metrics: compute CPU percentage metrics using float64 as base unit

metrics/cpu.go

@@ -17,8 +17,9 @@
 package metrics
 
 // CPUStats is the system and process CPU stats.
+// All values are in seconds.
 type CPUStats struct {
-	GlobalTime int64 // Time spent by the CPU working on all processes
-	GlobalWait int64 // Time spent by waiting on disk for all processes
-	LocalTime  int64 // Time spent by the CPU working on this process
+	GlobalTime float64 // Time spent by the CPU working on all processes
+	GlobalWait float64 // Time spent by waiting on disk for all processes
+	LocalTime  float64 // Time spent by the CPU working on this process
 }

metrics/cpu_enabled.go

@@ -38,7 +38,7 @@ func ReadCPUStats(stats *CPUStats) {
 	}
 	// requesting all cpu times will always return an array with only one time stats entry
 	timeStat := timeStats[0]
-	stats.GlobalTime = int64((timeStat.User + timeStat.Nice + timeStat.System) * cpu.ClocksPerSec)
-	stats.GlobalWait = int64((timeStat.Iowait) * cpu.ClocksPerSec)
+	stats.GlobalTime = timeStat.User + timeStat.Nice + timeStat.System
+	stats.GlobalWait = timeStat.Iowait
 	stats.LocalTime = getProcessCPUTime()
 }

metrics/cputime_nop.go

@@ -21,6 +21,6 @@ package metrics
 
 // getProcessCPUTime returns 0 on Windows as there is no system call to resolve
 // the actual process' CPU time.
-func getProcessCPUTime() int64 {
+func getProcessCPUTime() float64 {
 	return 0
 }

metrics/cputime_unix.go

@@ -26,11 +26,11 @@ import (
 )
 
 // getProcessCPUTime retrieves the process' CPU time since program startup.
-func getProcessCPUTime() int64 {
+func getProcessCPUTime() float64 {
 	var usage syscall.Rusage
 	if err := syscall.Getrusage(syscall.RUSAGE_SELF, &usage); err != nil {
 		log.Warn("Failed to retrieve CPU time", "err", err)
 		return 0
 	}
-	return int64(usage.Utime.Sec+usage.Stime.Sec)*100 + int64(usage.Utime.Usec+usage.Stime.Usec)/10000 //nolint:unconvert
+	return float64(usage.Utime.Sec+usage.Stime.Sec) + float64(usage.Utime.Usec+usage.Stime.Usec)/1000000 //nolint:unconvert
 }

metrics/metrics.go

@@ -141,9 +141,9 @@ func CollectProcessMetrics(refresh time.Duration) {
 
 	// Define the various metrics to collect
 	var (
-		cpuSysLoad            = GetOrRegisterGaugeFloat64("system/cpu/sysload", DefaultRegistry)
-		cpuSysWait            = GetOrRegisterGaugeFloat64("system/cpu/syswait", DefaultRegistry)
-		cpuProcLoad           = GetOrRegisterGaugeFloat64("system/cpu/procload", DefaultRegistry)
+		cpuSysLoad            = GetOrRegisterGauge("system/cpu/sysload", DefaultRegistry)
+		cpuSysWait            = GetOrRegisterGauge("system/cpu/syswait", DefaultRegistry)
+		cpuProcLoad           = GetOrRegisterGauge("system/cpu/procload", DefaultRegistry)
 		cpuThreads            = GetOrRegisterGauge("system/cpu/threads", DefaultRegistry)
 		cpuGoroutines         = GetOrRegisterGauge("system/cpu/goroutines", DefaultRegistry)
 		cpuSchedLatency       = getOrRegisterRuntimeHistogram("system/cpu/schedlatency", secondsToNs, nil)
@@ -167,12 +167,16 @@ func CollectProcessMetrics(refresh time.Duration) {
 	// Iterate loading the different stats and updating the meters.
 	now, prev := 0, 1
 	for ; ; now, prev = prev, now {
-		// CPU
+		// Gather CPU times.
 		ReadCPUStats(&cpustats[now])
 		refreshFreq := time.Since(lastCollectionTime).Seconds()
-		cpuSysLoad.Update(float64(cpustats[now].GlobalTime-cpustats[prev].GlobalTime) / refreshFreq)
-		cpuSysWait.Update(float64(cpustats[now].GlobalWait-cpustats[prev].GlobalWait) / refreshFreq)
-		cpuProcLoad.Update(float64(cpustats[now].LocalTime-cpustats[prev].LocalTime) / refreshFreq)
+		sysLoad := (cpustats[now].GlobalTime - cpustats[prev].GlobalTime) / refreshFreq
+		sysWait := (cpustats[now].GlobalWait - cpustats[prev].GlobalWait) / refreshFreq
+		procLoad := (cpustats[now].LocalTime - cpustats[prev].LocalTime) / refreshFreq
+		// Convert to integer percentage.
+		cpuSysLoad.Update(int64(sysLoad * 100))
+		cpuSysWait.Update(int64(sysWait * 100))
+		cpuProcLoad.Update(int64(procLoad * 100))
 
 		// Threads
 		cpuThreads.Update(int64(threadCreateProfile.Count()))