Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

support k100_ai #3

Merged
merged 1 commit into from
Nov 18, 2024
Merged

support k100_ai #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 18 additions & 13 deletions internal/pkg/dcu/corealloc.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,16 +19,18 @@ package dcu
import (
"fmt"
"strconv"
"strings"
)

func initCoreUsage(req int) string {
res := ""
i := 0
for i < req/4 {
res = res + "0"
i++
}
return res
return strings.Repeat("0", 16)
//res := ""
//i := 0
//for i <= req/4 {
// res = res + "0"
// i++
//}
//return res
}

func addCoreUsage(tot string, c string) (string, error) {
Expand Down Expand Up @@ -68,26 +70,29 @@ func byteAlloc(b int, req int) (int, int) {
remains--
res = res + 1
}
if remains <= 0 {
break
}
i++
}
return res, remains
}

func allocCoreUsage(tot string, req int) (string, error) {
i := 0
func allocCoreUsage(tot string, req int) (string, int, error) {
i := len(tot) - 1
res := ""
remains := req
for {
left := int64(0)
alloc := 0
if i < len(tot) && tot[i] != 0 {
if i >= 0 {
left, _ = strconv.ParseInt(string(tot[i]), 16, 0)
alloc, remains = byteAlloc(int(left), remains)
res = fmt.Sprintf("%s%x", res, alloc)
res = fmt.Sprintf("%x%s", alloc, res)
} else {
break
}
i++
i--
}
return res, nil
return res, remains, nil
}
4 changes: 2 additions & 2 deletions internal/pkg/dcu/corealloc_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,10 +43,10 @@ func TestAddCoreUsage(t *testing.T) {

func TestAllocCoreUsage(t *testing.T) {
str1 := "50200fff4000000"
res, _ := allocCoreUsage(str1, 16)
res, _, _ := allocCoreUsage(str1, 16)
t.Log("res=", res)
assert.Equal(t, strings.Compare(res, "afdfe0000000000"), 0)
str1 = "abcde000ad00012"
res, _ = allocCoreUsage(str1, 32)
res, _, _ = allocCoreUsage(str1, 32)
t.Log("res=", res)
}
107 changes: 79 additions & 28 deletions internal/pkg/dcu/server.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@ type Plugin struct {
Heartbeat chan bool
vidx []bool
pipeid [][]bool
coremask []string
coremask [][]string
cardtype []string
count int
}
Expand Down Expand Up @@ -87,9 +87,11 @@ func (p *Plugin) Start() error {
for idx := range p.cardtype {
p.cardtype[idx] = ""
}
p.coremask = make([]string, 16)
p.coremask = make([][]string, 16)
for idx := range p.coremask {
p.coremask[idx] = ""
p.coremask[idx] = make([]string, 2)
p.coremask[idx][0] = ""
p.coremask[idx][1] = ""
}
p.count = 0

Expand Down Expand Up @@ -136,7 +138,10 @@ func (p *Plugin) Start() error {
if index%2 == 0 {
_, err := fmt.Sscanf(val, "DCU[%d] : Card Series: %s\n", &idx, &cardtype)
if err != nil {
panic(err)
_, err := fmt.Sscanf(val, "DCU[%d] : Card Series: DCU %s\n", &idx, &cardtype)
if err != nil {
panic(err)
}
}
p.cardtype[idx] = fmt.Sprintf("%v-%v", "DCU", cardtype)
}
Expand Down Expand Up @@ -186,7 +191,8 @@ func (p *Plugin) Start() error {
}
fmt.Println("collecting pcibus=", p.pcibusid, "cores=", p.totalcores)
for idx, val := range p.totalcores {
p.coremask[idx] = initCoreUsage(val)
p.coremask[idx][0] = initCoreUsage(val)
p.coremask[idx][1] = initCoreUsage(val)
}
go p.WatchAndRegister()
return nil
Expand Down Expand Up @@ -295,7 +301,8 @@ func (p *Plugin) RefreshContainerDevices() error {
return err
}
for idx := range p.coremask {
p.coremask[idx] = initCoreUsage(p.totalcores[idx])
p.coremask[idx][0] = initCoreUsage(p.totalcores[idx])
p.coremask[idx][1] = initCoreUsage(p.totalcores[idx])
}

for _, f := range files {
Expand All @@ -312,7 +319,8 @@ func (p *Plugin) RefreshContainerDevices() error {
didx, _ = strconv.Atoi(tmpstr[2])
pid, _ = strconv.Atoi(tmpstr[3])
vdidx, _ = strconv.Atoi(tmpstr[4])
p.coremask[didx], _ = addCoreUsage(p.coremask[didx], tmpstr[5])
p.coremask[didx][0], _ = addCoreUsage(p.coremask[didx][0], tmpstr[5])
p.coremask[didx][1], _ = addCoreUsage(p.coremask[didx][1], tmpstr[6])
p.vidx[vdidx] = true
p.pipeid[didx][pid] = true
}
Expand All @@ -326,6 +334,7 @@ func (p *Plugin) RefreshContainerDevices() error {
p.vidx[vdidx] = false
p.pipeid[didx][pid] = false
os.RemoveAll("/usr/local/vgpu/dcu/" + f.Name())
os.Remove(fmt.Sprintf("/etc/vdev/vdev%d.conf", vdidx))
}
fmt.Println(f.Name())
}
Expand Down Expand Up @@ -430,10 +439,14 @@ func getIndexFromUUID(uid string) int {
}

// Create virtual vdev directory and file
func (p *Plugin) createvdevFile(current *corev1.Pod, ctr *corev1.Container, req util.ContainerDevices) (string, error) {
s := ""
func (p *Plugin) createvdevFiles(current *corev1.Pod, ctr *corev1.Container, req util.ContainerDevices) (string, error) {
var devidx, pipeid, vdevidx int
coremsk := ""
var pcibusId string
var reqcores, mem int32
var err error
coremsk1 := initCoreUsage(16)
coremsk2 := initCoreUsage(16)
reqtmp := 0
if len(req) > 1 {
return "", nil
}
Expand All @@ -442,44 +455,82 @@ func (p *Plugin) createvdevFile(current *corev1.Pod, ctr *corev1.Container, req
continue
}
idx := getIndexFromUUID(val.UUID)
pcibusId := p.pcibusid[idx]
s = fmt.Sprintf("PciBusId: %s\n", pcibusId)
reqcores := (val.Usedcores * int32(p.totalcores[idx])) / 100
coremsk, _ = allocCoreUsage(p.coremask[idx], int(reqcores))
s = s + fmt.Sprintf("cu_mask: 0x%s\n", coremsk)
s = s + fmt.Sprintf("cu_count: %d\n", reqcores)
s = s + fmt.Sprintf("mem: %d MiB\n", val.Usedmem)
s = s + fmt.Sprintf("device_id: %d\n", 0)
pcibusId = p.pcibusid[idx]
reqcores = (val.Usedcores * int32(p.totalcores[idx])) / 100
coremsk1, reqtmp, _ = allocCoreUsage(p.coremask[idx][0], int(reqcores))
if reqtmp > 0 {
coremsk2, _, _ = allocCoreUsage(p.coremask[idx][1], reqtmp)
}
mem = val.Usedmem
devidx = idx
vdevidx, err := p.AllocateVidx()
vdevidx, err = p.AllocateVidx()
if err != nil {
return "", err
}
s = s + fmt.Sprintf("vdev_id: %d\n", vdevidx)
pipeid, err = p.AllocatePipeID(idx)
if err != nil {
return "", err
}
s = s + fmt.Sprintf("pipe_id: %d\n", pipeid)
s = s + fmt.Sprintln("enable: 1")
}
cacheFileHostDirectory := "/usr/local/vgpu/dcu/" + string(current.UID) + "_" + ctr.Name + "_" + fmt.Sprint(devidx) + "_" + fmt.Sprint(pipeid) + "_" + fmt.Sprint(vdevidx) + "_" + coremsk
err := os.MkdirAll(cacheFileHostDirectory, 0777)
dirName := string(current.UID) + "_" + ctr.Name + "_" + fmt.Sprint(devidx) + "_" + fmt.Sprint(pipeid) + "_" + fmt.Sprint(vdevidx) + "_" + fmt.Sprint(coremsk1) + "_" + fmt.Sprint(coremsk2)
cacheFileHostDirectory := fmt.Sprintf("/usr/local/vgpu/dcu/%s", dirName)
err = createvdevFile(pcibusId, coremsk1, coremsk2, reqcores, mem, 0, vdevidx, pipeid, cacheFileHostDirectory, "vdev0.conf")
if err != nil {
return "", err
}
err = os.Chmod(cacheFileHostDirectory, 0777)
// support dcu-exporter
err = createvdevFile(pcibusId, coremsk1, coremsk2, reqcores, mem, devidx, vdevidx, pipeid, "/etc/vdev/", fmt.Sprintf("vdev%d.conf", vdevidx))
if err != nil {
return "", err
}
klog.Infoln("s=", s)
err = os.WriteFile(cacheFileHostDirectory+"/vdev0.conf", []byte(s), os.ModePerm)

coreUsage1, err := addCoreUsage(p.coremask[devidx][0], coremsk1)
if err != nil {
return "", err
}
p.coremask[devidx][0] = coreUsage1

coreUsage2, err := addCoreUsage(p.coremask[devidx][1], coremsk2)
if err != nil {
return "", err
}
p.coremask[devidx][1] = coreUsage2

return cacheFileHostDirectory, nil
}

func createvdevFile(pcibusId, coremsk1, coremsk2 string, reqcores, mem int32, deviceid, vdevidx, pipeid int, cacheFileHostDirectory, cacheFileName string) error {
s := ""
s = fmt.Sprintf("PciBusId: %s\n", pcibusId)
s = s + fmt.Sprintf("cu_mask: 0x%s\n", coremsk1)
s = s + fmt.Sprintf("cu_mask: 0x%s\n", coremsk2)
s = s + fmt.Sprintf("cu_count: %d\n", reqcores)
s = s + fmt.Sprintf("mem: %d MiB\n", mem)
s = s + fmt.Sprintf("device_id: %d\n", deviceid)
s = s + fmt.Sprintf("vdev_id: %d\n", vdevidx)
s = s + fmt.Sprintf("pipe_id: %d\n", pipeid)
s = s + fmt.Sprintln("enable: 1")
klog.Infoln("s=", s)

_, err := os.Stat(cacheFileHostDirectory)
if os.IsNotExist(err) {
err := os.MkdirAll(cacheFileHostDirectory, 0777)
if err != nil {
return err
}
err = os.Chmod(cacheFileHostDirectory, 0777)
if err != nil {
return err
}
}

err = os.WriteFile(fmt.Sprintf("%s/%s", cacheFileHostDirectory, cacheFileName), []byte(s), os.ModePerm)
if err != nil {
return err
}
return nil
}

func (p *Plugin) Allocate(ctx context.Context, reqs *kubeletdevicepluginv1beta1.AllocateRequest) (*kubeletdevicepluginv1beta1.AllocateResponse, error) {
var car kubeletdevicepluginv1beta1.ContainerAllocateResponse
var dev *kubeletdevicepluginv1beta1.DeviceSpec
Expand Down Expand Up @@ -544,7 +595,7 @@ func (p *Plugin) Allocate(ctx context.Context, reqs *kubeletdevicepluginv1beta1.
}
//Create vdev file
if len(devreq) < 2 && devreq[0].Usedmem < int32(p.totalmem[0]) {
filename, err := p.createvdevFile(current, &currentCtr, devreq)
filename, err := p.createvdevFiles(current, &currentCtr, devreq)
if err != nil {
util.PodAllocationFailed(nodename, current, NodeLockDCU)
return &responses, err
Expand Down
Loading