/
recommender.go
206 lines (189 loc) · 8.32 KB
/
recommender.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
package nodescorer
import (
"context"
"fmt"
"net/http"
"time"
"github.com/elankath/scaler-simulator/pricing"
"k8s.io/apimachinery/pkg/api/resource"
"github.com/gardener/gardener/pkg/apis/core/v1beta1"
scalesim "github.com/elankath/scaler-simulator"
"github.com/elankath/scaler-simulator/simutil"
"github.com/elankath/scaler-simulator/webutil"
corev1 "k8s.io/api/core/v1"
)
type Recommender struct {
engine scalesim.Engine
scenarioName string
shootName string
strategyWeights scalesim.StrategyWeights
logWriter http.ResponseWriter
podOrder string
}
func NewRecommender(engine scalesim.Engine, scenarioName, shootName, podOrder string, strategyWeights scalesim.StrategyWeights, logWriter http.ResponseWriter) *Recommender {
return &Recommender{
engine: engine,
scenarioName: scenarioName,
shootName: shootName,
strategyWeights: strategyWeights,
logWriter: logWriter,
podOrder: podOrder,
}
}
func (r *Recommender) Run(ctx context.Context, shoot *v1beta1.Shoot, unscheduledPods []corev1.Pod) (scalesim.Recommendations, error) {
startTime := time.Now()
defer func() {
webutil.Logf(r.logWriter, "Execution of scenario: %s completed in %v", r.scenarioName, time.Since(startTime))
}()
recommendations := make(scalesim.Recommendations)
var runCounter int
for {
runCounter++
var nodeScores scalesim.NodeRunResults
unscheduledPodCount := len(unscheduledPods)
if unscheduledPodCount == 0 {
webutil.Log(r.logWriter, "All pods are scheduled. Exiting the loop...")
break
}
webutil.Log(r.logWriter, fmt.Sprintf("Scenario-Run #%d", runCounter))
nodeScores = r.computeNodeScores(ctx, shoot, unscheduledPods)
if len(nodeScores) == 0 {
webutil.Log(r.logWriter, fmt.Sprintf("In Scenario-Run #%d, no pods could be assgined, exiting early", runCounter))
break
}
winnerNodeScore := nodeScores.GetWinner()
webutil.Log(r.logWriter, "Winning Score: "+winnerNodeScore.String())
recommendation := recommendations[winnerNodeScore.Pool.Name]
if recommendation == nil {
recommendation = &scalesim.Recommendation{
Waste: *resource.NewQuantity(0, resource.BinarySI),
Allocatable: *resource.NewQuantity(0, resource.BinarySI),
}
}
recommendation.WorkerPoolName = winnerNodeScore.Pool.Name
recommendation.Replicas = recommendation.Replicas + 1
recommendation.Cost = pricing.GetPricing(winnerNodeScore.Pool.Machine.Type) * (float64(recommendation.Replicas))
checkEventsSince := time.Now()
scaledNode, err := r.scaleWorker(ctx, winnerNodeScore.Pool)
if err != nil {
r.logError(err)
return recommendations, err
}
if err = r.engine.VirtualClusterAccess().CreatePods(ctx, r.podOrder, unscheduledPods...); err != nil {
r.logError(err)
return recommendations, err
}
_, _, err = simutil.WaitForAndRecordPodSchedulingEvents(ctx, r.engine.VirtualClusterAccess(), r.logWriter, checkEventsSince, unscheduledPods, 10*time.Second)
if err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return recommendations, err
}
assignedPods, err := simutil.GetPodsAssignedToNode(ctx, r.engine.VirtualClusterAccess(), scaledNode.Name)
if err != nil {
r.logError(err)
return recommendations, err
}
webutil.Log(r.logWriter, fmt.Sprintf("At the end of run #%d, winner node %s is assigned pods: %v", runCounter, scaledNode.Name, simutil.PodNames(assignedPods)))
if len(assignedPods) == 0 {
webutil.Log(r.logWriter, "No pods are assigned to the winning scaled node "+scaledNode.Name)
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error")
return recommendations, err
}
nodePodAssignments, err := simutil.GetNodePodAssignments(ctx, r.engine.VirtualClusterAccess())
if err != nil {
r.logError(err)
return recommendations, err
}
webutil.Log(r.logWriter, fmt.Sprintf("At the end of run #%d, Node Pod Assignments: %v", runCounter, nodePodAssignments))
recommendation.Waste.Add(simutil.ComputeNodeWaste(scaledNode, assignedPods))
recommendation.Allocatable.Add(*scaledNode.Status.Allocatable.Memory())
recommendations[winnerNodeScore.Pool.Name] = recommendation
allAssignedPods, err := simutil.GetAllAssignedPods(ctx, r.engine.VirtualClusterAccess())
if err != nil {
r.logError(err)
return recommendations, err
}
unscheduledPods = simutil.DeleteAssignedPods(unscheduledPods, allAssignedPods)
webutil.Log(r.logWriter, fmt.Sprintf("Removing unscheduled pods #%d, at the end of run #%d", len(unscheduledPods), runCounter))
if err = r.engine.VirtualClusterAccess().DeletePods(ctx, unscheduledPods...); err != nil {
r.logError(err)
return recommendations, err
}
}
return recommendations, nil
}
func (r *Recommender) scaleWorker(ctx context.Context, worker *v1beta1.Worker) (*corev1.Node, error) {
scaledNode, err := simutil.CreateNodeInWorkerGroup(ctx, r.engine.VirtualClusterAccess(), worker)
if err != nil {
r.logError(err)
return nil, err
}
return scaledNode, err
}
func (r *Recommender) computeNodeScores(ctx context.Context, shoot *v1beta1.Shoot, candidatePods []corev1.Pod) scalesim.NodeRunResults {
nodeScores := scalesim.NodeRunResults(make(map[string]scalesim.NodeRunResult))
var checkEventsSince time.Time
for _, pool := range shoot.Spec.Provider.Workers {
//if err = r.engine.VirtualClusterAccess().DeletePods(ctx, candidatePods...); err != nil {
// webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
// return nil, candidatePods
//}
webutil.Logf(r.logWriter, "Scaling workerpool %s...", pool.Name)
checkEventsSince = time.Now()
scaledNode, err := simutil.CreateNodeInWorkerGroup(ctx, r.engine.VirtualClusterAccess(), &pool)
if err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return nil
}
if scaledNode == nil {
webutil.Log(r.logWriter, "No new node can be created for pool "+pool.Name+" as it has reached its max. Skipping this pool.")
continue
}
if err = r.engine.VirtualClusterAccess().CreatePods(ctx, r.podOrder, candidatePods...); err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return nil
}
_, _, err = simutil.WaitForAndRecordPodSchedulingEvents(ctx, r.engine.VirtualClusterAccess(), r.logWriter, checkEventsSince, candidatePods, 10*time.Second)
if err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return nil
}
allPods, err := r.engine.VirtualClusterAccess().ListPods(ctx)
if err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return nil
}
candidatePods = simutil.GetMatchingPods(allPods, candidatePods)
nodeScore, err := computeNodeRunResult(r.strategyWeights, scaledNode, candidatePods, shoot.Spec.Provider.Workers)
if err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return nil
}
webutil.Log(r.logWriter, "Resetting virtual cluster state for next scale-up...")
if err = r.resetRunState(ctx, scaledNode.Name, candidatePods); err != nil {
return nil
}
if nodeScore.NumAssignedPodsToNode == 0 {
webutil.Log(r.logWriter, "No pods are scheduled on the candidate scaled node "+scaledNode.Name+". This node group will not be considered in this run")
continue
}
nodeScores[scaledNode.Name] = nodeScore
webutil.Log(r.logWriter, fmt.Sprintf("Node score for %s: %v", scaledNode.Name, nodeScores[scaledNode.Name]))
}
return nodeScores
}
func (r *Recommender) logError(err error) {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
}
func (r *Recommender) resetRunState(ctx context.Context, nodeName string, pods []corev1.Pod) error {
webutil.Log(r.logWriter, "Deleting scaled node..."+nodeName)
if err := r.engine.VirtualClusterAccess().DeleteNode(ctx, nodeName); err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return err
}
if err := r.engine.VirtualClusterAccess().DeletePods(ctx, pods...); err != nil {
webutil.Log(r.logWriter, "Execution of scenario: "+r.scenarioName+" completed with error: "+err.Error())
return err
}
return nil
}