-
Notifications
You must be signed in to change notification settings - Fork 126
/
nodeiteration.go
458 lines (418 loc) · 12.4 KB
/
nodeiteration.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
package nodedb
import (
"bytes"
"container/heap"
"github.com/hashicorp/go-memdb"
"github.com/pkg/errors"
log "github.com/sirupsen/logrus"
"golang.org/x/exp/slices"
"k8s.io/apimachinery/pkg/api/resource"
"github.com/armadaproject/armada/internal/scheduler/internaltypes"
)
type NodeIterator interface {
NextNode() *internaltypes.Node
}
// NodesIterator is an iterator over all nodes in the db.
type NodesIterator struct {
it memdb.ResultIterator
}
func NewNodesIterator(txn *memdb.Txn) (*NodesIterator, error) {
it, err := txn.LowerBound("nodes", "id", "")
if err != nil {
return nil, errors.WithStack(err)
}
return &NodesIterator{
it: it,
}, nil
}
func (it *NodesIterator) WatchCh() <-chan struct{} {
panic("not implemented")
}
func (it *NodesIterator) NextNode() *internaltypes.Node {
obj := it.it.Next()
if obj == nil {
return nil
}
return obj.(*internaltypes.Node)
}
func (it *NodesIterator) Next() interface{} {
return it.NextNode()
}
type NodePairIterator struct {
itA *NodesIterator
itB *NodesIterator
nodeA *internaltypes.Node
nodeB *internaltypes.Node
}
type NodePairIteratorItem struct {
NodeA *internaltypes.Node
NodeB *internaltypes.Node
}
func NewNodePairIterator(txnA, txnB *memdb.Txn) (*NodePairIterator, error) {
itA, err := NewNodesIterator(txnA)
if err != nil {
return nil, errors.WithStack(err)
}
itB, err := NewNodesIterator(txnB)
if err != nil {
return nil, errors.WithStack(err)
}
return &NodePairIterator{
itA: itA,
itB: itB,
}, nil
}
func (it *NodePairIterator) WatchCh() <-chan struct{} {
panic("not implemented")
}
func (it *NodePairIterator) NextItem() (rv *NodePairIteratorItem) {
defer func() {
if rv == nil {
return
}
if rv.NodeA != nil {
it.nodeA = nil
}
if rv.NodeB != nil {
it.nodeB = nil
}
}()
if it.nodeA == nil {
it.nodeA = it.itA.NextNode()
}
if it.nodeB == nil {
it.nodeB = it.itB.NextNode()
}
if it.nodeA == nil && it.nodeB == nil {
return nil
} else if it.nodeA == nil || it.nodeB == nil {
return &NodePairIteratorItem{
NodeA: it.nodeA,
NodeB: it.nodeB,
}
}
cmp := bytes.Compare([]byte(it.nodeA.GetId()), []byte(it.nodeB.GetId()))
if cmp == 0 {
return &NodePairIteratorItem{
NodeA: it.nodeA,
NodeB: it.nodeB,
}
} else if cmp == -1 {
return &NodePairIteratorItem{
NodeA: it.nodeA,
}
} else {
return &NodePairIteratorItem{
NodeB: it.nodeB,
}
}
}
func (it *NodePairIterator) Next() interface{} {
return it.NextItem()
}
// NodeIndex is an index for internaltypes.Node that returns node.NodeDbKeys[KeyIndex].
type NodeIndex struct {
KeyIndex int
}
// FromArgs computes the index key from a set of arguments.
// Takes a single argument resourceAmount of type []byte.
func (index *NodeIndex) FromArgs(args ...interface{}) ([]byte, error) {
if len(args) != 1 {
return nil, errors.New("must provide exactly one argument")
}
return args[0].([]byte), nil
}
// FromObject extracts the index key from a *Node.
func (index *NodeIndex) FromObject(raw interface{}) (bool, []byte, error) {
node := raw.(*internaltypes.Node)
return true, node.Keys[index.KeyIndex], nil
}
// NodeTypesIterator is an iterator over all nodes of the given nodeTypes
// with at least some specified amount of resources allocatable at a given priority.
// For example, all nodes of nodeType "foo" and "bar" with at least 2 cores and 1Gi memory allocatable at priority 2.
// Nodes are returned in sorted order, from least to most of the specified resource available.
type NodeTypesIterator struct {
pq *nodeTypesIteratorPQ
}
func NewNodeTypesIterator(
txn *memdb.Txn,
nodeTypeIds []uint64,
indexName string,
priority int32,
keyIndex int,
indexedResources []string,
indexedResourceRequests []resource.Quantity,
indexedResourceResolutionMillis []int64,
) (*NodeTypesIterator, error) {
pq := &nodeTypesIteratorPQ{
priority: priority,
indexedResources: indexedResources,
items: make([]*nodeTypesIteratorPQItem, 0, len(nodeTypeIds)),
}
for _, nodeTypeId := range nodeTypeIds {
it, err := NewNodeTypeIterator(
txn,
nodeTypeId,
indexName,
priority,
keyIndex,
indexedResources,
indexedResourceRequests,
indexedResourceResolutionMillis,
)
if err != nil {
return nil, err
}
node, err := it.NextNode()
if err != nil {
return nil, err
}
if node == nil {
continue
}
heap.Push(pq, &nodeTypesIteratorPQItem{
node: node,
it: it,
})
}
return &NodeTypesIterator{pq: pq}, nil
}
func (it *NodeTypesIterator) WatchCh() <-chan struct{} {
panic("not implemented")
}
func (it *NodeTypesIterator) Next() interface{} {
v, err := it.NextNode()
if err != nil {
panic(err)
}
return v
}
func (it *NodeTypesIterator) NextNode() (*internaltypes.Node, error) {
if it.pq.Len() == 0 {
return nil, nil
}
pqItem := heap.Pop(it.pq).(*nodeTypesIteratorPQItem)
node := pqItem.node
nextNode, err := pqItem.it.NextNode()
if err != nil {
return nil, err
}
if nextNode != nil {
pqItem.node = nextNode
heap.Push(it.pq, pqItem)
}
return node, nil
}
type nodeTypesIteratorPQ struct {
priority int32
indexedResources []string
items []*nodeTypesIteratorPQItem
}
type nodeTypesIteratorPQItem struct {
node *internaltypes.Node
it *NodeTypeIterator
// The index of the item in the heap. Maintained by the heap.Interface methods.
index int
}
func (pq *nodeTypesIteratorPQ) Len() int { return len(pq.items) }
func (pq *nodeTypesIteratorPQ) Less(i, j int) bool {
return pq.less(pq.items[i].node, pq.items[j].node)
}
func (it *nodeTypesIteratorPQ) less(a, b *internaltypes.Node) bool {
allocatableByPriorityA := a.AllocatableByPriority[it.priority]
allocatableByPriorityB := b.AllocatableByPriority[it.priority]
for _, t := range it.indexedResources {
qa := allocatableByPriorityA.Get(t)
qb := allocatableByPriorityB.Get(t)
if cmp := qa.Cmp(qb); cmp == -1 {
return true
} else if cmp == 1 {
return false
}
}
// Tie-break by id.
return a.GetId() < b.GetId()
}
func (pq *nodeTypesIteratorPQ) Swap(i, j int) {
pq.items[i], pq.items[j] = pq.items[j], pq.items[i]
pq.items[i].index = i
pq.items[j].index = j
}
func (pq *nodeTypesIteratorPQ) Push(x any) {
n := len(pq.items)
item := x.(*nodeTypesIteratorPQItem)
item.index = n
pq.items = append(pq.items, item)
}
func (pq *nodeTypesIteratorPQ) Pop() any {
old := pq.items
n := len(old)
item := old[n-1]
old[n-1] = nil // avoid memory leak
item.index = -1 // for safety
pq.items = old[0 : n-1]
return item
}
// NodeTypeIterator is an iterator over all nodes of a given nodeType
// with at least some specified amount of resources allocatable at a given priority.
// For example, all nodes of nodeType "foo" with at least 2 cores and 1Gi memory allocatable at priority 2.
// Nodes are returned in sorted order, from least to most of the specified resource available.
type NodeTypeIterator struct {
txn *memdb.Txn
// Only yield nodes of this nodeType.
nodeTypeId uint64
// Priority at which to consider allocatable resources on the node.
priority int32
// Used to index into node.keys to assert that keys are always increasing.
// This to detect if the iterator gets stuck.
// TODO(albin): With better testing we should be able to remove this.
keyIndex int
// Name of the memdb index used for node iteration.
// Should correspond to the priority set for this iterator.
indexName string
// NodeDb indexed resources.
indexedResources []string
// Pod requests for indexed resources in the same order as indexedResources.
indexedResourceRequests []resource.Quantity
// The resolution with which indexed resources are tracked. In the same order as indexedResources.
indexedResourceResolutionMillis []int64
// Current lower bound on node allocatable resources looked for.
// Updated in-place as the iterator makes progress.
lowerBound []resource.Quantity
// Tentative lower-bound.
newLowerBound []resource.Quantity
// memdb key computed from nodeTypeId and lowerBound.
// Stored here to avoid dynamic allocs.
key []byte
// Key for newLowerBound.
newKey []byte
// Current iterator into the underlying memdb.
// Updated in-place whenever lowerBound changes.
memdbIterator memdb.ResultIterator
// Used to detect if the iterator gets stuck in a loop.
previousKey []byte
previousNode *internaltypes.Node
}
func NewNodeTypeIterator(
txn *memdb.Txn,
nodeTypeId uint64,
indexName string,
priority int32,
keyIndex int,
indexedResources []string,
indexedResourceRequests []resource.Quantity,
indexedResourceResolutionMillis []int64,
) (*NodeTypeIterator, error) {
if len(indexedResources) != len(indexedResourceRequests) {
return nil, errors.Errorf("indexedResources and resourceRequirements are not of equal length")
}
if len(indexedResources) != len(indexedResourceResolutionMillis) {
return nil, errors.Errorf("indexedResources and indexedResourceResolutionMillis are not of equal length")
}
if keyIndex < 0 {
return nil, errors.Errorf("keyIndex is negative: %d", keyIndex)
}
it := &NodeTypeIterator{
txn: txn,
nodeTypeId: nodeTypeId,
priority: priority,
keyIndex: keyIndex,
indexName: indexName,
indexedResources: indexedResources,
indexedResourceRequests: indexedResourceRequests,
indexedResourceResolutionMillis: indexedResourceResolutionMillis,
lowerBound: slices.Clone(indexedResourceRequests),
newLowerBound: slices.Clone(indexedResourceRequests),
}
memdbIt, err := it.newNodeTypeIterator()
if err != nil {
return nil, err
}
it.memdbIterator = memdbIt
return it, nil
}
func (it *NodeTypeIterator) newNodeTypeIterator() (memdb.ResultIterator, error) {
// TODO(albin): We're re-computing the key unnecessarily here.
it.key = NodeIndexKey(it.key[0:0], it.nodeTypeId, it.lowerBound)
memdbIt, err := it.txn.LowerBound(
"nodes",
it.indexName,
it.key,
)
if err != nil {
return nil, errors.WithStack(err)
}
return memdbIt, nil
}
func (it *NodeTypeIterator) WatchCh() <-chan struct{} {
panic("not implemented")
}
func (it *NodeTypeIterator) Next() interface{} {
v, err := it.NextNode()
if err != nil {
panic(err)
}
return v
}
func (it *NodeTypeIterator) NextNode() (*internaltypes.Node, error) {
for {
v := it.memdbIterator.Next()
if v == nil {
return nil, nil
}
node := v.(*internaltypes.Node)
if it.keyIndex >= len(node.Keys) {
return nil, errors.Errorf("keyIndex is %d, but node %s has only %d keys", it.keyIndex, node.GetId(), len(node.Keys))
}
nodeKey := node.Keys[it.keyIndex]
if it.previousKey != nil && bytes.Compare(it.previousKey, nodeKey) != -1 {
return nil, errors.Errorf(
"iteration loop detected: key %x of node %#v is not greater than key %x of node %#v",
nodeKey, node, it.previousKey, it.previousNode,
)
}
it.previousKey = nodeKey
it.previousNode = node
if node.GetNodeTypeId() != it.nodeTypeId {
// There are no more nodes of this nodeType.
return nil, nil
}
allocatableByPriority := node.AllocatableByPriority[it.priority]
if len(allocatableByPriority.Resources) == 0 {
return nil, errors.Errorf("node %s has no resources registered at priority %d: %v", node.GetId(), it.priority, node.AllocatableByPriority)
}
for i, t := range it.indexedResources {
nodeQuantity := allocatableByPriority.Get(t).DeepCopy()
requestQuantity := it.indexedResourceRequests[i].DeepCopy()
it.newLowerBound[i] = roundQuantityToResolution(nodeQuantity, it.indexedResourceResolutionMillis[i])
// If nodeQuantity < requestQuantity, replace the iterator using the lowerBound.
// If nodeQuantity >= requestQuantity for all resources, return the node.
if nodeQuantity.Cmp(requestQuantity) == -1 {
for j := i; j < len(it.indexedResources); j++ {
it.newLowerBound[j] = it.indexedResourceRequests[j]
}
it.newKey = NodeIndexKey(it.newKey[0:0], it.nodeTypeId, it.newLowerBound)
if bytes.Compare(it.key, it.newKey) == -1 {
// TODO(albin): Temporary workaround. Shouldn't be necessary.
lowerBound := it.lowerBound
it.lowerBound = it.newLowerBound
it.newLowerBound = lowerBound
} else {
log.Warnf(
"new lower-bound %x is not greater than current bound %x",
it.newKey, it.key,
)
break
}
memdbIterator, err := it.newNodeTypeIterator()
if err != nil {
return nil, err
}
it.memdbIterator = memdbIterator
break
} else if i == len(it.indexedResources)-1 {
return node, nil
}
}
}
}