Skip to content

Commit 2c7df41

Browse files
committed
fix: get node in register
Signed-off-by: james <open4pd@4paradigm.com>
1 parent c16b095 commit 2c7df41

7 files changed

Lines changed: 24 additions & 24 deletions

File tree

internal/pkg/api/device/ascend/device.go

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -101,7 +101,7 @@ func (dev *Devices) GetResource(n *corev1.Node) map[string]int {
101101
resourceName: 0,
102102
}
103103
if !device.CheckHealthy(n, dev.config.ResourceName) {
104-
klog.Infof("device %s is not healthy on this node", dev.CommonWord())
104+
klog.Infof("device %s is unhealthy on this node", dev.CommonWord())
105105
return resourceMap
106106
}
107107
devInfos, err := dev.GetNodeDevices(n)
@@ -121,9 +121,9 @@ func (dev *Devices) GetResource(n *corev1.Node) map[string]int {
121121
return resourceMap
122122
}
123123

124-
func (dev *Devices) RunManager(n *corev1.Node) {
124+
func (dev *Devices) RunManager() {
125125
lmock := mock.NewMockLister(device.GetVendorName(dev.config.ResourceMemoryName))
126-
go device.Register(n, lmock, dev)
126+
go device.Register(lmock, dev)
127127
mockmanager := dpm.NewManager(lmock)
128128
klog.Infof("Running mocking dp: %s", dev.CommonWord())
129129
mockmanager.Run()

internal/pkg/api/device/device.go

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -74,7 +74,7 @@ type Devices interface {
7474
CommonWord() string
7575
GetNodeDevices(n *corev1.Node) ([]*DeviceInfo, error)
7676
GetResource(n *corev1.Node) map[string]int
77-
RunManager(n *corev1.Node)
77+
RunManager()
7878
}
7979

8080
type ResourceNames struct {
@@ -101,27 +101,27 @@ func GetDevices() map[string]Devices {
101101
}
102102

103103
func RunManagers() error {
104-
nodeName := os.Getenv("NODE_NAME")
105-
node, err := client.GetClient().CoreV1().Nodes().Get(context.Background(), nodeName, v1.GetOptions{})
106-
if err != nil {
107-
klog.Error("Get node error", err.Error())
108-
return err
109-
}
110104
for name, dev := range DevicesMap {
111105
klog.Infof("%s run manager", name)
112106
ch[name] = make(chan int)
113-
go dev.RunManager(node)
107+
go dev.RunManager()
114108
}
115109
for _, val := range ch {
116110
<-val
117111
}
118112
return nil
119113
}
120114

121-
func Register(n *corev1.Node, l *mock.MockLister, dev Devices) {
115+
func Register(l *mock.MockLister, dev Devices) {
116+
nodeName := os.Getenv("NODE_NAME")
122117
for {
123-
resourceMap := dev.GetResource(n)
124-
l.SetResource(resourceMap)
118+
node, err := client.GetClient().CoreV1().Nodes().Get(context.Background(), nodeName, v1.GetOptions{})
119+
if err != nil {
120+
klog.Error("Get node error", err.Error())
121+
} else {
122+
resourceMap := dev.GetResource(node)
123+
l.SetResource(resourceMap)
124+
}
125125
time.Sleep(time.Second * 30)
126126
}
127127
}

internal/pkg/api/device/hygon/device.go

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -90,7 +90,7 @@ func (dev *DCUDevices) GetResource(n *corev1.Node) map[string]int {
9090
memoryResourceName: 0,
9191
}
9292
if !device.CheckHealthy(n, HygonResourceCount) {
93-
klog.Infof("device %s is not healthy on this node", dev.CommonWord())
93+
klog.Infof("device %s is unhealthy on this node", dev.CommonWord())
9494
return resourceMap
9595
}
9696
devs, err := dev.GetNodeDevices(n)
@@ -110,9 +110,9 @@ func (dev *DCUDevices) GetResource(n *corev1.Node) map[string]int {
110110
return resourceMap
111111
}
112112

113-
func (dev *DCUDevices) RunManager(n *corev1.Node) {
113+
func (dev *DCUDevices) RunManager() {
114114
lmock := mock.NewMockLister(device.GetVendorName(HygonResourceMemory))
115-
device.Register(n, lmock, dev)
115+
device.Register(lmock, dev)
116116
mockmanager := dpm.NewManager(lmock)
117117
klog.Infof("Running mocking dp: %s", dev.CommonWord())
118118
mockmanager.Run()

internal/pkg/api/device/kunlun/device.go

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -100,9 +100,9 @@ func (dev *KunlunVDevices) GetResource(n *corev1.Node) map[string]int {
100100
return resourceMap
101101
}
102102

103-
func (dev *KunlunVDevices) RunManager(n *corev1.Node) {
103+
func (dev *KunlunVDevices) RunManager() {
104104
lmock := mock.NewMockLister(device.GetVendorName(KunlunResourceVCount))
105-
device.Register(n, lmock, dev)
105+
device.Register(lmock, dev)
106106
mockmanager := dpm.NewManager(lmock)
107107
klog.Infof("Running mocking dp: %s", dev.CommonWord())
108108
mockmanager.Run()

internal/pkg/api/device/nvidia/device.go

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -171,7 +171,7 @@ func (dev *NvidiaGPUDevices) GetResource(n *corev1.Node) map[string]int {
171171
memoryPercentageName: 0,
172172
}
173173
if !device.CheckHealthy(n, dev.config.ResourceCountName) {
174-
klog.Infof("device %s is not healthy on this node", dev.CommonWord())
174+
klog.Infof("device %s is unhealthy on this node", dev.CommonWord())
175175
return resourceMap
176176
}
177177
devs, err := dev.GetNodeDevices(n)
@@ -200,9 +200,9 @@ func (dev *NvidiaGPUDevices) GetResource(n *corev1.Node) map[string]int {
200200
return resourceMap
201201
}
202202

203-
func (dev *NvidiaGPUDevices) RunManager(n *corev1.Node) {
203+
func (dev *NvidiaGPUDevices) RunManager() {
204204
lmock := mock.NewMockLister(Vendor)
205-
go device.Register(n, lmock, dev)
205+
go device.Register(lmock, dev)
206206
mockmanager := dpm.NewManager(lmock)
207207
klog.Infof("Running mocking dp: %s", dev.CommonWord())
208208
mockmanager.Run()

internal/pkg/api/device/nvidia/device_test.go

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -176,7 +176,7 @@ func TestGetNodeDevices(t *testing.T) {
176176
node := tt.setupNode()
177177
dev := tt.setupDev()
178178

179-
devices, err := dev.GetNodeDevices(node)
179+
devices, err := dev.GetNodeDevices(&node)
180180

181181
if (err != nil) != tt.wantErr {
182182
t.Errorf("GetNodeDevices() error = %v, wantErr %v", err, tt.wantErr)

internal/pkg/mock/lister.go

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -84,9 +84,9 @@ func (l *MockLister) SetResource(resourceMap map[string]int) {
8484
return
8585
}
8686
l.mutex.Lock()
87+
defer l.mutex.Unlock()
8788
l.counts = resourceMap
8889
pluginNums := len(l.pluginsMap)
89-
l.mutex.Unlock()
9090

9191
if pluginNums == 0 {
9292
resourceNames := make([]string, 0, len(resourceMap))

0 commit comments

Comments
 (0)