Merge pull request #43 from m0ar/master

Add `CpuTime` and `Duration` quantiles to worker metrics

README.md

@@ -68,6 +68,8 @@ The original method of zone filtering by using env variables `ZONE_<name>` is no
 ## List of available metrics
 
 ```
+# HELP cloudflare_worker_cpu_time CPU time quantiles by script name
+# HELP cloudflare_worker_duration Duration quantiles by script name (GB*s)
 # HELP cloudflare_worker_errors_count Number of errors by script name
 # HELP cloudflare_worker_requests_count Number of requests sent to worker by script name
 # HELP cloudflare_zone_bandwidth_cached Cached bandwidth per zone in bytes

cloudflare.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"time"
 
-	cloudflare "github.com/cloudflare/cloudflare-go"
+	"github.com/cloudflare/cloudflare-go"
 	"github.com/machinebox/graphql"
 	log "github.com/sirupsen/logrus"
 )
@@ -43,6 +43,17 @@ type accountResp struct {
 			Errors   uint64  `json:"errors"`
 			Duration float64 `json:"duration"`
 		} `json:"sum"`
+
+		Quantiles struct {
+			CpuTimeP50   float32 `json:"cpuTimeP50"`
+			CpuTimeP75   float32 `json:"cpuTimeP75"`
+			CpuTimeP99   float32 `json:"cpuTimeP99"`
+			CpuTimeP999  float32 `json:"cpuTimeP999"`
+			DurationP50  float32 `json:"durationP50"`
+			DurationP75  float32 `json:"durationP75"`
+			DurationP99  float32 `json:"durationP99"`
+			DurationP999 float32 `json:"durationP999"`
+		} `json:"quantiles"`
 	} `json:"workersInvocationsAdaptive"`
 }
 
@@ -402,6 +413,17 @@ func fetchWorkerTotals(accountID string) (*cloudflareResponseAccts, error) {
 						errors
 						duration
 					}
+
+					quantiles {
+						cpuTimeP50
+						cpuTimeP75
+						cpuTimeP99
+						cpuTimeP999
+						durationP50
+						durationP75
+						durationP99
+						durationP999
+					}
 				}
 			}
 		}

main.go

@@ -7,7 +7,7 @@ import (
 	"sync"
 	"time"
 
-	cloudflare "github.com/cloudflare/cloudflare-go"
+	"github.com/cloudflare/cloudflare-go"
 	"github.com/namsral/flag"
 	"github.com/nelkinda/health-go"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -22,6 +22,7 @@ var (
 	cfgMetricsPath = "/metrics"
 	cfgZones       = ""
 	cfgScrapeDelay = 300
+	cfgFreeTier    = false
 )
 
 func getTargetZones() []string {
@@ -30,7 +31,7 @@ func getTargetZones() []string {
 	if len(cfgZones) > 0 {
 		zoneIDs = strings.Split(cfgZones, ",")
 	} else {
-		//depricated
+		// deprecated
 		for _, e := range os.Environ() {
 			if strings.HasPrefix(e, "ZONE_") {
 				split := strings.SplitN(e, "=", 2)
@@ -64,7 +65,6 @@ func fetchMetrics() {
 	var wg sync.WaitGroup
 	zones := fetchZones()
 	accounts := fetchAccounts()
-
 	filteredZones := filterZones(zones, getTargetZones())
 
 	for _, a := range accounts {
@@ -96,7 +96,8 @@ func main() {
 	flag.StringVar(&cfgCfAPIEmail, "cf_api_email", cfgCfAPIEmail, "cloudflare api email, works with api_key flag")
 	flag.StringVar(&cfgCfAPIToken, "cf_api_token", cfgCfAPIToken, "cloudflare api token (preferred)")
 	flag.StringVar(&cfgZones, "cf_zones", cfgZones, "cloudflare zones to export, comma delimited list")
-	flag.IntVar(&cfgScrapeDelay, "scrape_delay", cfgScrapeDelay , "scrape delay in seconds, defaults to 300")
+	flag.IntVar(&cfgScrapeDelay, "scrape_delay", cfgScrapeDelay, "scrape delay in seconds, defaults to 300")
+	flag.BoolVar(&cfgFreeTier, "free_tier", cfgFreeTier, "scrape only metrics included in free plan")
 	flag.Parse()
 	if !(len(cfgCfAPIToken) > 0 || (len(cfgCfAPIEmail) > 0 && len(cfgCfAPIKey) > 0)) {
 		log.Fatal("Please provide CF_API_KEY+CF_API_EMAIL or CF_API_TOKEN")

prometheus.go

@@ -185,6 +185,18 @@ var (
 		Help: "Number of errors by script name",
 	}, []string{"script_name"},
 	)
+
+	workerCpuTime = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "cloudflare_worker_cpu_time",
+		Help: "CPU time quantiles by script name",
+	}, []string{"script_name", "quantile"},
+	)
+
+	workerDuration = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "cloudflare_worker_duration",
+		Help: "Duration quantiles by script name (GB*s)",
+	}, []string{"script_name", "quantile"},
+	)
 )
 
 func fetchWorkerAnalytics(account cloudflare.Account, wg *sync.WaitGroup) {
@@ -200,13 +212,26 @@ func fetchWorkerAnalytics(account cloudflare.Account, wg *sync.WaitGroup) {
 		for _, w := range a.WorkersInvocationsAdaptive {
 			workerRequests.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName}).Add(float64(w.Sum.Requests))
 			workerErrors.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName}).Add(float64(w.Sum.Errors))
+			workerCpuTime.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P50"}).Set(float64(w.Quantiles.CpuTimeP50))
+			workerCpuTime.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P75"}).Set(float64(w.Quantiles.CpuTimeP75))
+			workerCpuTime.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P99"}).Set(float64(w.Quantiles.CpuTimeP99))
+			workerCpuTime.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P999"}).Set(float64(w.Quantiles.CpuTimeP999))
+			workerDuration.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P50"}).Set(float64(w.Quantiles.DurationP50))
+			workerDuration.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P75"}).Set(float64(w.Quantiles.DurationP75))
+			workerDuration.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P99"}).Set(float64(w.Quantiles.DurationP99))
+			workerDuration.With(prometheus.Labels{"script_name": w.Dimensions.ScriptName, "quantile": "P999"}).Set(float64(w.Quantiles.DurationP999))
 		}
 	}
 }
 
 func fetchZoneColocationAnalytics(zones []cloudflare.Zone, wg *sync.WaitGroup) {
 	wg.Add(1)
 	defer wg.Done()
+
+	// Colocation metrics are not available in non-enterprise zones
+	if cfgFreeTier {
+		return
+	}
 	zoneIDs := extractZoneIDs(zones)
 
 	r, err := fetchColoTotals(zoneIDs)
@@ -229,6 +254,11 @@ func fetchZoneAnalytics(zones []cloudflare.Zone, wg *sync.WaitGroup) {
 	wg.Add(1)
 	defer wg.Done()
 
+	// None of the below referenced metrics are available in the free tier
+	if cfgFreeTier {
+		return
+	}
+
 	zoneIDs := extractZoneIDs(zones)
 
 	r, err := fetchZoneTotals(zoneIDs)