-
Notifications
You must be signed in to change notification settings - Fork 1.9k
/
rank.go
844 lines (722 loc) · 26.4 KB
/
rank.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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
package scheduler
import (
"fmt"
"math"
"github.com/hashicorp/nomad/lib/cpuset"
"github.com/hashicorp/nomad/nomad/structs"
)
const (
// binPackingMaxFitScore is the maximum possible bin packing fitness score.
// This is used to normalize bin packing score to a value between 0 and 1
binPackingMaxFitScore = 18.0
)
// Rank is used to provide a score and various ranking metadata
// along with a node when iterating. This state can be modified as
// various rank methods are applied.
type RankedNode struct {
Node *structs.Node
FinalScore float64
Scores []float64
TaskResources map[string]*structs.AllocatedTaskResources
TaskLifecycles map[string]*structs.TaskLifecycleConfig
AllocResources *structs.AllocatedSharedResources
// Proposed is used to cache the proposed allocations on the
// node. This can be shared between iterators that require it.
Proposed []*structs.Allocation
// PreemptedAllocs is used by the BinpackIterator to identify allocs
// that should be preempted in order to make the placement
PreemptedAllocs []*structs.Allocation
}
func (r *RankedNode) GoString() string {
return fmt.Sprintf("<Node: %s Score: %0.3f>", r.Node.ID, r.FinalScore)
}
func (r *RankedNode) ProposedAllocs(ctx Context) ([]*structs.Allocation, error) {
if r.Proposed != nil {
return r.Proposed, nil
}
p, err := ctx.ProposedAllocs(r.Node.ID)
if err != nil {
return nil, err
}
r.Proposed = p
return p, nil
}
func (r *RankedNode) SetTaskResources(task *structs.Task,
resource *structs.AllocatedTaskResources) {
if r.TaskResources == nil {
r.TaskResources = make(map[string]*structs.AllocatedTaskResources)
r.TaskLifecycles = make(map[string]*structs.TaskLifecycleConfig)
}
r.TaskResources[task.Name] = resource
r.TaskLifecycles[task.Name] = task.Lifecycle
}
// RankIterator is used to iteratively yield nodes along
// with ranking metadata. The iterators may manage some state for
// performance optimizations.
type RankIterator interface {
// Next yields a ranked option or nil if exhausted
Next() *RankedNode
// Reset is invoked when an allocation has been placed
// to reset any stale state.
Reset()
}
// FeasibleRankIterator is used to consume from a FeasibleIterator
// and return an unranked node with base ranking.
type FeasibleRankIterator struct {
ctx Context
source FeasibleIterator
}
// NewFeasibleRankIterator is used to return a new FeasibleRankIterator
// from a FeasibleIterator source.
func NewFeasibleRankIterator(ctx Context, source FeasibleIterator) *FeasibleRankIterator {
iter := &FeasibleRankIterator{
ctx: ctx,
source: source,
}
return iter
}
func (iter *FeasibleRankIterator) Next() *RankedNode {
option := iter.source.Next()
if option == nil {
return nil
}
ranked := &RankedNode{
Node: option,
}
return ranked
}
func (iter *FeasibleRankIterator) Reset() {
iter.source.Reset()
}
// StaticRankIterator is a RankIterator that returns a static set of results.
// This is largely only useful for testing.
type StaticRankIterator struct {
ctx Context
nodes []*RankedNode
offset int
seen int
}
// NewStaticRankIterator returns a new static rank iterator over the given nodes
func NewStaticRankIterator(ctx Context, nodes []*RankedNode) *StaticRankIterator {
iter := &StaticRankIterator{
ctx: ctx,
nodes: nodes,
}
return iter
}
func (iter *StaticRankIterator) Next() *RankedNode {
// Check if exhausted
n := len(iter.nodes)
if iter.offset == n || iter.seen == n {
if iter.seen != n {
iter.offset = 0
} else {
return nil
}
}
// Return the next offset
offset := iter.offset
iter.offset += 1
iter.seen += 1
return iter.nodes[offset]
}
func (iter *StaticRankIterator) Reset() {
iter.seen = 0
}
// BinPackIterator is a RankIterator that scores potential options
// based on a bin-packing algorithm.
type BinPackIterator struct {
ctx Context
source RankIterator
evict bool
priority int
jobId structs.NamespacedID
taskGroup *structs.TaskGroup
memoryOversubscription bool
scoreFit func(*structs.Node, *structs.ComparableResources) float64
}
// NewBinPackIterator returns a BinPackIterator which tries to fit tasks
// potentially evicting other tasks based on a given priority.
func NewBinPackIterator(ctx Context, source RankIterator, evict bool, priority int, schedConfig *structs.SchedulerConfiguration) *BinPackIterator {
algorithm := schedConfig.EffectiveSchedulerAlgorithm()
scoreFn := structs.ScoreFitBinPack
if algorithm == structs.SchedulerAlgorithmSpread {
scoreFn = structs.ScoreFitSpread
}
iter := &BinPackIterator{
ctx: ctx,
source: source,
evict: evict,
priority: priority,
memoryOversubscription: schedConfig != nil && schedConfig.MemoryOversubscriptionEnabled,
scoreFit: scoreFn,
}
iter.ctx.Logger().Named("binpack").Trace("NewBinPackIterator created", "algorithm", algorithm)
return iter
}
func (iter *BinPackIterator) SetJob(job *structs.Job) {
iter.priority = job.Priority
iter.jobId = job.NamespacedID()
}
func (iter *BinPackIterator) SetTaskGroup(taskGroup *structs.TaskGroup) {
iter.taskGroup = taskGroup
}
func (iter *BinPackIterator) Next() *RankedNode {
OUTER:
for {
// Get the next potential option
option := iter.source.Next()
if option == nil {
return nil
}
// Get the proposed allocations
proposed, err := option.ProposedAllocs(iter.ctx)
if err != nil {
iter.ctx.Logger().Named("binpack").Error("failed retrieving proposed allocations", "error", err)
continue
}
// Index the existing network usage
netIdx := structs.NewNetworkIndex()
netIdx.SetNode(option.Node)
netIdx.AddAllocs(proposed)
// Create a device allocator
devAllocator := newDeviceAllocator(iter.ctx, option.Node)
devAllocator.AddAllocs(proposed)
// Track the affinities of the devices
totalDeviceAffinityWeight := 0.0
sumMatchingAffinities := 0.0
// Assign the resources for each task
total := &structs.AllocatedResources{
Tasks: make(map[string]*structs.AllocatedTaskResources,
len(iter.taskGroup.Tasks)),
TaskLifecycles: make(map[string]*structs.TaskLifecycleConfig,
len(iter.taskGroup.Tasks)),
Shared: structs.AllocatedSharedResources{
DiskMB: int64(iter.taskGroup.EphemeralDisk.SizeMB),
},
}
var allocsToPreempt []*structs.Allocation
// Initialize preemptor with node
preemptor := NewPreemptor(iter.priority, iter.ctx, &iter.jobId)
preemptor.SetNode(option.Node)
// Count the number of existing preemptions
allPreemptions := iter.ctx.Plan().NodePreemptions
var currentPreemptions []*structs.Allocation
for _, allocs := range allPreemptions {
currentPreemptions = append(currentPreemptions, allocs...)
}
preemptor.SetPreemptions(currentPreemptions)
// Check if we need task group network resource
if len(iter.taskGroup.Networks) > 0 {
ask := iter.taskGroup.Networks[0].Copy()
for i, port := range ask.DynamicPorts {
if port.HostNetwork != "" {
if hostNetworkValue, hostNetworkOk := resolveTarget(port.HostNetwork, option.Node); hostNetworkOk {
ask.DynamicPorts[i].HostNetwork = hostNetworkValue.(string)
} else {
iter.ctx.Logger().Named("binpack").Error(fmt.Sprintf("Invalid template for %s host network in port %s", port.HostNetwork, port.Label))
netIdx.Release()
continue OUTER
}
}
}
for i, port := range ask.ReservedPorts {
if port.HostNetwork != "" {
if hostNetworkValue, hostNetworkOk := resolveTarget(port.HostNetwork, option.Node); hostNetworkOk {
ask.ReservedPorts[i].HostNetwork = hostNetworkValue.(string)
} else {
iter.ctx.Logger().Named("binpack").Error(fmt.Sprintf("Invalid template for %s host network in port %s", port.HostNetwork, port.Label))
netIdx.Release()
continue OUTER
}
}
}
offer, err := netIdx.AssignPorts(ask)
if err != nil {
// If eviction is not enabled, mark this node as exhausted and continue
if !iter.evict {
iter.ctx.Metrics().ExhaustedNode(option.Node,
fmt.Sprintf("network: %s", err))
netIdx.Release()
continue OUTER
}
// Look for preemptible allocations to satisfy the network resource for this task
preemptor.SetCandidates(proposed)
netPreemptions := preemptor.PreemptForNetwork(ask, netIdx)
if netPreemptions == nil {
iter.ctx.Logger().Named("binpack").Debug("preemption not possible ", "network_resource", ask)
netIdx.Release()
continue OUTER
}
allocsToPreempt = append(allocsToPreempt, netPreemptions...)
// First subtract out preempted allocations
proposed = structs.RemoveAllocs(proposed, netPreemptions)
// Reset the network index and try the offer again
netIdx.Release()
netIdx = structs.NewNetworkIndex()
netIdx.SetNode(option.Node)
netIdx.AddAllocs(proposed)
offer, err = netIdx.AssignPorts(ask)
if err != nil {
iter.ctx.Logger().Named("binpack").Debug("unexpected error, unable to create network offer after considering preemption", "error", err)
netIdx.Release()
continue OUTER
}
}
// Reserve this to prevent another task from colliding
netIdx.AddReservedPorts(offer)
// Update the network ask to the offer
nwRes := structs.AllocatedPortsToNetworkResouce(ask, offer, option.Node.NodeResources)
total.Shared.Networks = []*structs.NetworkResource{nwRes}
total.Shared.Ports = offer
option.AllocResources = &structs.AllocatedSharedResources{
Networks: []*structs.NetworkResource{nwRes},
DiskMB: int64(iter.taskGroup.EphemeralDisk.SizeMB),
Ports: offer,
}
}
for _, task := range iter.taskGroup.Tasks {
// Allocate the resources
taskResources := &structs.AllocatedTaskResources{
Cpu: structs.AllocatedCpuResources{
CpuShares: int64(task.Resources.CPU),
},
Memory: structs.AllocatedMemoryResources{
MemoryMB: int64(task.Resources.MemoryMB),
},
}
if iter.memoryOversubscription {
taskResources.Memory.MemoryMaxMB = int64(task.Resources.MemoryMaxMB)
}
// Check if we need a network resource
if len(task.Resources.Networks) > 0 {
ask := task.Resources.Networks[0].Copy()
offer, err := netIdx.AssignNetwork(ask)
if offer == nil {
// If eviction is not enabled, mark this node as exhausted and continue
if !iter.evict {
iter.ctx.Metrics().ExhaustedNode(option.Node,
fmt.Sprintf("network: %s", err))
netIdx.Release()
continue OUTER
}
// Look for preemptible allocations to satisfy the network resource for this task
preemptor.SetCandidates(proposed)
netPreemptions := preemptor.PreemptForNetwork(ask, netIdx)
if netPreemptions == nil {
iter.ctx.Logger().Named("binpack").Debug("preemption not possible ", "network_resource", ask)
netIdx.Release()
continue OUTER
}
allocsToPreempt = append(allocsToPreempt, netPreemptions...)
// First subtract out preempted allocations
proposed = structs.RemoveAllocs(proposed, netPreemptions)
// Reset the network index and try the offer again
netIdx.Release()
netIdx = structs.NewNetworkIndex()
netIdx.SetNode(option.Node)
netIdx.AddAllocs(proposed)
offer, err = netIdx.AssignNetwork(ask)
if offer == nil {
iter.ctx.Logger().Named("binpack").Debug("unexpected error, unable to create network offer after considering preemption", "error", err)
netIdx.Release()
continue OUTER
}
}
// Reserve this to prevent another task from colliding
netIdx.AddReserved(offer)
// Update the network ask to the offer
taskResources.Networks = []*structs.NetworkResource{offer}
}
// Check if we need to assign devices
for _, req := range task.Resources.Devices {
offer, sumAffinities, err := devAllocator.AssignDevice(req)
if offer == nil {
// If eviction is not enabled, mark this node as exhausted and continue
if !iter.evict {
iter.ctx.Metrics().ExhaustedNode(option.Node, fmt.Sprintf("devices: %s", err))
continue OUTER
}
// Attempt preemption
preemptor.SetCandidates(proposed)
devicePreemptions := preemptor.PreemptForDevice(req, devAllocator)
if devicePreemptions == nil {
iter.ctx.Logger().Named("binpack").Debug("preemption not possible", "requested_device", req)
netIdx.Release()
continue OUTER
}
allocsToPreempt = append(allocsToPreempt, devicePreemptions...)
// First subtract out preempted allocations
proposed = structs.RemoveAllocs(proposed, allocsToPreempt)
// Reset the device allocator with new set of proposed allocs
devAllocator := newDeviceAllocator(iter.ctx, option.Node)
devAllocator.AddAllocs(proposed)
// Try offer again
offer, sumAffinities, err = devAllocator.AssignDevice(req)
if offer == nil {
iter.ctx.Logger().Named("binpack").Debug("unexpected error, unable to create device offer after considering preemption", "error", err)
continue OUTER
}
}
// Store the resource
devAllocator.AddReserved(offer)
taskResources.Devices = append(taskResources.Devices, offer)
// Add the scores
if len(req.Affinities) != 0 {
for _, a := range req.Affinities {
totalDeviceAffinityWeight += math.Abs(float64(a.Weight))
}
sumMatchingAffinities += sumAffinities
}
}
// Check if we need to allocate any reserved cores
if task.Resources.Cores > 0 {
// set of reservable CPUs for the node
nodeCPUSet := cpuset.New(option.Node.NodeResources.Cpu.ReservableCpuCores...)
// set of all reserved CPUs on the node
allocatedCPUSet := cpuset.New()
for _, alloc := range proposed {
allocatedCPUSet = allocatedCPUSet.Union(cpuset.New(alloc.ComparableResources().Flattened.Cpu.ReservedCores...))
}
// add any cores that were reserved for other tasks
for _, tr := range total.Tasks {
allocatedCPUSet = allocatedCPUSet.Union(cpuset.New(tr.Cpu.ReservedCores...))
}
// set of CPUs not yet reserved on the node
availableCPUSet := nodeCPUSet.Difference(allocatedCPUSet)
// If not enough cores are available mark the node as exhausted
if availableCPUSet.Size() < task.Resources.Cores {
// TODO preemption
iter.ctx.Metrics().ExhaustedNode(option.Node, "cores")
continue OUTER
}
// Set the task's reserved cores
taskResources.Cpu.ReservedCores = availableCPUSet.ToSlice()[0:task.Resources.Cores]
// Total CPU usage on the node is still tracked by CPUShares. Even though the task will have the entire
// core reserved, we still track overall usage by cpu shares.
taskResources.Cpu.CpuShares = option.Node.NodeResources.Cpu.SharesPerCore() * int64(task.Resources.Cores)
}
// Store the task resource
option.SetTaskResources(task, taskResources)
// Accumulate the total resource requirement
total.Tasks[task.Name] = taskResources
total.TaskLifecycles[task.Name] = task.Lifecycle
}
// Store current set of running allocs before adding resources for the task group
current := proposed
// Add the resources we are trying to fit
proposed = append(proposed, &structs.Allocation{AllocatedResources: total})
// Check if these allocations fit, if they do not, simply skip this node
fit, dim, util, _ := structs.AllocsFit(option.Node, proposed, netIdx, false)
netIdx.Release()
if !fit {
// Skip the node if evictions are not enabled
if !iter.evict {
iter.ctx.Metrics().ExhaustedNode(option.Node, dim)
continue
}
// If eviction is enabled and the node doesn't fit the alloc, check if
// any allocs can be preempted
// Initialize preemptor with candidate set
preemptor.SetCandidates(current)
preemptedAllocs := preemptor.PreemptForTaskGroup(total)
allocsToPreempt = append(allocsToPreempt, preemptedAllocs...)
// If we were unable to find preempted allocs to meet these requirements
// mark as exhausted and continue
if len(preemptedAllocs) == 0 {
iter.ctx.Metrics().ExhaustedNode(option.Node, dim)
continue
}
}
if len(allocsToPreempt) > 0 {
option.PreemptedAllocs = allocsToPreempt
}
// Score the fit normally otherwise
fitness := iter.scoreFit(option.Node, util)
normalizedFit := fitness / binPackingMaxFitScore
option.Scores = append(option.Scores, normalizedFit)
iter.ctx.Metrics().ScoreNode(option.Node, "binpack", normalizedFit)
// Score the device affinity
if totalDeviceAffinityWeight != 0 {
sumMatchingAffinities /= totalDeviceAffinityWeight
option.Scores = append(option.Scores, sumMatchingAffinities)
iter.ctx.Metrics().ScoreNode(option.Node, "devices", sumMatchingAffinities)
}
return option
}
}
func (iter *BinPackIterator) Reset() {
iter.source.Reset()
}
// JobAntiAffinityIterator is used to apply an anti-affinity to allocating
// along side other allocations from this job. This is used to help distribute
// load across the cluster.
type JobAntiAffinityIterator struct {
ctx Context
source RankIterator
jobID string
taskGroup string
desiredCount int
}
// NewJobAntiAffinityIterator is used to create a JobAntiAffinityIterator that
// applies the given penalty for co-placement with allocs from this job.
func NewJobAntiAffinityIterator(ctx Context, source RankIterator, jobID string) *JobAntiAffinityIterator {
iter := &JobAntiAffinityIterator{
ctx: ctx,
source: source,
jobID: jobID,
}
return iter
}
func (iter *JobAntiAffinityIterator) SetJob(job *structs.Job) {
iter.jobID = job.ID
}
func (iter *JobAntiAffinityIterator) SetTaskGroup(tg *structs.TaskGroup) {
iter.taskGroup = tg.Name
iter.desiredCount = tg.Count
}
func (iter *JobAntiAffinityIterator) Next() *RankedNode {
for {
option := iter.source.Next()
if option == nil {
return nil
}
// Get the proposed allocations
proposed, err := option.ProposedAllocs(iter.ctx)
if err != nil {
iter.ctx.Logger().Named("job_anti_affinity").Error("failed retrieving proposed allocations", "error", err)
continue
}
// Determine the number of collisions
collisions := 0
for _, alloc := range proposed {
if alloc.JobID == iter.jobID && alloc.TaskGroup == iter.taskGroup {
collisions += 1
}
}
// Calculate the penalty based on number of collisions
// TODO(preetha): Figure out if batch jobs need a different scoring penalty where collisions matter less
if collisions > 0 {
scorePenalty := -1 * float64(collisions+1) / float64(iter.desiredCount)
option.Scores = append(option.Scores, scorePenalty)
iter.ctx.Metrics().ScoreNode(option.Node, "job-anti-affinity", scorePenalty)
} else {
iter.ctx.Metrics().ScoreNode(option.Node, "job-anti-affinity", 0)
}
return option
}
}
func (iter *JobAntiAffinityIterator) Reset() {
iter.source.Reset()
}
// NodeReschedulingPenaltyIterator is used to apply a penalty to
// a node that had a previous failed allocation for the same job.
// This is used when attempting to reschedule a failed alloc
type NodeReschedulingPenaltyIterator struct {
ctx Context
source RankIterator
penaltyNodes map[string]struct{}
}
// NewNodeReschedulingPenaltyIterator is used to create a NodeReschedulingPenaltyIterator that
// applies the given scoring penalty for placement onto nodes in penaltyNodes
func NewNodeReschedulingPenaltyIterator(ctx Context, source RankIterator) *NodeReschedulingPenaltyIterator {
iter := &NodeReschedulingPenaltyIterator{
ctx: ctx,
source: source,
}
return iter
}
func (iter *NodeReschedulingPenaltyIterator) SetPenaltyNodes(penaltyNodes map[string]struct{}) {
iter.penaltyNodes = penaltyNodes
}
func (iter *NodeReschedulingPenaltyIterator) Next() *RankedNode {
option := iter.source.Next()
if option == nil {
return nil
}
_, ok := iter.penaltyNodes[option.Node.ID]
if ok {
option.Scores = append(option.Scores, -1)
iter.ctx.Metrics().ScoreNode(option.Node, "node-reschedule-penalty", -1)
} else {
iter.ctx.Metrics().ScoreNode(option.Node, "node-reschedule-penalty", 0)
}
return option
}
func (iter *NodeReschedulingPenaltyIterator) Reset() {
iter.penaltyNodes = make(map[string]struct{})
iter.source.Reset()
}
// NodeAffinityIterator is used to resolve any affinity rules in the job or task group,
// and apply a weighted score to nodes if they match.
type NodeAffinityIterator struct {
ctx Context
source RankIterator
jobAffinities []*structs.Affinity
affinities []*structs.Affinity
}
// NewNodeAffinityIterator is used to create a NodeAffinityIterator that
// applies a weighted score according to whether nodes match any
// affinities in the job or task group.
func NewNodeAffinityIterator(ctx Context, source RankIterator) *NodeAffinityIterator {
return &NodeAffinityIterator{
ctx: ctx,
source: source,
}
}
func (iter *NodeAffinityIterator) SetJob(job *structs.Job) {
iter.jobAffinities = job.Affinities
}
func (iter *NodeAffinityIterator) SetTaskGroup(tg *structs.TaskGroup) {
// Merge job affinities
if iter.jobAffinities != nil {
iter.affinities = append(iter.affinities, iter.jobAffinities...)
}
// Merge task group affinities and task affinities
if tg.Affinities != nil {
iter.affinities = append(iter.affinities, tg.Affinities...)
}
for _, task := range tg.Tasks {
if task.Affinities != nil {
iter.affinities = append(iter.affinities, task.Affinities...)
}
}
}
func (iter *NodeAffinityIterator) Reset() {
iter.source.Reset()
// This method is called between each task group, so only reset the merged list
iter.affinities = nil
}
func (iter *NodeAffinityIterator) hasAffinities() bool {
return len(iter.affinities) > 0
}
func (iter *NodeAffinityIterator) Next() *RankedNode {
option := iter.source.Next()
if option == nil {
return nil
}
if !iter.hasAffinities() {
iter.ctx.Metrics().ScoreNode(option.Node, "node-affinity", 0)
return option
}
// TODO(preetha): we should calculate normalized weights once and reuse it here
sumWeight := 0.0
for _, affinity := range iter.affinities {
sumWeight += math.Abs(float64(affinity.Weight))
}
totalAffinityScore := 0.0
for _, affinity := range iter.affinities {
if matchesAffinity(iter.ctx, affinity, option.Node) {
totalAffinityScore += float64(affinity.Weight)
}
}
normScore := totalAffinityScore / sumWeight
if totalAffinityScore != 0.0 {
option.Scores = append(option.Scores, normScore)
iter.ctx.Metrics().ScoreNode(option.Node, "node-affinity", normScore)
}
return option
}
func matchesAffinity(ctx Context, affinity *structs.Affinity, option *structs.Node) bool {
//TODO(preetha): Add a step here that filters based on computed node class for potential speedup
// Resolve the targets
lVal, lOk := resolveTarget(affinity.LTarget, option)
rVal, rOk := resolveTarget(affinity.RTarget, option)
// Check if satisfied
return checkAffinity(ctx, affinity.Operand, lVal, rVal, lOk, rOk)
}
// ScoreNormalizationIterator is used to combine scores from various prior
// iterators and combine them into one final score. The current implementation
// averages the scores together.
type ScoreNormalizationIterator struct {
ctx Context
source RankIterator
}
// NewScoreNormalizationIterator is used to create a ScoreNormalizationIterator that
// averages scores from various iterators into a final score.
func NewScoreNormalizationIterator(ctx Context, source RankIterator) *ScoreNormalizationIterator {
return &ScoreNormalizationIterator{
ctx: ctx,
source: source}
}
func (iter *ScoreNormalizationIterator) Reset() {
iter.source.Reset()
}
func (iter *ScoreNormalizationIterator) Next() *RankedNode {
option := iter.source.Next()
if option == nil || len(option.Scores) == 0 {
return option
}
numScorers := len(option.Scores)
sum := 0.0
for _, score := range option.Scores {
sum += score
}
option.FinalScore = sum / float64(numScorers)
//TODO(preetha): Turn map in allocmetrics into a heap of topK scores
iter.ctx.Metrics().ScoreNode(option.Node, "normalized-score", option.FinalScore)
return option
}
// PreemptionScoringIterator is used to score nodes according to the
// combination of preemptible allocations in them
type PreemptionScoringIterator struct {
ctx Context
source RankIterator
}
// NewPreemptionScoringIterator is used to create a score based on net
// aggregate priority of preempted allocations.
func NewPreemptionScoringIterator(ctx Context, source RankIterator) RankIterator {
return &PreemptionScoringIterator{
ctx: ctx,
source: source,
}
}
func (iter *PreemptionScoringIterator) Reset() {
iter.source.Reset()
}
func (iter *PreemptionScoringIterator) Next() *RankedNode {
option := iter.source.Next()
if option == nil || option.PreemptedAllocs == nil {
return option
}
netPriority := netPriority(option.PreemptedAllocs)
// preemption score is inversely proportional to netPriority
preemptionScore := preemptionScore(netPriority)
option.Scores = append(option.Scores, preemptionScore)
iter.ctx.Metrics().ScoreNode(option.Node, "preemption", preemptionScore)
return option
}
// netPriority is a scoring heuristic that represents a combination of two factors.
// First factor is the max priority in the set of allocations, with
// an additional factor that takes into account the individual priorities of allocations
func netPriority(allocs []*structs.Allocation) float64 {
sumPriority := 0
max := 0.0
for _, alloc := range allocs {
if float64(alloc.Job.Priority) > max {
max = float64(alloc.Job.Priority)
}
sumPriority += alloc.Job.Priority
}
// We use the maximum priority across all allocations
// with an additional penalty that increases proportional to the
// ratio of the sum by max
// This ensures that we penalize nodes that have a low max but a high
// number of preemptible allocations
ret := max + (float64(sumPriority) / max)
return ret
}
// preemptionScore is calculated using a logistic function
// see https://www.desmos.com/calculator/alaeiuaiey for a visual representation of the curve.
// Lower values of netPriority get a score closer to 1 and the inflection point is around 2048
// The score is modelled to be between 0 and 1 because its combined with other
// scoring factors like bin packing
func preemptionScore(netPriority float64) float64 {
// These values were chosen such that a net priority of 2048 would get a preemption score of 0.5
// rate is the decay parameter of the logistic function used in scoring preemption options
const rate = 0.0048
// origin controls the inflection point of the logistic function used in scoring preemption options
const origin = 2048.0
// This function manifests as an s curve that asympotically moves towards zero for large values of netPriority
return 1.0 / (1 + math.Exp(rate*(netPriority-origin)))
}