(poolSizeMetric)
+	CabooseMetrics.MustRegister(poolHealthMetric)
 	CabooseMetrics.MustRegister(fetchResponseMetric)
 	CabooseMetrics.MustRegister(fetchSpeedMetric)
 	CabooseMetrics.MustRegister(fetchLatencyMetric)

pool.go

@@ -164,6 +164,21 @@ func (p *pool) doRefresh() {
 		}
 		p.lk.Unlock()
 		poolSizeMetric.Set(float64(len(n)))
+
+		// periodic update of a pool health metric
+		p.lk.RLock()
+		byWeight := make(map[int]int)
+		for _, m := range p.endpoints {
+			if _, ok := byWeight[m.replication]; !ok {
+				byWeight[m.replication] = 0
+			}
+			byWeight[m.replication] += 1
+		}
+		poolHealthMetric.Reset()
+		for weight, cnt := range byWeight {
+			poolHealthMetric.WithLabelValues(fmt.Sprintf("%d", weight)).Set(float64(cnt))
+		}
+		p.lk.RUnlock()
 	} else {
 		poolErrorMetric.Add(1)
 	}