-
Notifications
You must be signed in to change notification settings - Fork 134
/
reports.go
405 lines (378 loc) · 17.3 KB
/
reports.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
package scheduler
import (
"context"
"fmt"
"strings"
"sync"
"sync/atomic"
"text/tabwriter"
"github.com/gogo/protobuf/types"
lru "github.com/hashicorp/golang-lru"
"github.com/oklog/ulid"
"github.com/openconfig/goyang/pkg/indent"
"github.com/pkg/errors"
"golang.org/x/exp/maps"
"golang.org/x/exp/slices"
"github.com/armadaproject/armada/internal/common/armadaerrors"
schedulercontext "github.com/armadaproject/armada/internal/scheduler/context"
"github.com/armadaproject/armada/internal/scheduler/schedulerobjects"
)
// SchedulingContextRepository stores scheduling contexts associated with recent scheduling attempts.
// On adding a context, a map is cloned, then mutated, and then swapped for the previous map using atomic pointers.
// Hence, reads concurrent with writes are safe and don't need locking.
// A mutex protects against concurrent writes.
type SchedulingContextRepository struct {
// Maps executor id to *schedulercontext.SchedulingContext.
// The most recent attempt.
mostRecentSchedulingContextByExecutorP atomic.Pointer[SchedulingContextByExecutor]
// The most recent attempt where a non-zero amount of resources were scheduled.
mostRecentSuccessfulSchedulingContextByExecutorP atomic.Pointer[SchedulingContextByExecutor]
// Maps queue name to QueueSchedulingContextByExecutor.
// The most recent attempt.
mostRecentQueueSchedulingContextByExecutorByQueueP atomic.Pointer[map[string]QueueSchedulingContextByExecutor]
// The most recent attempt where a non-zero amount of resources were scheduled.
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueueP atomic.Pointer[map[string]QueueSchedulingContextByExecutor]
// Maps job id to JobSchedulingContextByExecutor.
// We limit the number of job contexts to store to control memory usage.
mostRecentJobSchedulingContextByExecutorByJobId *lru.Cache
// Store all executor ids seen so far in a set.
// Used to ensure all executors are included in reports.
executorIds map[string]bool
// All executors in sorted order.
sortedExecutorIdsP atomic.Pointer[[]string]
// Protects the fields in this struct from concurrent and dirty writes.
mu sync.Mutex
}
type (
SchedulingContextByExecutor map[string]*schedulercontext.SchedulingContext
QueueSchedulingContextByExecutor map[string]*schedulercontext.QueueSchedulingContext
JobSchedulingContextByExecutor map[string]*schedulercontext.JobSchedulingContext
)
func NewSchedulingContextRepository(maxJobSchedulingContextsPerExecutor uint) (*SchedulingContextRepository, error) {
jobSchedulingContextByExecutorByJobId, err := lru.New(int(maxJobSchedulingContextsPerExecutor))
if err != nil {
return nil, err
}
rv := &SchedulingContextRepository{
mostRecentJobSchedulingContextByExecutorByJobId: jobSchedulingContextByExecutorByJobId,
executorIds: make(map[string]bool),
}
mostRecentSchedulingContextByExecutor := make(SchedulingContextByExecutor)
mostRecentSuccessfulSchedulingContextByExecutor := make(SchedulingContextByExecutor)
mostRecentQueueSchedulingContextByExecutorByQueue := make(map[string]QueueSchedulingContextByExecutor)
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue := make(map[string]QueueSchedulingContextByExecutor)
sortedExecutorIds := make([]string, 0)
rv.mostRecentSchedulingContextByExecutorP.Store(&mostRecentSchedulingContextByExecutor)
rv.mostRecentSuccessfulSchedulingContextByExecutorP.Store(&mostRecentSuccessfulSchedulingContextByExecutor)
rv.mostRecentQueueSchedulingContextByExecutorByQueueP.Store(&mostRecentQueueSchedulingContextByExecutorByQueue)
rv.mostRecentSuccessfulQueueSchedulingContextByExecutorByQueueP.Store(&mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue)
rv.sortedExecutorIdsP.Store(&sortedExecutorIds)
return rv, nil
}
// AddSchedulingContext adds a scheduling context to the repo.
// It also extracts the queue and job scheduling contexts it contains and stores those separately.
//
// It's safe to call this method concurrently with itself and with methods getting contexts from the repo.
// It's not safe to mutate contexts once they've been provided to this method.
//
// Job contexts are stored first, then queue contexts, and finally the scheduling context itself.
// This avoids having a stored scheduling (queue) context referring to a queue (job) context that isn't stored yet.
func (repo *SchedulingContextRepository) AddSchedulingContext(sctx *schedulercontext.SchedulingContext) error {
queueSchedulingContextByQueue, jobSchedulingContextByJobId := extractQueueAndJobContexts(sctx)
repo.mu.Lock()
defer repo.mu.Unlock()
for _, jctx := range jobSchedulingContextByJobId {
if err := repo.addJobSchedulingContext(jctx); err != nil {
return err
}
}
if err := repo.addQueueSchedulingContexts(maps.Values(queueSchedulingContextByQueue)); err != nil {
return err
}
if err := repo.addSchedulingContext(sctx); err != nil {
return err
}
if err := repo.addExecutorId(sctx.ExecutorId); err != nil {
return err
}
return nil
}
// Should only be called from AddSchedulingContext to avoid concurrent and/or dirty writes.
func (repo *SchedulingContextRepository) addExecutorId(executorId string) error {
n := len(repo.executorIds)
repo.executorIds[executorId] = true
if len(repo.executorIds) != n {
sortedExecutorIds := maps.Keys(repo.executorIds)
slices.Sort(sortedExecutorIds)
repo.sortedExecutorIdsP.Store(&sortedExecutorIds)
}
return nil
}
// Should only be called from AddSchedulingContext to avoid dirty writes.
func (repo *SchedulingContextRepository) addSchedulingContext(sctx *schedulercontext.SchedulingContext) error {
mostRecentSchedulingContextByExecutor := *repo.mostRecentSchedulingContextByExecutorP.Load()
mostRecentSchedulingContextByExecutor = maps.Clone(mostRecentSchedulingContextByExecutor)
mostRecentSuccessfulSchedulingContextByExecutor := *repo.mostRecentSuccessfulSchedulingContextByExecutorP.Load()
mostRecentSuccessfulSchedulingContextByExecutor = maps.Clone(mostRecentSuccessfulSchedulingContextByExecutor)
mostRecentSchedulingContextByExecutor[sctx.ExecutorId] = sctx
if !sctx.ScheduledResourcesByPriority.IsZero() {
mostRecentSuccessfulSchedulingContextByExecutor[sctx.ExecutorId] = sctx
}
repo.mostRecentSchedulingContextByExecutorP.Store(&mostRecentSchedulingContextByExecutor)
repo.mostRecentSuccessfulSchedulingContextByExecutorP.Store(&mostRecentSuccessfulSchedulingContextByExecutor)
return nil
}
// Should only be called from AddSchedulingContext to avoid dirty writes.
func (repo *SchedulingContextRepository) addQueueSchedulingContexts(qctxs []*schedulercontext.QueueSchedulingContext) error {
mostRecentQueueSchedulingContextByExecutorByQueue := *repo.mostRecentQueueSchedulingContextByExecutorByQueueP.Load()
mostRecentQueueSchedulingContextByExecutorByQueue = maps.Clone(mostRecentQueueSchedulingContextByExecutorByQueue)
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue := *repo.mostRecentSuccessfulQueueSchedulingContextByExecutorByQueueP.Load()
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue = maps.Clone(mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue)
for _, qctx := range qctxs {
if qctx.ExecutorId == "" {
return errors.WithStack(&armadaerrors.ErrInvalidArgument{
Name: "ExecutorId",
Value: "",
Message: "received empty executorId",
})
}
if qctx.Queue == "" {
return errors.WithStack(&armadaerrors.ErrInvalidArgument{
Name: "Queue",
Value: "",
Message: "received empty queue name",
})
}
if previous := mostRecentQueueSchedulingContextByExecutorByQueue[qctx.Queue]; previous != nil {
previous = maps.Clone(previous)
previous[qctx.ExecutorId] = qctx
mostRecentQueueSchedulingContextByExecutorByQueue[qctx.Queue] = previous
} else {
mostRecentQueueSchedulingContextByExecutorByQueue[qctx.Queue] = QueueSchedulingContextByExecutor{
qctx.ExecutorId: qctx,
}
}
if !qctx.ScheduledResourcesByPriority.IsZero() {
if previous := mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue[qctx.Queue]; previous != nil {
previous = maps.Clone(previous)
previous[qctx.ExecutorId] = qctx
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue[qctx.Queue] = previous
} else {
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue[qctx.Queue] = QueueSchedulingContextByExecutor{
qctx.ExecutorId: qctx,
}
}
}
}
repo.mostRecentQueueSchedulingContextByExecutorByQueueP.Store(&mostRecentQueueSchedulingContextByExecutorByQueue)
repo.mostRecentSuccessfulQueueSchedulingContextByExecutorByQueueP.Store(&mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue)
return nil
}
// Should only be called from AddSchedulingContext to avoid dirty writes.
func (repo *SchedulingContextRepository) addJobSchedulingContext(jctx *schedulercontext.JobSchedulingContext) error {
if jctx.ExecutorId == "" {
return errors.WithStack(&armadaerrors.ErrInvalidArgument{
Name: "ExecutorId",
Value: "",
Message: "received empty executorId",
})
}
if jctx.JobId == "" {
return errors.WithStack(&armadaerrors.ErrInvalidArgument{
Name: "JobId",
Value: "",
Message: "received empty jobId",
})
}
previous, ok, _ := repo.mostRecentJobSchedulingContextByExecutorByJobId.PeekOrAdd(
jctx.JobId,
JobSchedulingContextByExecutor{jctx.ExecutorId: jctx},
)
if ok {
jobSchedulingContextByExecutor := previous.(JobSchedulingContextByExecutor)
jobSchedulingContextByExecutor[jctx.ExecutorId] = jctx
repo.mostRecentJobSchedulingContextByExecutorByJobId.Add(jctx.JobId, jobSchedulingContextByExecutor)
}
return nil
}
// extractQueueAndJobContexts extracts the job and queue scheduling contexts from the scheduling context,
// and returns those separately.
func extractQueueAndJobContexts(sctx *schedulercontext.SchedulingContext) (map[string]*schedulercontext.QueueSchedulingContext, map[string]*schedulercontext.JobSchedulingContext) {
queueSchedulingContextByQueue := make(map[string]*schedulercontext.QueueSchedulingContext)
jobSchedulingContextByJobId := make(map[string]*schedulercontext.JobSchedulingContext)
for queue, qctx := range sctx.QueueSchedulingContexts {
for jobId, jctx := range qctx.SuccessfulJobSchedulingContexts {
jobSchedulingContextByJobId[jobId] = jctx
}
for jobId, jctx := range qctx.UnsuccessfulJobSchedulingContexts {
jobSchedulingContextByJobId[jobId] = jctx
}
queueSchedulingContextByQueue[queue] = qctx
}
return queueSchedulingContextByQueue, jobSchedulingContextByJobId
}
// GetSchedulingReport is a gRPC endpoint for querying scheduler reports.
// TODO: Further separate this from internal contexts.
func (repo *SchedulingContextRepository) GetSchedulingReport(_ context.Context, _ *types.Empty) (*schedulerobjects.SchedulingReport, error) {
return &schedulerobjects.SchedulingReport{
Report: repo.getSchedulingReportString(),
}, nil
}
func (repo *SchedulingContextRepository) getSchedulingReportString() string {
sortedExecutorIds := repo.GetSortedExecutorIds()
mostRecentSchedulingContextByExecutor := repo.GetMostRecentSchedulingContextByExecutor()
mostRecentSuccessfulSchedulingContextByExecutor := repo.GetMostRecentSuccessfulSchedulingContextByExecutor()
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
for _, executorId := range sortedExecutorIds {
fmt.Fprintf(w, "%s:\n", executorId)
sctx := mostRecentSchedulingContextByExecutor[executorId]
if sctx != nil {
fmt.Fprint(w, indent.String("\t", "Most recent attempt:\n"))
fmt.Fprint(w, indent.String("\t\t", sctx.String()))
} else {
fmt.Fprint(w, indent.String("\t", "Most recent attempt: none\n"))
}
sctx = mostRecentSuccessfulSchedulingContextByExecutor[executorId]
if sctx != nil {
fmt.Fprint(w, indent.String("\t", "Most recent successful attempt:\n"))
fmt.Fprint(w, indent.String("\t\t", sctx.String()))
} else {
fmt.Fprint(w, indent.String("\t", "Most recent successful attempt: none\n"))
}
}
w.Flush()
return sb.String()
}
// GetQueueReport is a gRPC endpoint for querying queue reports.
// TODO: Further separate this from internal contexts.
func (repo *SchedulingContextRepository) GetQueueReport(_ context.Context, queue *schedulerobjects.Queue) (*schedulerobjects.QueueReport, error) {
queueName := strings.TrimSpace(queue.Name)
return &schedulerobjects.QueueReport{
Report: repo.getQueueReportString(queueName),
}, nil
}
func (repo *SchedulingContextRepository) getQueueReportString(queue string) string {
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
sortedExecutorIds := repo.GetSortedExecutorIds()
mostRecentQueueSchedulingContextByExecutor, _ := repo.GetMostRecentQueueSchedulingContextByExecutor(queue)
mostRecentSuccessfulQueueSchedulingContextByExecutor, _ := repo.GetMostRecentSuccessfulQueueSchedulingContextByExecutor(queue)
for _, executorId := range sortedExecutorIds {
fmt.Fprintf(w, "%s:\n", executorId)
qctx := mostRecentQueueSchedulingContextByExecutor[executorId]
if qctx != nil {
fmt.Fprint(w, indent.String("\t", "Most recent attempt:\n"))
fmt.Fprint(w, indent.String("\t\t", qctx.String()))
} else {
fmt.Fprint(w, indent.String("\t", "Most recent attempt: none\n"))
}
qctx = mostRecentSuccessfulQueueSchedulingContextByExecutor[executorId]
if qctx != nil {
fmt.Fprint(w, indent.String("\t", "Most recent successful attempt:\n"))
fmt.Fprint(w, indent.String("\t\t", qctx.String()))
} else {
fmt.Fprint(w, indent.String("\t", "Most recent successful attempt: none\n"))
}
}
w.Flush()
return sb.String()
}
// GetJobReport is a gRPC endpoint for querying job reports.
// TODO: Further separate this from internal contexts.
func (repo *SchedulingContextRepository) GetJobReport(_ context.Context, jobId *schedulerobjects.JobId) (*schedulerobjects.JobReport, error) {
key := strings.TrimSpace(jobId.Id)
if _, err := ulid.Parse(key); err != nil {
return nil, &armadaerrors.ErrInvalidArgument{
Name: "jobId",
Value: jobId.Id,
Message: fmt.Sprintf("%s is not a valid jobId", jobId.Id),
}
}
return &schedulerobjects.JobReport{
Report: repo.getJobReportString(key),
}, nil
}
func (repo *SchedulingContextRepository) getJobReportString(jobId string) string {
sortedExecutorIds := repo.GetSortedExecutorIds()
jobSchedulingContextByExecutor, _ := repo.GetMostRecentJobSchedulingContextByExecutor(jobId)
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
for _, executorId := range sortedExecutorIds {
jctx := jobSchedulingContextByExecutor[executorId]
if jctx != nil {
fmt.Fprintf(w, "%s:\n", executorId)
fmt.Fprint(w, indent.String("\t", jctx.String()))
} else {
fmt.Fprintf(w, "%s: no recent attempt\n", executorId)
}
}
w.Flush()
return sb.String()
}
func (repo *SchedulingContextRepository) GetMostRecentSchedulingContextByExecutor() SchedulingContextByExecutor {
return *repo.mostRecentSchedulingContextByExecutorP.Load()
}
func (repo *SchedulingContextRepository) GetMostRecentSuccessfulSchedulingContextByExecutor() SchedulingContextByExecutor {
return *repo.mostRecentSuccessfulSchedulingContextByExecutorP.Load()
}
func (repo *SchedulingContextRepository) GetMostRecentQueueSchedulingContextByExecutor(queue string) (QueueSchedulingContextByExecutor, bool) {
mostRecentQueueSchedulingContextByExecutorByQueue := *repo.mostRecentQueueSchedulingContextByExecutorByQueueP.Load()
mostRecentQueueSchedulingContextByExecutor, ok := mostRecentQueueSchedulingContextByExecutorByQueue[queue]
return mostRecentQueueSchedulingContextByExecutor, ok
}
func (repo *SchedulingContextRepository) GetMostRecentSuccessfulQueueSchedulingContextByExecutor(queue string) (QueueSchedulingContextByExecutor, bool) {
mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue := *repo.mostRecentSuccessfulQueueSchedulingContextByExecutorByQueueP.Load()
mostRecentSuccessfulQueueSchedulingContextByExecutor, ok := mostRecentSuccessfulQueueSchedulingContextByExecutorByQueue[queue]
return mostRecentSuccessfulQueueSchedulingContextByExecutor, ok
}
func (repo *SchedulingContextRepository) GetMostRecentJobSchedulingContextByExecutor(jobId string) (JobSchedulingContextByExecutor, bool) {
if v, ok := repo.mostRecentJobSchedulingContextByExecutorByJobId.Get(jobId); ok {
jobSchedulingContextByExecutor := v.(JobSchedulingContextByExecutor)
return jobSchedulingContextByExecutor, true
} else {
return nil, false
}
}
func (repo *SchedulingContextRepository) GetSortedExecutorIds() []string {
return *repo.sortedExecutorIdsP.Load()
}
func (m SchedulingContextByExecutor) String() string {
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
executorIds := maps.Keys(m)
slices.Sort(executorIds)
for _, executorId := range executorIds {
sctx := m[executorId]
fmt.Fprintf(w, "%s:\n", executorId)
fmt.Fprint(w, indent.String("\t", sctx.String()))
}
w.Flush()
return sb.String()
}
func (m QueueSchedulingContextByExecutor) String() string {
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
executorIds := maps.Keys(m)
slices.Sort(executorIds)
for _, executorId := range executorIds {
qctx := m[executorId]
fmt.Fprintf(w, "%s:\n", executorId)
fmt.Fprint(w, indent.String("\t", qctx.String()))
}
w.Flush()
return sb.String()
}
func (m JobSchedulingContextByExecutor) String() string {
var sb strings.Builder
w := tabwriter.NewWriter(&sb, 1, 1, 1, ' ', 0)
executorIds := maps.Keys(m)
slices.Sort(executorIds)
for _, executorId := range executorIds {
jctx := m[executorId]
fmt.Fprintf(w, "%s:\n", executorId)
fmt.Fprint(w, indent.String("\t", jctx.String()))
}
w.Flush()
return sb.String()
}