process_metrics_linux.go: avoid precision loss when calculating process_cpu_seconds_total metric value

Previously the sum of two floating-point numbers - utime and stime - could lead to precision loss,
which is usually visible as too many digits after the floating point such as 1.29999999999999 instead of 1.3
or 2.00000000000001 instead of 2.

This should improve the compression rate at VictoriaMetrics side for this metric.

Author: valyala

process_metrics_linux.go

@@ -79,8 +79,13 @@ func writeProcessMetrics(w io.Writer) {
 
 	utime := float64(p.Utime) / userHZ
 	stime := float64(p.Stime) / userHZ
+
+	// Calculate totalTime by dividing the sum of p.Utime and p.Stime by userHZ.
+	// This reduces possible floating-point precision loss
+	totalTime := float64(p.Utime+p.Stime) / userHZ
+
 	WriteCounterFloat64(w, "process_cpu_seconds_system_total", stime)
-	WriteCounterFloat64(w, "process_cpu_seconds_total", utime+stime)
+	WriteCounterFloat64(w, "process_cpu_seconds_total", totalTime)
 	WriteCounterFloat64(w, "process_cpu_seconds_user_total", utime)
 	WriteCounterUint64(w, "process_major_pagefaults_total", uint64(p.Majflt))
 	WriteCounterUint64(w, "process_minor_pagefaults_total", uint64(p.Minflt))