/
poolcache.go
463 lines (430 loc) · 15.1 KB
/
poolcache.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
/*
Copyright 2016 The Fission 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 fscache
import (
"bufio"
"context"
"errors"
"fmt"
"io"
"go.uber.org/zap"
"k8s.io/apimachinery/pkg/api/resource"
"k8s.io/apimachinery/pkg/types"
"github.com/fission/fission/pkg/crd"
ferror "github.com/fission/fission/pkg/error"
otelUtils "github.com/fission/fission/pkg/utils/otel"
)
type requestType int
const (
getValue requestType = iota
listAvailableValue
setValue
markAvailable
deleteValue
setCPUUtilization
markSpecializationFailure
logFuncSvc
markDeleted
)
type (
funcSvcInfo struct {
val *FuncSvc
activeRequests int // number of requests served by function pod
currentCPUUsage resource.Quantity // current cpu usage of the specialized function pod
cpuLimit resource.Quantity // if currentCPUUsage is more than cpuLimit cache miss occurs in getValue request
}
funcSvcGroup struct {
svcWaiting int
svcRetain int
svcs map[string]*funcSvcInfo
queue *Queue
deleted bool
}
// PoolCache implements a simple cache implementation having values mapped by two keys [function][address].
// As of now PoolCache is only used by poolmanager executor
PoolCache struct {
cache map[crd.CacheKeyURG]*funcSvcGroup
requestChannel chan *request
logger *zap.Logger
}
request struct {
requestType
ctx context.Context
function crd.CacheKeyURG
address string
dumpWriter io.Writer
value *FuncSvc
requestsPerPod int
cpuUsage resource.Quantity
responseChannel chan *response
concurrency int
svcsRetain int
}
response struct {
error
allValues []*FuncSvc
value *FuncSvc
svcWaitValue *svcWait
}
svcWait struct {
svcChannel chan *FuncSvc
ctx context.Context
}
)
// NewPoolCache create a Cache object
func NewPoolCache(logger *zap.Logger) *PoolCache {
c := &PoolCache{
cache: make(map[crd.CacheKeyURG]*funcSvcGroup),
requestChannel: make(chan *request),
logger: logger,
}
go c.service()
return c
}
func NewFuncSvcGroup() *funcSvcGroup {
return &funcSvcGroup{
svcs: make(map[string]*funcSvcInfo),
queue: NewQueue(),
}
}
func (c *PoolCache) service() {
for {
req := <-c.requestChannel
resp := &response{}
switch req.requestType {
case getValue:
funcSvcGroup, ok := c.cache[req.function]
if !ok {
// first request for this function, create a new group
c.cache[req.function] = NewFuncSvcGroup()
c.cache[req.function].svcWaiting++
resp.error = ferror.MakeError(ferror.ErrorNotFound,
fmt.Sprintf("function Name '%s' not found", req.function))
req.responseChannel <- resp
continue
}
found := false
totalActiveRequests := 0
// check if any specialized pod is available
for addr := range funcSvcGroup.svcs {
totalActiveRequests += funcSvcGroup.svcs[addr].activeRequests
if funcSvcGroup.svcs[addr].activeRequests < req.requestsPerPod &&
funcSvcGroup.svcs[addr].currentCPUUsage.Cmp(funcSvcGroup.svcs[addr].cpuLimit) < 1 {
// mark active
funcSvcGroup.svcs[addr].activeRequests++
if c.logger.Core().Enabled(zap.DebugLevel) {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Debug("Increase active requests with getValue", zap.String("function", req.function.String()), zap.String("address", addr), zap.Int("activeRequests", funcSvcGroup.svcs[addr].activeRequests))
}
resp.value = funcSvcGroup.svcs[addr].val
found = true
break
}
}
// if specialized pod is available then return svc
if found {
req.responseChannel <- resp
continue
}
concurrencyUsed := len(funcSvcGroup.svcs) + (funcSvcGroup.svcWaiting - funcSvcGroup.queue.Len())
// if concurrency is available then be aggressive and use it as we are not sure if specialization will complete for other requests
if req.concurrency > 0 && concurrencyUsed < req.concurrency {
funcSvcGroup.svcWaiting++
resp.error = ferror.MakeError(ferror.ErrorNotFound, fmt.Sprintf("function '%s' not found", req.function))
req.responseChannel <- resp
continue
}
// if no concurrency is available then check if there is any virtual capacity in the existing pods to serve the request in future
// if specialization doesnt complete within request then request will be timeout
capacity := (concurrencyUsed * req.requestsPerPod) - (totalActiveRequests + funcSvcGroup.svcWaiting)
if capacity > 0 {
funcSvcGroup.svcWaiting++
svcWait := &svcWait{
svcChannel: make(chan *FuncSvc),
ctx: req.ctx,
}
resp.svcWaitValue = svcWait
funcSvcGroup.queue.Push(svcWait)
req.responseChannel <- resp
continue
}
// concurrency should not be set to zero and
// sum of specialization in progress and specialized pods should be less then req.concurrency
if req.concurrency > 0 && concurrencyUsed >= req.concurrency {
resp.error = ferror.MakeError(ferror.ErrorTooManyRequests, fmt.Sprintf("function '%s' concurrency '%d' limit reached.", req.function, req.concurrency))
} else {
funcSvcGroup.svcWaiting++
resp.error = ferror.MakeError(ferror.ErrorNotFound, fmt.Sprintf("function '%s' all functions are busy", req.function))
}
req.responseChannel <- resp
case setValue:
if _, ok := c.cache[req.function]; !ok {
c.cache[req.function] = NewFuncSvcGroup()
}
if _, ok := c.cache[req.function].svcs[req.address]; !ok {
c.cache[req.function].svcs[req.address] = &funcSvcInfo{}
}
c.cache[req.function].svcRetain = req.svcsRetain
c.cache[req.function].svcs[req.address].val = req.value
c.cache[req.function].svcs[req.address].activeRequests++
if c.cache[req.function].svcWaiting > 0 {
c.cache[req.function].svcWaiting--
svcCapacity := req.requestsPerPod - c.cache[req.function].svcs[req.address].activeRequests
queueLen := c.cache[req.function].queue.Len()
if svcCapacity > queueLen {
svcCapacity = queueLen
}
for i := 0; i <= svcCapacity; {
popped := c.cache[req.function].queue.Pop()
if popped == nil {
break
}
if popped.ctx.Err() == nil {
popped.svcChannel <- req.value
c.cache[req.function].svcs[req.address].activeRequests++
i++
}
close(popped.svcChannel)
c.cache[req.function].svcWaiting--
}
}
if c.logger.Core().Enabled(zap.DebugLevel) {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Debug("Increase active requests with setValue", zap.String("function", req.function.String()), zap.String("address", req.address), zap.Int("activeRequests", c.cache[req.function].svcs[req.address].activeRequests))
}
c.cache[req.function].svcs[req.address].cpuLimit = req.cpuUsage
case markDeleted:
for key := range c.cache {
if key.UID == req.function.UID {
c.cache[key].deleted = true
break
}
}
case listAvailableValue:
vals := make([]*FuncSvc, 0)
latestFuncGen := make(map[types.UID]int64)
// find the latest generation of each function
for key := range c.cache {
if currentFuncGen, ok := latestFuncGen[key.UID]; ok {
if key.Generation > currentFuncGen {
latestFuncGen[key.UID] = key.Generation
}
} else {
latestFuncGen[key.UID] = key.Generation
}
}
for key1, values := range c.cache {
svcRetain := values.svcRetain
// if the function is not latest generation, then we don't need to retain any pods
if latestFuncGen[key1.UID] != key1.Generation || values.deleted {
svcRetain = 0
}
svcCleanQuota := len(values.svcs) - svcRetain
if svcCleanQuota <= 0 {
continue
}
for key2, value := range values.svcs {
debugLevel := c.logger.Core().Enabled(zap.DebugLevel)
if debugLevel {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Debug("Reading active requests", zap.String("function", key1.String()), zap.String("address", key2), zap.Int("activeRequests", value.activeRequests))
}
if value.activeRequests == 0 && svcCleanQuota > 0 {
if debugLevel {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Debug("Function service with no active requests", zap.String("function", key1.String()), zap.String("address", key2), zap.Int("activeRequests", value.activeRequests))
}
vals = append(vals, value.val)
svcCleanQuota--
}
}
}
resp.allValues = vals
req.responseChannel <- resp
case setCPUUtilization:
if _, ok := c.cache[req.function]; !ok {
c.cache[req.function] = NewFuncSvcGroup()
}
if _, ok := c.cache[req.function].svcs[req.address]; ok {
c.cache[req.function].svcs[req.address].currentCPUUsage = req.cpuUsage
}
case markAvailable:
if _, ok := c.cache[req.function]; ok {
if _, ok = c.cache[req.function].svcs[req.address]; ok {
if c.cache[req.function].svcs[req.address].activeRequests > 0 {
c.cache[req.function].svcs[req.address].activeRequests--
if c.logger.Core().Enabled(zap.DebugLevel) {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Debug("Decrease active requests", zap.String("function", req.function.String()), zap.String("address", req.address), zap.Int("activeRequests", c.cache[req.function].svcs[req.address].activeRequests))
}
} else {
otelUtils.LoggerWithTraceID(req.ctx, c.logger).Error("Invalid request to decrease active requests", zap.String("function", req.function.String()), zap.String("address", req.address), zap.Int("activeRequests", c.cache[req.function].svcs[req.address].activeRequests))
}
}
}
case markSpecializationFailure:
if c.cache[req.function].svcWaiting > c.cache[req.function].queue.Len() {
c.cache[req.function].svcWaiting--
if c.cache[req.function].svcWaiting == c.cache[req.function].queue.Len() {
expiredRequests := c.cache[req.function].queue.Expired()
c.cache[req.function].svcWaiting = c.cache[req.function].svcWaiting - expiredRequests
}
}
case deleteValue:
if funcSvcGroup, ok := c.cache[req.function]; ok {
delete(c.cache[req.function].svcs, req.address)
if funcSvcGroup.deleted && len(c.cache[req.function].svcs) == 0 {
delete(c.cache, req.function)
}
}
req.responseChannel <- resp
case logFuncSvc:
datawriter := bufio.NewWriter(req.dumpWriter)
writefnSvcGrp := func(svcGrp *funcSvcGroup) error {
_, err := datawriter.WriteString(fmt.Sprintf("svc_waiting:%d\tqueue_len:%d", svcGrp.svcWaiting, svcGrp.queue.Len()))
if err != nil {
return err
}
if len(svcGrp.svcs) == 0 {
_, err := datawriter.WriteString("\n")
if err != nil {
return err
}
}
for addr, fnSvc := range svcGrp.svcs {
_, err := datawriter.WriteString(fmt.Sprintf("\tfunction_name:%s\tfn_svc_address:%s\tactive_req:%d\tcurrent_cpu_usage:%v\tcpu_limit:%v\n",
fnSvc.val.Function.Name, addr, fnSvc.activeRequests, fnSvc.currentCPUUsage, fnSvc.cpuLimit))
if err != nil {
return err
}
}
return nil
}
for _, fnSvcGrp := range c.cache {
err := writefnSvcGrp(fnSvcGrp)
if err != nil {
resp.error = err
break
}
}
err := datawriter.Flush()
if err != nil {
resp.error = errors.Join(resp.error, err)
}
req.responseChannel <- resp
default:
resp.error = ferror.MakeError(ferror.ErrorInvalidArgument,
fmt.Sprintf("invalid request type: %v", req.requestType))
req.responseChannel <- resp
}
}
}
func (c *PoolCache) MarkFuncDeleted(function crd.CacheKeyURG) {
c.requestChannel <- &request{
requestType: markDeleted,
function: function,
}
}
// GetValue returns a function service with status in Active else return error
func (c *PoolCache) GetSvcValue(ctx context.Context, function crd.CacheKeyURG, requestsPerPod int, concurrency int) (*FuncSvc, error) {
respChannel := make(chan *response)
c.requestChannel <- &request{
ctx: ctx,
requestType: getValue,
function: function,
concurrency: concurrency,
requestsPerPod: requestsPerPod,
responseChannel: respChannel,
}
resp := <-respChannel
if resp.svcWaitValue != nil {
select {
case <-ctx.Done():
return resp.value, ctx.Err()
case funcSvc := <-resp.svcWaitValue.svcChannel:
return funcSvc, nil
}
}
return resp.value, resp.error
}
// ListAvailableValue returns a list of the available function services stored in the Cache
func (c *PoolCache) ListAvailableValue() []*FuncSvc {
respChannel := make(chan *response)
c.requestChannel <- &request{
requestType: listAvailableValue,
responseChannel: respChannel,
}
resp := <-respChannel
return resp.allValues
}
// SetValue marks the value at key [function][address] as active(begin used)
func (c *PoolCache) SetSvcValue(ctx context.Context, function crd.CacheKeyURG, address string, value *FuncSvc, cpuLimit resource.Quantity, requestsPerPod, svcsRetain int) {
respChannel := make(chan *response)
c.requestChannel <- &request{
ctx: ctx,
requestType: setValue,
function: function,
address: address,
value: value,
cpuUsage: cpuLimit,
requestsPerPod: requestsPerPod,
svcsRetain: svcsRetain,
responseChannel: respChannel,
}
}
// SetCPUUtilization updates/sets the CPU utilization limit for the pod
func (c *PoolCache) SetCPUUtilization(function crd.CacheKeyURG, address string, cpuUsage resource.Quantity) {
c.requestChannel <- &request{
requestType: setCPUUtilization,
function: function,
address: address,
cpuUsage: cpuUsage,
responseChannel: make(chan *response),
}
}
// MarkAvailable marks the value at key [function][address] as available
func (c *PoolCache) MarkAvailable(function crd.CacheKeyURG, address string) {
respChannel := make(chan *response)
c.requestChannel <- &request{
requestType: markAvailable,
function: function,
address: address,
responseChannel: respChannel,
}
}
// DeleteValue deletes the value at key composed of [function][address]
func (c *PoolCache) DeleteValue(ctx context.Context, function crd.CacheKeyURG, address string) error {
respChannel := make(chan *response)
c.requestChannel <- &request{
ctx: ctx,
requestType: deleteValue,
function: function,
address: address,
responseChannel: respChannel,
}
resp := <-respChannel
return resp.error
}
// ReduceSpecializationInProgress reduces the svcWaiting count
func (c *PoolCache) MarkSpecializationFailure(function crd.CacheKeyURG) {
c.requestChannel <- &request{
requestType: markSpecializationFailure,
function: function,
responseChannel: make(chan *response),
}
}
func (c *PoolCache) LogFnSvcGroup(ctx context.Context, file io.Writer) error {
respChannel := make(chan *response)
c.requestChannel <- &request{
requestType: logFuncSvc,
dumpWriter: file,
responseChannel: respChannel,
}
resp := <-respChannel
return resp.error
}