-
Notifications
You must be signed in to change notification settings - Fork 3.8k
/
provider.go
349 lines (311 loc) · 9.13 KB
/
provider.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
// Copyright 2018 The Cockroach 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 provider
import (
"context"
"math"
"sync"
"time"
"github.com/cockroachdb/cockroach/pkg/roachpb"
"github.com/cockroachdb/cockroach/pkg/settings/cluster"
"github.com/cockroachdb/cockroach/pkg/storage/closedts"
"github.com/cockroachdb/cockroach/pkg/storage/closedts/ctpb"
"github.com/cockroachdb/cockroach/pkg/util/hlc"
"github.com/cockroachdb/cockroach/pkg/util/log"
"github.com/cockroachdb/cockroach/pkg/util/log/logtags"
"github.com/cockroachdb/cockroach/pkg/util/stop"
"github.com/cockroachdb/cockroach/pkg/util/syncutil"
"github.com/cockroachdb/cockroach/pkg/util/timeutil"
)
// Config holds the information necessary to create a Provider.
type Config struct {
// NodeID is the ID of the node on which the Provider is housed.
NodeID roachpb.NodeID
Settings *cluster.Settings
Stopper *stop.Stopper
Storage closedts.Storage
Clock closedts.LiveClockFn
Close closedts.CloseFn
}
type subscriber struct {
ch chan<- ctpb.Entry
queue []ctpb.Entry
}
// Provider implements closedts.Provider. It orchestrates the flow of closed
// timestamps and lets callers check whether they can serve reads.
type Provider struct {
cfg *Config
mu struct {
syncutil.RWMutex
*sync.Cond // on RWMutex.RLocker()
// The current subscribers. The goroutine associated to each
// subscriber uses the RLock to mutate its slot. Thus, when
// accessing this slice for any other reason, the write lock
// needs to be acquired.
subscribers []*subscriber
draining bool // tell subscribers to terminate
}
everyClockLog log.EveryN
}
var _ closedts.Provider = (*Provider)(nil)
// NewProvider initializes a Provider, that has yet to be started.
func NewProvider(cfg *Config) *Provider {
p := &Provider{
cfg: cfg,
everyClockLog: log.Every(time.Minute),
}
p.mu.Cond = sync.NewCond(p.mu.RLocker())
return p
}
// Start implements closedts.Provider.
//
// TODO(tschottdorf): the closer functionality could be extracted into its own
// component, which would make the interfaces a little cleaner. Decide whether
// it's worth it during testing.
func (p *Provider) Start() {
p.cfg.Stopper.RunWorker(logtags.AddTag(context.Background(), "ct-closer", nil), p.runCloser)
}
func (p *Provider) drain() {
p.mu.Lock()
p.mu.draining = true
p.mu.Unlock()
for {
p.mu.Broadcast()
p.mu.Lock()
done := true
for _, sub := range p.mu.subscribers {
done = done && sub == nil
}
p.mu.Unlock()
if done {
return
}
}
}
func (p *Provider) runCloser(ctx context.Context) {
// The loop below signals the subscribers, so when it exits it needs to do
// extra work to help the subscribers terminate.
defer p.drain()
if p.cfg.NodeID == 0 {
// This Provider is likely misconfigured.
panic("can't use NodeID zero")
}
ch := p.Notify(p.cfg.NodeID)
defer close(ch)
confCh := make(chan struct{}, 1)
confChanged := func() {
select {
case confCh <- struct{}{}:
default:
}
}
closedts.TargetDuration.SetOnChange(&p.cfg.Settings.SV, confChanged)
var t timeutil.Timer
defer t.Stop()
var lastEpoch ctpb.Epoch
for {
closeFraction := closedts.CloseFraction.Get(&p.cfg.Settings.SV)
targetDuration := float64(closedts.TargetDuration.Get(&p.cfg.Settings.SV))
t.Reset(time.Duration(closeFraction * targetDuration))
select {
case <-p.cfg.Stopper.ShouldQuiesce():
return
case <-ctx.Done():
return
case <-t.C:
t.Read = true
case <-confCh:
// Loop around to use the updated timer.
continue
}
next, epoch, err := p.cfg.Clock(p.cfg.NodeID)
next.WallTime -= int64(targetDuration)
if err != nil {
if p.everyClockLog.ShouldLog() {
log.Warningf(ctx, "unable to move closed timestamp forward: %s", err)
}
// Broadcast even if nothing new was queued, so that the subscribers
// loop to check their client's context.
p.mu.Broadcast()
} else {
closed, m := p.cfg.Close(next)
if log.V(1) {
log.Infof(ctx, "closed ts=%s with %+v, next closed timestamp should be %s", closed, m, next)
}
entry := ctpb.Entry{
Epoch: lastEpoch,
ClosedTimestamp: closed,
MLAI: m,
}
// TODO(tschottdorf): this one-off between the epoch is awkward. Clock() gives us the epoch for `next`
// but the entry wants the epoch for the current closed timestamp. Better to pass both into Close and
// to get both back from it as well.
lastEpoch = epoch
// Simulate a subscription to the local node, so that the new information
// is added to the storage (and thus becomes available to future subscribers
// as well, not only to existing ones). The other end of the chan will Broadcast().
//
// TODO(tschottdorf): the transport should ignore connection requests from
// the node to itself. Those connections would pointlessly loop this around
// once more.
ch <- entry
}
}
}
// Notify implements closedts.Notifyee. It passes the incoming stream of Entries
// to the local Storage.
func (p *Provider) Notify(nodeID roachpb.NodeID) chan<- ctpb.Entry {
ch := make(chan ctpb.Entry)
p.cfg.Stopper.RunWorker(context.Background(), func(ctx context.Context) {
handle := func(entry ctpb.Entry) {
p.cfg.Storage.Add(nodeID, entry)
}
// Special-case data about the origin node, which folks can subscribe to.
// This is easily generalized to also allow subscriptions for data that
// originated on other nodes, but this doesn't seem necessary right now.
if nodeID == p.cfg.NodeID {
handle = func(entry ctpb.Entry) {
// Add to the Storage first.
p.cfg.Storage.Add(nodeID, entry)
// Notify existing subscribers.
p.mu.Lock()
for _, sub := range p.mu.subscribers {
if sub == nil {
continue
}
sub.queue = append(sub.queue, entry)
}
p.mu.Unlock()
// Wake up all clients.
p.mu.Broadcast()
}
}
for entry := range ch {
handle(entry)
}
})
return ch
}
// Subscribe implements closedts.Producer. It produces a stream of Entries
// pertaining to the local Node.
//
// TODO(tschottdorf): consider not forcing the caller to launch the goroutine.
func (p *Provider) Subscribe(ctx context.Context, ch chan<- ctpb.Entry) {
var i int
sub := &subscriber{ch, nil}
p.mu.Lock()
for i = 0; i < len(p.mu.subscribers); i++ {
if p.mu.subscribers[i] == nil {
p.mu.subscribers[i] = sub
break
}
}
if i == len(p.mu.subscribers) {
p.mu.subscribers = append(p.mu.subscribers, sub)
}
draining := p.mu.draining
p.mu.Unlock()
defer func() {
p.mu.Lock()
p.mu.subscribers[i] = nil
p.mu.Unlock()
close(ch)
}()
if draining {
return
}
if log.V(1) {
log.Infof(ctx, "new subscriber (slot %d) connected", i)
}
// The subscription is already active, so any storage snapshot from now on is
// going to fully catch up the subscriber without a gap.
{
var entries []ctpb.Entry
p.cfg.Storage.VisitAscending(p.cfg.NodeID, func(entry ctpb.Entry) (done bool) {
// Don't block in this method.
entries = append(entries, entry)
return false // not done
})
for _, entry := range entries {
select {
case ch <- entry:
case <-ctx.Done():
return
}
}
}
for {
p.mu.RLock()
var done bool
for len(p.mu.subscribers[i].queue) == 0 {
if ctx.Err() != nil || p.mu.draining {
done = true
break
}
p.mu.Wait()
}
var queue []ctpb.Entry
// When only readers are around (as they are now), we can actually
// mutate our slot because that's all the others do as well.
queue, p.mu.subscribers[i].queue = p.mu.subscribers[i].queue, nil
p.mu.RUnlock()
if done {
return
}
shouldLog := log.V(1)
var n int
minMLAI := ctpb.LAI(math.MaxInt64)
var minRangeID, maxRangeID roachpb.RangeID
var maxMLAI ctpb.LAI
for _, entry := range queue {
if shouldLog {
n += len(entry.MLAI)
for rangeID, mlai := range entry.MLAI {
if mlai < minMLAI {
minMLAI = mlai
minRangeID = rangeID
}
if mlai > maxMLAI {
maxMLAI = mlai
maxRangeID = rangeID
}
}
}
select {
case ch <- entry:
case <-ctx.Done():
return
}
}
if shouldLog {
log.Infof(ctx, "sent %d closed timestamp entries to client %d (%d range updates total, min/max mlai: %d@r%d / %d@r%d)", len(queue), i, n, minMLAI, minRangeID, maxMLAI, maxRangeID)
}
}
}
// MaxClosed implements closedts.Provider.
func (p *Provider) MaxClosed(
nodeID roachpb.NodeID, rangeID roachpb.RangeID, epoch ctpb.Epoch, lai ctpb.LAI,
) hlc.Timestamp {
var maxTS hlc.Timestamp
p.cfg.Storage.VisitDescending(nodeID, func(entry ctpb.Entry) (done bool) {
if mlai, found := entry.MLAI[rangeID]; found {
if entry.Epoch == epoch && mlai <= lai {
maxTS = entry.ClosedTimestamp
return true
}
}
return false
})
return maxTS
}