-
Notifications
You must be signed in to change notification settings - Fork 91
/
advisor_helper.go
289 lines (251 loc) · 9.58 KB
/
advisor_helper.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
/*
Copyright 2022 The Katalyst 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 cpu
import (
"fmt"
v1 "k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/klog/v2"
"k8s.io/kubelet/pkg/apis/resourceplugin/v1alpha1"
"github.com/kubewharf/katalyst-core/pkg/agent/qrm-plugins/cpu/dynamicpolicy/state"
"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/assembler/headroomassembler"
"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/assembler/provisionassembler"
"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/region"
"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/types"
"github.com/kubewharf/katalyst-core/pkg/util/general"
"github.com/kubewharf/katalyst-core/pkg/util/machine"
)
func (cra *cpuResourceAdvisor) getRegionsByRegionNames(names sets.String) []region.QoSRegion {
var regions []region.QoSRegion = nil
for regionName := range names {
r, ok := cra.regionMap[regionName]
if !ok {
return nil
}
regions = append(regions, r)
}
return regions
}
func (cra *cpuResourceAdvisor) getRegionsByPodUID(podUID string) []region.QoSRegion {
var regions []region.QoSRegion = nil
for _, r := range cra.regionMap {
podSet := r.GetPods()
for uid := range podSet {
if uid == podUID {
regions = append(regions, r)
}
}
}
return regions
}
func (cra *cpuResourceAdvisor) getContainerRegions(ci *types.ContainerInfo, regionType types.QoSRegionType) ([]region.QoSRegion, error) {
var regions []region.QoSRegion
// For non-newly allocated containers, they already had regionNames,
// we can directly get the regions by regionMap.
for _, r := range cra.getRegionsByRegionNames(ci.RegionNames) {
if r.Type() == regionType {
regions = append(regions, r)
}
}
if len(regions) > 0 {
return regions, nil
}
// The regionNames of newly allocated containers are empty, if other containers of the same pod have been assigned regions,
// we can get regions by pod UID, otherwise create new region.
for _, r := range cra.getRegionsByPodUID(ci.PodUID) {
if r.Type() == regionType {
regions = append(regions, r)
}
}
return regions, nil
}
func (cra *cpuResourceAdvisor) setContainerRegions(ci *types.ContainerInfo, regions []region.QoSRegion) {
ci.RegionNames = sets.NewString()
for _, r := range regions {
ci.RegionNames.Insert(r.Name())
}
}
func (cra *cpuResourceAdvisor) getPoolRegions(poolName string) []region.QoSRegion {
pool, ok := cra.metaCache.GetPoolInfo(poolName)
if !ok || pool == nil {
return nil
}
var regions []region.QoSRegion = nil
for regionName := range pool.RegionNames {
r, ok := cra.regionMap[regionName]
if !ok {
return nil
}
regions = append(regions, r)
}
return regions
}
func (cra *cpuResourceAdvisor) setPoolRegions(poolName string, regions []region.QoSRegion) error {
pool, ok := cra.metaCache.GetPoolInfo(poolName)
if !ok {
klog.Warningf("pool %s doesn't exist, create a new pool by advisor", poolName)
return nil
}
pool.RegionNames = sets.NewString()
for _, r := range regions {
pool.RegionNames.Insert(r.Name())
}
return cra.metaCache.SetPoolInfo(poolName, pool)
}
func (cra *cpuResourceAdvisor) initializeProvisionAssembler() error {
assemblerName := cra.conf.CPUAdvisorConfiguration.ProvisionAssembler
initializers := provisionassembler.GetRegisteredInitializers()
initializer, ok := initializers[assemblerName]
if !ok {
return fmt.Errorf("unsupported provision assembler %v", assemblerName)
}
cra.provisionAssembler = initializer(cra.conf, cra.extraConf, &cra.regionMap, &cra.reservedForReclaim, &cra.numaAvailable, &cra.nonBindingNumas, cra.metaCache, cra.metaServer, cra.emitter)
return nil
}
func (cra *cpuResourceAdvisor) initializeHeadroomAssembler() error {
assemblerName := cra.conf.CPUAdvisorConfiguration.HeadroomAssembler
initializers := headroomassembler.GetRegisteredInitializers()
initializer, ok := initializers[assemblerName]
if !ok {
return fmt.Errorf("unsupported headroom assembler %v", assemblerName)
}
cra.headroomAssembler = initializer(cra.conf, cra.extraConf, &cra.regionMap, &cra.reservedForReclaim, &cra.numaAvailable, &cra.nonBindingNumas, cra.metaCache, cra.metaServer, cra.emitter)
return nil
}
// updateNumasAvailableResource updates available resource of all numa nodes.
// available = total - reserved pool - reserved for reclaim
func (cra *cpuResourceAdvisor) updateNumasAvailableResource() {
cra.numaAvailable = make(map[int]int)
reservePoolInfo, _ := cra.metaCache.GetPoolInfo(state.PoolNameReserve)
cpusPerNuma := cra.metaServer.CPUsPerNuma()
for id := 0; id < cra.metaServer.NumNUMANodes; id++ {
reservePoolNuma := 0
if cpuset, ok := reservePoolInfo.TopologyAwareAssignments[id]; ok {
reservePoolNuma = cpuset.Size()
}
reservedForReclaimNuma := 0
if v, ok := cra.reservedForReclaim[id]; ok {
reservedForReclaimNuma = v
}
cra.numaAvailable[id] = cpusPerNuma - reservePoolNuma - reservedForReclaimNuma
}
}
func (cra *cpuResourceAdvisor) getNumasReservedForAllocate(numas machine.CPUSet) float64 {
reserved := cra.conf.GetDynamicConfiguration().ReservedResourceForAllocate[v1.ResourceCPU]
return float64(reserved.Value()*int64(numas.Size())) / float64(cra.metaServer.NumNUMANodes)
}
func (cra *cpuResourceAdvisor) getRegionMaxRequirement(r region.QoSRegion) float64 {
res := 0.0
switch r.Type() {
case types.QoSRegionTypeIsolation:
cra.metaCache.RangeContainer(func(podUID string, containerName string, ci *types.ContainerInfo) bool {
if _, ok := r.GetPods()[podUID]; ok && ci.ContainerType == v1alpha1.ContainerType_MAIN {
// for pods without limits, fallback to requests instead
res += general.MaxFloat64(ci.CPULimit, ci.CPURequest)
}
return true
})
res = general.MaxFloat64(1, res)
default:
for _, numaID := range r.GetBindingNumas().ToSliceInt() {
res += float64(cra.numaAvailable[numaID])
}
}
return res
}
func (cra *cpuResourceAdvisor) getRegionMinRequirement(r region.QoSRegion) float64 {
switch r.Type() {
case types.QoSRegionTypeShare:
return types.MinShareCPURequirement
case types.QoSRegionTypeIsolation:
res := 0.0
cra.metaCache.RangeContainer(func(podUID string, containerName string, ci *types.ContainerInfo) bool {
if _, ok := r.GetPods()[podUID]; ok && ci.ContainerType == v1alpha1.ContainerType_MAIN {
// todo: to be compatible with resource over-commit,
// set lower-bound as limit too, but we need to reconsider this in the future
res += general.MaxFloat64(ci.CPULimit, ci.CPURequest)
}
return true
})
res = general.MaxFloat64(1, res)
return res
case types.QoSRegionTypeDedicatedNumaExclusive:
return types.MinDedicatedCPURequirement
default:
klog.Errorf("[qosaware-cpu] unknown region type %v", r.Type())
return 0.0
}
}
func (cra *cpuResourceAdvisor) getRegionReservedForReclaim(r region.QoSRegion) float64 {
res := 0.0
for _, numaID := range r.GetBindingNumas().ToSliceInt() {
res += float64(cra.reservedForReclaim[numaID])
}
return res
}
func (cra *cpuResourceAdvisor) getRegionReservedForAllocate(r region.QoSRegion) float64 {
res := 0.0
for _, numaID := range r.GetBindingNumas().ToSliceInt() {
divider := cra.numRegionsPerNuma[numaID]
if divider < 1 {
divider = 1
}
res += cra.getNumasReservedForAllocate(machine.NewCPUSet(numaID)) / float64(divider)
}
return res
}
func (cra *cpuResourceAdvisor) updateRegionEntries() {
entries := make(types.RegionEntries)
for regionName, r := range cra.regionMap {
regionInfo := &types.RegionInfo{
RegionName: r.Name(),
RegionType: r.Type(),
OwnerPoolName: r.OwnerPoolName(),
BindingNumas: r.GetBindingNumas(),
}
if r.Type() == types.QoSRegionTypeShare || r.Type() == types.QoSRegionTypeDedicatedNumaExclusive {
headroom, err := r.GetHeadroom()
if err != nil {
general.ErrorS(err, "failed to get region headroom", "regionName", r.Name())
headroom = types.InvalidHeadroom
}
regionInfo.Headroom = headroom
regionInfo.HeadroomPolicyTopPriority, regionInfo.HeadroomPolicyInUse = r.GetHeadRoomPolicy()
controlKnobMap, err := r.GetProvision()
if err != nil {
controlKnobMap = types.InvalidControlKnob
general.ErrorS(err, "failed to get region provision", "regionName", r.Name())
}
regionInfo.ControlKnobMap = controlKnobMap
regionInfo.ProvisionPolicyTopPriority, regionInfo.ProvisionPolicyInUse = r.GetProvisionPolicy()
}
entries[regionName] = regionInfo
general.InfoS("region info", "HeadroomPolicyTopPriority", regionInfo.HeadroomPolicyTopPriority,
"HeadroomPolicyInUse", regionInfo.HeadroomPolicyInUse, "ProvisionPolicyTopPriority", regionInfo.ProvisionPolicyTopPriority,
"ProvisionPolicyInUse", regionInfo.ProvisionPolicyInUse, "regionName", r.Name(), "bindingNUMAs", r.GetBindingNumas().String(),
"headroom", regionInfo.Headroom)
}
_ = cra.metaCache.SetRegionEntries(entries)
}
func (cra *cpuResourceAdvisor) updateRegionStatus() {
for regionName, r := range cra.regionMap {
r.UpdateStatus()
regionInfo, ok := cra.metaCache.GetRegionInfo(regionName)
if !ok {
continue
}
status := r.GetStatus()
regionInfo.RegionStatus = status
_ = cra.metaCache.SetRegionInfo(regionName, regionInfo)
}
}