cmd/syncthing: Start CPU usage monitoring not from init (fixes #4183)

Starting stuff from init() is an antipattern, and the innerProcess
variable isn't 100% reliable. We should sort out the other uses of it as
well in due time.

Also removing the hack on innerProcess as I happened to see it and the
affected versions are now <1% users.

GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4185

cmd/syncthing/cpuusage.go

@@ -0,0 +1,59 @@
+// Copyright (C) 2017 The Syncthing Authors.
+//
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this file,
+// You can obtain one at https://mozilla.org/MPL/2.0/.
+
+package main
+
+import (
+	"math"
+	"time"
+
+	metrics "github.com/rcrowley/go-metrics"
+)
+
+const cpuTickRate = 5 * time.Second
+
+type cpuService struct {
+	avg  metrics.EWMA
+	stop chan struct{}
+}
+
+func newCPUService() *cpuService {
+	return &cpuService{
+		// 10 second average. Magic alpha value comes from looking at EWMA package
+		// definitions of EWMA1, EWMA5. The tick rate *must* be five seconds (hard
+		// coded in the EWMA package).
+		avg:  metrics.NewEWMA(1 - math.Exp(-float64(cpuTickRate)/float64(time.Second)/10.0)),
+		stop: make(chan struct{}),
+	}
+}
+
+func (s *cpuService) Serve() {
+	// Initialize prevUsage to an actual value returned by cpuUsage
+	// instead of zero, because at least Windows returns a huge negative
+	// number here that then slowly increments...
+	prevUsage := cpuUsage()
+	ticker := time.NewTicker(cpuTickRate)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ticker.C:
+			curUsage := cpuUsage()
+			s.avg.Update(int64((curUsage - prevUsage) / time.Millisecond))
+			prevUsage = curUsage
+			s.avg.Tick()
+		case <-s.stop:
+			return
+		}
+	}
+}
+
+func (s *cpuService) Stop() {
+	close(s.stop)
+}
+
+func (s *cpuService) Rate() float64 {
+	return s.avg.Rate()
+}