chore (Telemetry) Make time bucket count of Prometheus configurable (#99

) * Make time bucket count configurable * lint * Update telemetry/prometheus/metrics.go Co-authored-by: Cal Bera <[email protected]> Signed-off-by: gordonbear <[email protected]> * Update to use seconds * more comments * Remove unneeded conversion * lint --------- Signed-off-by: gordonbear <[email protected]> Co-authored-by: Gordon <[email protected]> Co-authored-by: Cal Bera <[email protected]>
berachain · Jun 11, 2024 · 27bf367 · 27bf367
1 parent 3a23c8b
commit 27bf367
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 5 deletions.
diff --git a/telemetry/prometheus/config.go b/telemetry/prometheus/config.go
@@ -5,10 +5,19 @@ import (
 	"regexp"
 )
 
+const (
+	// Default bucket count 1000 can satisfy the precision of p99 for most histogram stats.
+	DefaultBucketCount = 1000
+)
+
 type Config struct {
-	Enabled   bool
-	Namespace string // optional
-	Subsystem string // optional
+	Enabled              bool
+	Namespace            string // optional
+	Subsystem            string // optional
+	HistogramBucketCount int    // Number of buckets for histogram, default to 1000
+	// Number of buckets for time buckets, default to 1000.
+	// The bucket size is 0.01s(10ms), so the maximum covered time range is 10ms * TimeBucketCount.
+	TimeBucketCount int
 }
 
 func (c *Config) Validate() error {

diff --git a/telemetry/prometheus/metrics.go b/telemetry/prometheus/metrics.go
@@ -24,6 +24,7 @@ type metrics struct {
 
 // NewMetrics initializes a new instance of Prometheus metrics.
 func NewMetrics(cfg *Config) (*metrics, error) { //nolint:revive // only used as Metrics interface.
+	setDefaultCfg(cfg)
 	if err := cfg.Validate(); err != nil {
 		return nil, err
 	}
@@ -173,7 +174,8 @@ func (p *metrics) Histogram(name string, value float64, tags []string, rate floa
 			Namespace: p.cfg.Namespace,
 			Subsystem: p.cfg.Subsystem,
 			Help:      name + " histogram",
-			Buckets:   prometheus.LinearBuckets(0, rate, 10), // Adjust bucketing as necessary
+			// The maximum covered stats range is rate * HistogramBucketCount
+			Buckets: prometheus.LinearBuckets(0, rate, p.cfg.HistogramBucketCount),
 		}, labels)
 		prometheus.MustRegister(histogramVec)
 		p.histogramVecs[name] = histogramVec
@@ -196,13 +198,15 @@ func (p *metrics) Time(name string, value time.Duration, tags []string) {
 			Namespace: p.cfg.Namespace,
 			Subsystem: p.cfg.Subsystem,
 			Help:      name + " timing histogram",
-			Buckets:   prometheus.LinearBuckets(0, 1, 10), // Adjust bucketing as necessary
+			// Given bucket=0.01s(10ms), the maximum covered time range is 10ms * TimeBucketCount
+			Buckets: prometheus.LinearBuckets(0, 0.01, p.cfg.TimeBucketCount),
 		}, labels)
 		prometheus.MustRegister(histogramVec)
 		p.histogramVecs[name] = histogramVec
 	}
 
 	// Convert time.Duration to seconds since Prometheus prefers base units
+	// see https://prometheus.io/docs/practices/naming/#base-units
 	histogramVec.WithLabelValues(labelValues...).Observe(value.Seconds())
 }
 
@@ -250,3 +254,13 @@ func forceValidName(name string) string {
 
 	return string(runes)
 }
+
+// Set default values if not provided.
+func setDefaultCfg(cfg *Config) {
+	if cfg.HistogramBucketCount <= 0 {
+		cfg.HistogramBucketCount = DefaultBucketCount
+	}
+	if cfg.TimeBucketCount <= 0 {
+		cfg.TimeBucketCount = DefaultBucketCount
+	}
+}