cpu_ctrl/main.go

package main

import (
	"context"
	"errors"
	"fmt"
	"io/ioutil"
	"os"
	"os/exec"
	"os/signal"
	"path"
	"path/filepath"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"

	"amuz.es/src/infra/cpu_ctrl/pid"

	"github.com/coreos/go-systemd/daemon"
	"github.com/shirou/gopsutil/cpu"
)

const (
	auto = 0x0
	min  = 0x04
	max  = 0x64
)

var (
	prefix = [...]int{0x3a, 0x01}
	suffix = [...]int{auto, auto, auto, auto, auto, auto}
)

type notifyType string

const (
	DaemonStarted  = notifyType("READY=1")
	DaemonStopping = notifyType("STOPPING=1")
)

func NotifyDaemon(status notifyType) {
	daemon.SdNotify(false, string(status))
}

func getProcessorCount() int {
	var maxcpu int
	stat, err := cpu.Info()
	if err != nil {
		panic(err)
	}
	for _, info := range stat {
		physicalId, err := strconv.Atoi(info.PhysicalID)
		if err != nil {
			panic(err)
		} else if maxcpu < physicalId {
			maxcpu = physicalId
		}
	}
	return maxcpu + 1
}

type Processor struct {
	Id            int
	TempeturePath string
}

type TempetureChange struct {
	Id        int
	Tempeture float64
}

func getProcessorInfo(processorId int) (*Processor, error) {
	if matches, err := filepath.Glob(fmt.Sprintf("/sys/devices/platform/coretemp.%d/hwmon/hwmon?", processorId)); err != nil {
		return nil, err
	} else if matches == nil {
		return nil, errors.New("hwmon not found!")
	} else {
		return &Processor{
			Id:            processorId,
			TempeturePath: matches[0],
		}, nil
	}
}
func ReadTempeture(path string, senseChan chan<- float64, errorChan chan<- error, waiter *sync.WaitGroup) {
	defer waiter.Done()
	if dat, err := ioutil.ReadFile(path); err != nil {
		errorChan <- err
	} else if tempetureSense, err := strconv.Atoi(strings.TrimSpace(string(dat))); err != nil {
		errorChan <- err
	} else {
		senseChan <- float64(tempetureSense) / 1000.0
	}
}

func CpuTempetureMonitoring(info *Processor, sampleDuration time.Duration, notifier chan<- TempetureChange, errorChan chan<- error, ctx context.Context, waiter *sync.WaitGroup) {
	defer waiter.Done()
	tempeturePathGlob := path.Join(info.TempeturePath, "temp?_input")
	ticker := time.Tick(sampleDuration)
	for {
		select {
		case <-ticker:
			matches, err := filepath.Glob(tempeturePathGlob)
			if err != nil {
				panic(err)
			}
			tempetureReadWaiter := &sync.WaitGroup{}
			queue := make(chan float64, len(matches))
			// exclude package temp
			for _, path := range matches[1:] {
				tempetureReadWaiter.Add(1)
				go ReadTempeture(path, queue, errorChan, tempetureReadWaiter)
			}
			tempetureReadWaiter.Wait()
			close(queue)
			var tempeture float64
			for sense := range queue {
				if tempeture < sense {
					tempeture = sense
				}
			}
			notifier <- TempetureChange{
				Id:        info.Id,
				Tempeture: tempeture,
			}
		case <-ctx.Done():
			return
		}
	}
}
func CpuTempetureScraper(processorCount int, notifier <-chan TempetureChange, errorChan chan<- error, ctx context.Context, waiter *sync.WaitGroup) {
	defer waiter.Done()
	var (
		P, I, D          = 1.5, 0.4, 2.0
		SetPoint         = 37.0
		SampleTime       = time.Second
		maxNoob, minNoob = 0x64, 0x4
		WindupGuard      = float64(maxNoob - minNoob)
	)
	noobs := make([]int, processorCount)
	controllers := make([]pid.Controller, 0, processorCount)
	for i := 0; i < processorCount; i++ {
		controller := pid.New(P, I, D)
		controller.SetSetPoint(SetPoint)
		controller.SetSampleTime(SampleTime)
		controller.SetWindupGuard(WindupGuard)
		controllers = append(controllers, controller)
	}

	for {
		select {
		case change := <-notifier:
			controller := controllers[change.Id]
			adj_noob := int(-controller.Update(change.Tempeture))
			if adj_noob < minNoob {
				adj_noob = minNoob
			} else if adj_noob > maxNoob {
				adj_noob = maxNoob
			}
			if noobs[change.Id] == adj_noob {
				continue
			}
			noobs[change.Id] = adj_noob
			fmt.Printf("cpu %d fan 0x%x\n", change.Id, adj_noob)

			args := make([]string, 0)
			args = append(args,
				"raw",
				"0x3a", "0x01",
			)
			for _, item := range noobs {
				args = append(args, fmt.Sprintf("0x%x", item))
			}
			args = append(args,
				"0x0", "0x0", "0x0", "0x0", "0x0", "0x0",
			)
			cmd := exec.Command("ipmitool", args...)
			if err := cmd.Run(); err != nil {
				errorChan <- err
				return
			}
		case <-ctx.Done():
			return
		}
	}
}

func main() {
	var (
		processorCount  = getProcessorCount()
		ctx, canceled   = context.WithCancel(context.Background())
		waiter          = &sync.WaitGroup{}
		exitSignal      = make(chan os.Signal, 1)
		errorChan       = make(chan error, 1)
		tempetureChange = make(chan TempetureChange)
		sampleDuration  = time.Second
	)

	if processorCount == 0 {
		errorChan <- errors.New("cpu not found!")
	}
	for i := 0; i < processorCount; i++ {
		if info, err := getProcessorInfo(i); err != nil {
			errorChan <- err
		} else {
			waiter.Add(1)
			go CpuTempetureMonitoring(info, sampleDuration, tempetureChange, errorChan, ctx, waiter)
		}
	}
	waiter.Add(1)
	go CpuTempetureScraper(processorCount, tempetureChange, errorChan, ctx, waiter)
	defer waiter.Wait()
	signal.Notify(exitSignal, syscall.SIGINT, syscall.SIGTERM, syscall.SIGHUP)
	NotifyDaemon(DaemonStarted)
	defer NotifyDaemon(DaemonStopping)

	select {
	case <-ctx.Done():
		fmt.Println("Service request to close this application")
	case err := <-errorChan:
		canceled()
		fmt.Printf("error! %s\n", err.Error())
	case sysSignal := <-exitSignal:
		canceled()
		fmt.Printf("SYSCALL! %s\n", sysSignal.String())
	}
}