-
Notifications
You must be signed in to change notification settings - Fork 2
/
device-plugin.go
340 lines (303 loc) · 10.9 KB
/
device-plugin.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
/*
Copyright 2022 The KubeService-Stack Authors.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package main
import (
"flag"
"fmt"
"net"
"os"
"os/signal"
"path"
"path/filepath"
"strconv"
"syscall"
"time"
"github.com/fsnotify/fsnotify"
"github.com/kubeservice-stack/common/pkg/logger"
"github.com/kubeservice-stack/cpusets-controller/pkg/client"
"github.com/kubeservice-stack/cpusets-controller/pkg/config"
"github.com/kubeservice-stack/cpusets-controller/third_party/forked/nokia/CPU-Pooler/pkg/checkpoint"
"github.com/kubeservice-stack/cpusets-controller/third_party/forked/nokia/CPU-Pooler/pkg/types"
"golang.org/x/net/context"
grpc "google.golang.org/grpc"
"k8s.io/client-go/kubernetes"
pluginapi "k8s.io/kubelet/pkg/apis/deviceplugin/v1beta1"
"k8s.io/kubernetes/pkg/kubelet/cm/cpuset"
)
var (
resourceBaseName = "cmss.cn"
cdms []*cpuDeviceManager
mainLogger = logger.GetLogger("cmd/cpusets-device-plugin", "main")
)
type cpuDeviceManager struct {
pool types.Pool
socketFile string
grpcServer *grpc.Server
sharedPoolCPUs string
poolType string
nodeTopology map[int]int
htTopology map[int]string
}
//TODO: PoC if cpuset setting could be implemented in this hook? cpuset cgroup of the container should already exist at this point (kinda)
//The DeviceIDs could be used to determine which container has them, once we have a container name parsed out from the allocation backend we could manipulate its cpuset before it is even started
//Long shot, but if it works both cpusetter and process starter would become unnecessary
func (cdm *cpuDeviceManager) PreStartContainer(ctx context.Context, psRqt *pluginapi.PreStartContainerRequest) (*pluginapi.PreStartContainerResponse, error) {
return &pluginapi.PreStartContainerResponse{}, nil
}
func (cdm *cpuDeviceManager) Start() error {
pluginEndpoint := filepath.Join(pluginapi.DevicePluginPath, cdm.socketFile)
mainLogger.Info("Starting CPU Device Plugin server", logger.Any("endpoint", pluginEndpoint))
lis, err := net.Listen("unix", pluginEndpoint)
if err != nil {
mainLogger.Error("Error. Starting CPU Device Plugin server failed", logger.Error(err))
}
cdm.grpcServer = grpc.NewServer()
// Register all services
pluginapi.RegisterDevicePluginServer(cdm.grpcServer, cdm)
go cdm.grpcServer.Serve(lis)
// Wait for server to start by launching a blocking connection
conn, err := grpc.Dial(pluginEndpoint, grpc.WithInsecure(), grpc.WithBlock(),
grpc.WithTimeout(5*time.Second),
grpc.WithDialer(func(addr string, timeout time.Duration) (net.Conn, error) {
return net.DialTimeout("unix", addr, timeout)
}),
)
if err != nil {
mainLogger.Error("Error. Could not establish connection with gRPC server", logger.Error(err))
return err
}
mainLogger.Info("CPU Device Plugin server started serving")
conn.Close()
return nil
}
func (cdm *cpuDeviceManager) cleanup() error {
pluginEndpoint := filepath.Join(pluginapi.DevicePluginPath, cdm.socketFile)
if err := os.Remove(pluginEndpoint); err != nil && !os.IsNotExist(err) {
return err
}
return nil
}
func (cdm *cpuDeviceManager) Stop() error {
mainLogger.Info("CPU Device Plugin gRPC server..")
if cdm.grpcServer == nil {
return nil
}
cdm.grpcServer.Stop()
cdm.grpcServer = nil
return cdm.cleanup()
}
func (cdm *cpuDeviceManager) ListAndWatch(e *pluginapi.Empty, stream pluginapi.DevicePlugin_ListAndWatchServer) error {
var updateNeeded = true
for {
if updateNeeded {
resp := new(pluginapi.ListAndWatchResponse)
if cdm.poolType == "shared" {
nbrOfCPUs := cdm.pool.CPUset.Size()
for i := 0; i < nbrOfCPUs*1000; i++ {
cpuID := strconv.Itoa(i)
resp.Devices = append(resp.Devices, &pluginapi.Device{ID: cpuID, Health: pluginapi.Healthy})
}
} else {
for _, cpuID := range cdm.pool.CPUset.ToSlice() {
exclusiveCore := pluginapi.Device{ID: strconv.Itoa(cpuID), Health: pluginapi.Healthy}
if numaNode, exists := cdm.nodeTopology[cpuID]; exists {
exclusiveCore.Topology = &pluginapi.TopologyInfo{Nodes: []*pluginapi.NUMANode{{ID: int64(numaNode)}}}
}
resp.Devices = append(resp.Devices, &exclusiveCore)
}
}
if err := stream.Send(resp); err != nil {
mainLogger.Error("Error. Cannot update device states", logger.Error(err))
return err
}
updateNeeded = false
}
//TODO: When is update needed ?
time.Sleep(5 * time.Second)
}
return nil
}
func (cdm *cpuDeviceManager) Allocate(ctx context.Context, rqt *pluginapi.AllocateRequest) (*pluginapi.AllocateResponse, error) {
resp := new(pluginapi.AllocateResponse)
for _, container := range rqt.ContainerRequests {
envmap := make(map[string]string)
cpusAllocated, _ := cpuset.Parse("")
for _, id := range container.DevicesIDs {
tempSet, _ := cpuset.Parse(id)
cpusAllocated = cpusAllocated.Union(tempSet)
}
if cdm.pool.HTPolicy == types.MultiThreadHTPolicy {
cpusAllocated = topology.AddHTSiblingsToCPUSet(cpusAllocated, cdm.htTopology)
}
if cdm.poolType == "shared" {
envmap["SHARED_CPUS"] = cdm.sharedPoolCPUs
} else {
envmap["EXCLUSIVE_CPUS"] = cpusAllocated.String()
}
containerResp := new(pluginapi.ContainerAllocateResponse)
mainLogger.Info("CPUs allocated: " + cpusAllocated.String() + ": Num of CPUs " +
strconv.Itoa(cpusAllocated.Size()))
containerResp.Envs = envmap
resp.ContainerResponses = append(resp.ContainerResponses, containerResp)
}
return resp, nil
}
func (cdm *cpuDeviceManager) GetDevicePluginOptions(context.Context, *pluginapi.Empty) (*pluginapi.DevicePluginOptions, error) {
dpOptions := pluginapi.DevicePluginOptions{
PreStartRequired: false,
GetPreferredAllocationAvailable: false,
}
return &dpOptions, nil
}
func (cdm *cpuDeviceManager) Register(kubeletEndpoint, resourceName string) error {
conn, err := grpc.Dial(kubeletEndpoint, grpc.WithInsecure(),
grpc.WithDialer(func(addr string, timeout time.Duration) (net.Conn, error) {
return net.DialTimeout("unix", addr, timeout)
}))
if err != nil {
mainLogger.Error("CPU Device Plugin cannot connect to Kubelet service", logger.Error(err))
return err
}
defer conn.Close()
client := pluginapi.NewRegistrationClient(conn)
request := &pluginapi.RegisterRequest{
Version: pluginapi.Version,
Endpoint: cdm.socketFile,
ResourceName: resourceName,
}
if _, err = client.Register(context.Background(), request); err != nil {
mainLogger.Error("CPU Device Plugin cannot register to Kubelet service", logger.Error(err))
return err
}
return nil
}
func (cdm *cpuDeviceManager) GetPreferredAllocation(context.Context, *pluginapi.PreferredAllocationRequest) (*pluginapi.PreferredAllocationResponse, error) {
return &pluginapi.PreferredAllocationResponse{}, nil
}
func newCPUDeviceManager(poolName string, pool types.Pool, sharedCPUs string) *cpuDeviceManager {
mainLogger.Info("Starting plugin for pool: " + poolName)
return &cpuDeviceManager{
pool: pool,
socketFile: fmt.Sprintf("cpudp_%s.sock", poolName),
sharedPoolCPUs: sharedCPUs,
poolType: types.DeterminePoolType(poolName),
nodeTopology: topology.GetNodeTopology(),
htTopology: topology.GetHTTopology(),
}
}
func validatePools(poolConf types.PoolConfig) (string, error) {
var sharedCPUs string
var err error
for poolName, pool := range poolConf.Pools {
poolType := types.DeterminePoolType(poolName)
if poolType == types.SharedPoolID {
if sharedCPUs != "" {
err = fmt.Errorf("Only one shared pool allowed")
mainLogger.Error("Pool config error", logger.Any("poolConf", poolConf), logger.Error(err))
break
}
sharedCPUs = pool.CPUset.String()
}
}
return sharedCPUs, err
}
func createCDMs(poolConf types.PoolConfig, sharedCPUs string) error {
var err error
for poolName, pool := range poolConf.Pools {
poolType := types.DeterminePoolType(poolName)
//Deault or unrecognizable pools need not be made available to Device Manager as schedulable devices
if poolType == types.DefaultPoolID {
continue
}
cdm := newCPUDeviceManager(poolName, pool, sharedCPUs)
cdms = append(cdms, cdm)
if err := cdm.Start(); err != nil {
mainLogger.Error("cpuDeviceManager.Start() failed", logger.Error(err))
break
}
resourceName := resourceBaseName + "/" + poolName
err := cdm.Register(path.Join(pluginapi.DevicePluginPath, "kubelet.sock"), resourceName)
if err != nil {
// Stop server
cdm.grpcServer.Stop()
mainLogger.Error("Register error", logger.Error(err))
break
}
mainLogger.Info("CPU device plugin registered with the Kubelet")
}
return err
}
func createPluginsForPools(c kubernetes.Interface) error {
files, err := filepath.Glob(filepath.Join(pluginapi.DevicePluginPath, config.FileMatch))
if err != nil {
mainLogger.Error("filepath glob error!", logger.Error(err))
}
for _, f := range files {
if err := os.Remove(f); err != nil {
mainLogger.Error("os.Remove error!", logger.Error(err))
}
}
poolConf, err := types.DeterminePoolConfig(c, config.FileMatch, config.NodeName)
if err != nil {
mainLogger.Error("types.DeterminePoolConfig error!", logger.Error(err))
}
mainLogger.Info("Pool configuration", logger.Any("poolconf", poolConf))
var sharedCPUs string
sharedCPUs, err = validatePools(poolConf)
if err != nil {
return err
}
if err := createCDMs(poolConf, sharedCPUs); err != nil {
for _, cdm := range cdms {
cdm.Stop()
}
}
return err
}
func main() {
flag.Parse()
watcher, _ := fsnotify.NewWatcher()
watcher.Add(path.Join(pluginapi.DevicePluginPath, "kubelet.sock"))
defer watcher.Close()
// respond to syscalls for termination
sigCh := make(chan os.Signal, 1)
signal.Notify(sigCh, syscall.SIGHUP, syscall.SIGINT, syscall.SIGTERM, syscall.SIGQUIT)
_ = client.KubeClient()
if err := createPluginsForPools(client.Clientset); err != nil {
mainLogger.Error("Failed to start device plugin", logger.Error(err))
}
/* Monitor file changes for kubelet socket file and termination signals */
for {
select {
case sig := <-sigCh:
switch sig {
case syscall.SIGTERM, syscall.SIGQUIT, syscall.SIGINT:
mainLogger.Info("Received signal, shutting down.", logger.Any("signal", sig))
for _, cdm := range cdms {
cdm.Stop()
}
return
}
mainLogger.Info("Received signal!", logger.Any("signal", sig))
case event := <-watcher.Events:
mainLogger.Info("Kubelet change event in pluginpath ", logger.Any("event", event))
for _, cdm := range cdms {
cdm.Stop()
}
cdms = nil
if err := createPluginsForPools(client.Clientset); err != nil {
panic("Failed to restart device plugin")
}
}
}
}