-
Notifications
You must be signed in to change notification settings - Fork 793
/
scheduler.go
327 lines (288 loc) · 10.5 KB
/
scheduler.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
package ruler
import (
"context"
"fmt"
"hash"
"hash/fnv"
"math"
"sync"
"time"
"github.com/go-kit/kit/log/level"
"github.com/jonboulle/clockwork"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
"github.com/prometheus/prometheus/rules"
"github.com/cortexproject/cortex/pkg/configs"
config_client "github.com/cortexproject/cortex/pkg/configs/client"
"github.com/cortexproject/cortex/pkg/util"
)
var backoffConfig = util.BackoffConfig{
// Backoff for loading initial configuration set.
MinBackoff: 100 * time.Millisecond,
MaxBackoff: 2 * time.Second,
}
const (
timeLogFormat = "2006-01-02T15:04:05"
)
var (
totalConfigs = promauto.NewGauge(prometheus.GaugeOpts{
Namespace: "cortex",
Name: "scheduler_configs_total",
Help: "How many configs the scheduler knows about.",
})
totalRuleGroups = promauto.NewGauge(prometheus.GaugeOpts{
Namespace: "cortex",
Name: "scheduler_groups_total",
Help: "How many rule groups the scheduler is currently evaluating",
})
configUpdates = promauto.NewCounter(prometheus.CounterOpts{
Namespace: "cortex",
Name: "scheduler_config_updates_total",
Help: "How many config updates the scheduler has made.",
})
)
type workItem struct {
userID string
groupName string
hash uint32
group *group
scheduled time.Time
generation configs.ID // a monotonically increasing number used to spot out of date work items
}
// Key implements ScheduledItem
func (w workItem) Key() string {
return w.userID + ":" + w.groupName
}
// Scheduled implements ScheduledItem
func (w workItem) Scheduled() time.Time {
return w.scheduled
}
// Defer returns a work item with updated rules, rescheduled to a later time.
func (w workItem) Defer(interval time.Duration) workItem {
return workItem{w.userID, w.groupName, w.hash, w.group, w.scheduled.Add(interval), w.generation}
}
func (w workItem) String() string {
return fmt.Sprintf("%s:%s:%d@%s", w.userID, w.groupName, len(w.group.Rules()), w.scheduled.Format(timeLogFormat))
}
type userConfig struct {
rules map[string][]rules.Rule
generation configs.ID // a monotonically increasing number used to spot out of date work items
}
type groupFactory func(userID string, groupName string, rls []rules.Rule) (*group, error)
type scheduler struct {
ruleStore config_client.Client
evaluationInterval time.Duration // how often we re-evaluate each rule set
q *SchedulingQueue
pollInterval time.Duration // how often we check for new config
cfgs map[string]userConfig // all rules for all users
latestConfig configs.ID // # of last update received from config
groupFn groupFactory // function to create a new group
sync.RWMutex
done chan struct{}
}
// newScheduler makes a new scheduler.
func newScheduler(ruleStore config_client.Client, evaluationInterval, pollInterval time.Duration, groupFn groupFactory) *scheduler {
return &scheduler{
ruleStore: ruleStore,
evaluationInterval: evaluationInterval,
pollInterval: pollInterval,
q: NewSchedulingQueue(clockwork.NewRealClock()),
cfgs: map[string]userConfig{},
groupFn: groupFn,
done: make(chan struct{}),
}
}
// Run polls the source of configurations for changes.
func (s *scheduler) Run() {
level.Debug(util.Logger).Log("msg", "scheduler started")
// Load initial set of all configurations before polling for new ones.
s.addNewConfigs(time.Now(), s.loadAllConfigs())
ticker := time.NewTicker(s.pollInterval)
for {
select {
case now := <-ticker.C:
err := s.updateConfigs(now)
if err != nil {
level.Warn(util.Logger).Log("msg", "scheduler: error updating configs", "err", err)
}
case <-s.done:
ticker.Stop()
level.Debug(util.Logger).Log("msg", "scheduler config polling stopped")
return
}
}
}
func (s *scheduler) Stop() {
close(s.done)
s.q.Close()
level.Debug(util.Logger).Log("msg", "scheduler stopped")
}
// Load the full set of configurations from the server, retrying with backoff
// until we can get them.
func (s *scheduler) loadAllConfigs() map[string]configs.VersionedRulesConfig {
backoff := util.NewBackoff(context.Background(), backoffConfig)
for {
cfgs, err := s.poll()
if err == nil {
level.Debug(util.Logger).Log("msg", "scheduler: initial configuration load", "num_configs", len(cfgs))
return cfgs
}
level.Warn(util.Logger).Log("msg", "scheduler: error fetching all configurations, backing off", "err", err)
backoff.Wait()
}
}
func (s *scheduler) updateConfigs(now time.Time) error {
cfgs, err := s.poll()
if err != nil {
return err
}
s.addNewConfigs(now, cfgs)
return nil
}
// poll the configuration server. Not re-entrant.
func (s *scheduler) poll() (map[string]configs.VersionedRulesConfig, error) {
s.Lock()
configID := s.latestConfig
s.Unlock()
cfgs, err := s.ruleStore.GetRules(context.Background(), configID) // Warning: this will produce an incorrect result if the configID ever overflows
if err != nil {
level.Warn(util.Logger).Log("msg", "scheduler: configs server poll failed", "err", err)
return nil, err
}
s.Lock()
s.latestConfig = getLatestConfigID(cfgs, configID)
s.Unlock()
return cfgs, nil
}
// getLatestConfigID gets the latest configs ID.
// max [latest, max (map getID cfgs)]
func getLatestConfigID(cfgs map[string]configs.VersionedRulesConfig, latest configs.ID) configs.ID {
ret := latest
for _, config := range cfgs {
if config.ID > ret {
ret = config.ID
}
}
return ret
}
// computeNextEvalTime Computes when a user's rules should be next evaluated, based on how far we are through an evaluation cycle
func (s *scheduler) computeNextEvalTime(hasher hash.Hash64, now time.Time, userID string) time.Time {
intervalNanos := float64(s.evaluationInterval.Nanoseconds())
// Compute how far we are into the current evaluation cycle
currentEvalCyclePoint := math.Mod(float64(now.UnixNano()), intervalNanos)
hasher.Reset()
hasher.Write([]byte(userID))
offset := math.Mod(
// We subtract our current point in the cycle to cause the entries
// before 'now' to wrap around to the end.
// We don't want this to come out negative, so we add the interval to it
float64(hasher.Sum64())+intervalNanos-currentEvalCyclePoint,
intervalNanos)
return now.Add(time.Duration(int64(offset)))
}
func (s *scheduler) addNewConfigs(now time.Time, cfgs map[string]configs.VersionedRulesConfig) {
// TODO: instrument how many configs we have, both valid & invalid.
level.Debug(util.Logger).Log("msg", "adding configurations", "num_configs", len(cfgs))
hasher := fnv.New64a()
s.Lock()
generation := s.latestConfig
s.Unlock()
for userID, config := range cfgs {
s.addUserConfig(now, hasher, generation, userID, config)
}
configUpdates.Add(float64(len(cfgs)))
s.Lock()
lenCfgs := len(s.cfgs)
s.Unlock()
totalConfigs.Set(float64(lenCfgs))
}
func (s *scheduler) addUserConfig(now time.Time, hasher hash.Hash64, generation configs.ID, userID string, config configs.VersionedRulesConfig) {
rulesByGroup, err := config.Config.Parse()
if err != nil {
// XXX: This means that if a user has a working configuration and
// they submit a broken one, we'll keep processing the last known
// working configuration, and they'll never know.
// TODO: Provide a way of deleting / cancelling recording rules.
level.Warn(util.Logger).Log("msg", "scheduler: invalid Cortex configuration", "user_id", userID, "err", err)
return
}
level.Info(util.Logger).Log("msg", "scheduler: updating rules for user", "user_id", userID, "num_groups", len(rulesByGroup), "is_deleted", config.IsDeleted())
s.Lock()
// if deleted remove from map, otherwise - update map
if config.IsDeleted() {
delete(s.cfgs, userID)
s.Unlock()
return
}
s.cfgs[userID] = userConfig{rules: rulesByGroup, generation: generation}
s.Unlock()
ringHasher := fnv.New32a()
evalTime := s.computeNextEvalTime(hasher, now, userID)
workItems := []workItem{}
for group, rules := range rulesByGroup {
level.Debug(util.Logger).Log("msg", "scheduler: updating rules for user and group", "user_id", userID, "group", group, "num_rules", len(rules))
g, err := s.groupFn(userID, group, rules)
if err != nil {
// XXX: similarly to above if a user has a working configuration and
// for some reason we cannot create a group for the new one we'll use
// the last known working configuration
level.Warn(util.Logger).Log("msg", "scheduler: failed to create group for user", "user_id", userID, "group", group, "err", err)
return
}
ringHasher.Reset()
ringHasher.Write([]byte(userID + ":" + group))
hash := ringHasher.Sum32()
workItems = append(workItems, workItem{userID, group, hash, g, evalTime, generation})
}
for _, i := range workItems {
totalRuleGroups.Inc()
s.addWorkItem(i)
}
}
func (s *scheduler) addWorkItem(i workItem) {
select {
case <-s.done:
level.Debug(util.Logger).Log("msg", "scheduler: work item not added, scheduler stoped", "item", i)
return
default:
// The queue is keyed by userID+groupName, so items for existing userID+groupName will be replaced.
s.q.Enqueue(i)
level.Debug(util.Logger).Log("msg", "scheduler: work item added", "item", i)
}
}
// Get the next scheduled work item, blocking if none.
//
// Call `workItemDone` on the returned item to indicate that it is ready to be
// rescheduled.
func (s *scheduler) nextWorkItem() *workItem {
level.Debug(util.Logger).Log("msg", "scheduler: work item requested, pending...")
// TODO: We are blocking here on the second Dequeue event. Write more
// tests for the scheduling queue.
op := s.q.Dequeue()
if op == nil {
level.Info(util.Logger).Log("msg", "queue closed; no more work items")
return nil
}
item := op.(workItem)
level.Debug(util.Logger).Log("msg", "scheduler: work item granted", "item", item)
return &item
}
// workItemDone marks the given item as being ready to be rescheduled.
func (s *scheduler) workItemDone(i workItem) {
s.Lock()
config, found := s.cfgs[i.userID]
var currentRules []rules.Rule
if found {
currentRules = config.rules[i.groupName]
}
s.Unlock()
if !found || len(currentRules) == 0 || i.generation < config.generation {
// Warning: this test will produce an incorrect result if the generation ever overflows
level.Debug(util.Logger).Log("msg", "scheduler: stopping item", "user_id", i.userID, "group", i.groupName, "found", found, "len", len(currentRules))
totalRuleGroups.Dec()
return
}
next := i.Defer(s.evaluationInterval)
level.Debug(util.Logger).Log("msg", "scheduler: work item rescheduled", "item", i, "time", next.scheduled.Format(timeLogFormat))
s.addWorkItem(next)
}