main.go

package backend

import (
	"fmt"
	"get-container/docker"
	"get-container/gpu"
	"get-container/utils"
	"maps"
	"math"
	"os"
	"strconv"
	"strings"
	"sync"
	"time"

	"github.com/shirou/gopsutil/v4/process"
)

/*
	backend包为tui包提供数据
*/

var (
	MapIdDCU      = sync.Map{} // 记录dcu信息
	ContainerInfo *docker.ContainersInfo
	DockerPidInfo *DockerProcessMap = nil
	User                            = ""
	HostName                        = ""
)

func init() {
	ContainerInfo = docker.NewContainersInfo()
	ContainerInfo.Update()
	i, err := docker.ContainerInfo.GetProcessIdInDocker(false)
	DockerPidInfo = &DockerProcessMap{lock: sync.RWMutex{}, pids: make(map[int32]bool)}
	if err == nil && i != nil {
		for _, v := range i {
			for _, pidInfo := range v {
				DockerPidInfo.pids[pidInfo.Pid] = true
			}
		}
	}
	HostName, _ = os.Hostname()
	uid := os.Getuid()
	u, err := utils.GetSysUserById(uid)
	if err == nil && u != nil {
		User = u.Name
	} else {
		User = strconv.Itoa(uid)
	}
}

type DCUInfo struct {
	Id               int
	Name             string // full
	PerformanceLevel string
	Fan              string // full
	Temp             float32
	PwrAvg           float32
	PwrCap           float32
	BusId            string // full
	MemTotal         int    // full
	MemUsed          int    // full
	MemUsedPerent    float32
	Mig              bool // full
	DCUUTil          float32
	Ecc              bool // full
	PwrMode          string
}

// UpdateDCUInfo 更新dcu信息，full表示是否全部更新
func UpdateDCUInfo(full bool) {
	wg := sync.WaitGroup{}
	var smiAll map[int]*gpu.SMIAllOutput
	var eccInfo, migInfo map[int]bool
	var memInfo map[int]gpu.DCUMemInfo
	var runInfo map[int]gpu.DCURunningInfo
	var errSmiAll, errEcc, errMem, errRun error
	if full {
		wg.Add(5)
		go func() {
			smiAll, errSmiAll = gpu.GetSMIAllOutput()
			wg.Done()
		}()
		go func() {
			eccInfo, errEcc = gpu.GetEccInfo()
			wg.Done()
		}()
	} else {
		wg.Add(3)
	}
	go func() {
		migInfo = gpu.GetMigInfo()
		wg.Done()
	}()
	go func() {
		memInfo, errMem = gpu.GetDCUMemInfo()
		wg.Done()
	}()
	go func() {
		runInfo, errRun = gpu.GetRunningInfo()
		wg.Done()
	}()
	wg.Wait()
	cache := make(map[int]DCUInfo)
	if errSmiAll == nil && smiAll != nil {
		for k, v := range smiAll {
			i, have := cache[k]
			if !have {
				i = DCUInfo{}
				i.Id = k
			}
			i.Name = v.CardSeries
			i.PerformanceLevel = v.PerLevel
			i.Fan = gpu.NA
			i.PwrCap = v.MaxPwr
			i.PwrAvg = v.AvgPwr
			i.BusId = v.PCIBus
			i.MemUsedPerent = v.HCUMemUsage
			i.DCUUTil = v.HCUUsage
			cache[k] = i
		}
	}
	if errEcc == nil && eccInfo != nil {
		for k, v := range eccInfo {
			i, have := cache[k]
			if !have {
				i = DCUInfo{}
				i.Id = k
			}
			i.Ecc = v
			cache[k] = i
		}
	}
	for k, v := range migInfo {
		i, have := cache[k]
		if !have {
			i = DCUInfo{}
			i.Id = k
		}
		i.Mig = v
		cache[k] = i
	}
	if memInfo != nil && errMem == nil {
		for k, v := range memInfo {
			i, have := cache[k]
			if !have {
				i = DCUInfo{}
				i.Id = k
			}
			i.MemTotal = int(v.Total.Num)
			i.MemUsed = int(v.Used.Num)
			cache[k] = i
		}
	}
	if errRun == nil && runInfo != nil {
		for k, v := range runInfo {
			i, have := cache[k]
			if !have {
				i = DCUInfo{}
				i.Id = k
			}
			i.Temp = v.Temp
			i.PwrAvg = v.AvgPower
			i.PerformanceLevel = v.PerformanceLevel
			i.MemUsedPerent = v.MemPerc
			i.DCUUTil = v.HCUPerc
			i.PwrMode = v.Mode
			cache[k] = i
		}
	}
	for k, v := range cache {
		old, have := MapIdDCU.LoadOrStore(k, &v)
		if !have {
			continue
		}
		d := old.(*DCUInfo)
		if full {
			d.Name = v.Name
			d.PerformanceLevel = v.PerformanceLevel
			d.Fan = v.Fan
			d.PwrCap = v.PwrCap
			d.BusId = v.BusId
			d.Mig = v.Mig
			d.Ecc = v.Ecc
		}
		d.Temp = v.Temp
		d.PwrAvg = v.PwrAvg
		d.MemTotal = v.MemTotal
		d.MemUsed = v.MemUsed
		d.MemUsedPerent = v.MemUsedPerent
		d.DCUUTil = v.DCUUTil
		d.PwrMode = v.PwrMode
	}
}

func GetDCUInfo() map[int]DCUInfo {
	result := make(map[int]DCUInfo)
	MapIdDCU.Range(func(key, value any) bool {
		id := key.(int)
		val := value.(*DCUInfo)
		result[id] = *val
		return true
	})
	return result
}

type DockerProcessMap struct {
	lock sync.RWMutex
	pids map[int32]bool
}

func (dpm *DockerProcessMap) GetPidInfo() map[int32]bool {
	rl := dpm.lock.RLocker()
	rl.Lock()
	defer rl.Unlock()
	return maps.Clone(dpm.pids)
}

func (dpm *DockerProcessMap) Update(dinfo *docker.ContainersInfo) (map[int32]bool, sync.Locker) {
	dpm.lock.Lock()
	clear(dpm.pids)
	dpm.lock.Unlock()
	rl := dpm.lock.RLocker()
	i, err := dinfo.GetProcessIdInDocker(false)
	if err != nil || i == nil {
		return dpm.pids, rl
	}
	for _, v := range i {
		for _, pidInfo := range v {
			DockerPidInfo.pids[pidInfo.Pid] = true
		}
	}
	return dpm.pids, rl
}

type DCUProcessInfo struct {
	DCU    int    // 使用的dcu号
	DCUMem string // 使用的dcu内存容量
	SDMA   int
	Info   ProcessInfo // 通用进程信息
}

type ProcessInfo struct {
	Pid      int32   // 进程号
	User     string  // 用户名或uid
	CPU      float64 // CPU使用率
	Mem      float32 // 内存使用率
	Time     string  // 占用的CPU时间
	Cmd      string  // 命令
	InDocker bool    // 是否在docker容器里
}

func getProcessInfo(pids []int32) map[int32]ProcessInfo {
	result := make(map[int32]ProcessInfo)
	if len(pids) == 0 {
		return result
	}
	dockerPids := DockerPidInfo.GetPidInfo()
	for _, pid := range pids {
		p, err := process.NewProcess(int32(pid))
		if err != nil {
			continue
		}
		item := ProcessInfo{Pid: p.Pid}
		item.User, _ = p.Username()
		item.CPU, _ = p.CPUPercent()
		item.Mem, _ = p.MemoryPercent()
		t, err := p.Times()
		if err == nil {
			item.Time = durationStr(time.Duration((t.System + t.User)) * time.Second)
		}
		item.Cmd, _ = p.Cmdline()
		a, b := dockerPids[item.Pid]
		item.InDocker = a && b
		result[p.Pid] = item
	}
	return result
}

// GetDCUProcessInfo 返回值的key为dcu index
func GetDCUProcessInfo() map[int][]DCUProcessInfo {
	result := make(map[int][]DCUProcessInfo)
	info, err := gpu.GetDCUPidInfo()
	if err != nil {
		return result
	}
	pids := make([]int32, 0)
	for _, v := range info {
		pids = append(pids, v.Pid)
	}
	pinfo := getProcessInfo(pids)
	for _, v := range info {
		index := make([]int, 0)
		for _, i := range v.HCUIndex {
			ii, err := strconv.Atoi(i)
			if err != nil {
				continue
			}
			index = append(index, ii)
		}
		for _, i := range index {
			l, have := result[i]
			if !have {
				result[i] = make([]DCUProcessInfo, 0)
				l = result[i]
			}
			item := DCUProcessInfo{DCU: i}
			item.Info = pinfo[v.Pid]
			item.DCUMem = v.VRamUsed.HumanReadStr(1)
			item.SDMA = v.SDMAUsed
			l = append(l, item)
			result[i] = l
		}
	}
	return result
}

// durationStr 将时间段格式化为 小时:分钟:秒s的格式
func durationStr(d time.Duration) string {
	h := int(math.Floor(d.Hours()))
	m := int(d.Minutes()) % 60
	s := int(math.Floor(d.Seconds())) % 60
	return strings.Replace(fmt.Sprintf("%d:%2d:%2d", h, m, s), " ", "0", -1)
}