forked from libp2p/go-libp2p
-
Notifications
You must be signed in to change notification settings - Fork 0
/
addr_book_gc.go
408 lines (355 loc) · 12.2 KB
/
addr_book_gc.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
package pstoreds
import (
"context"
"fmt"
"strconv"
"time"
"github.com/AstaFrode/go-libp2p/core/peer"
"github.com/AstaFrode/go-libp2p/p2p/host/peerstore/pstoreds/pb"
"google.golang.org/protobuf/proto"
ds "github.com/ipfs/go-datastore"
"github.com/ipfs/go-datastore/query"
b32 "github.com/multiformats/go-base32"
)
var (
// GC lookahead entries are stored in key pattern:
// /peers/gc/addrs/<unix timestamp of next visit>/<peer ID b32> => nil
// in databases with lexicographical key order, this time-indexing allows us to visit
// only the timeslice we are interested in.
gcLookaheadBase = ds.NewKey("/peers/gc/addrs")
// queries
purgeLookaheadQuery = query.Query{
Prefix: gcLookaheadBase.String(),
Orders: []query.Order{query.OrderByFunction(orderByTimestampInKey)},
KeysOnly: true,
}
purgeStoreQuery = query.Query{
Prefix: addrBookBase.String(),
Orders: []query.Order{query.OrderByKey{}},
KeysOnly: false,
}
populateLookaheadQuery = query.Query{
Prefix: addrBookBase.String(),
Orders: []query.Order{query.OrderByKey{}},
KeysOnly: true,
}
)
// dsAddrBookGc is responsible for garbage collection in a datastore-backed address book.
type dsAddrBookGc struct {
ctx context.Context
ab *dsAddrBook
running chan struct{}
lookaheadEnabled bool
purgeFunc func()
currWindowEnd int64
}
func newAddressBookGc(ctx context.Context, ab *dsAddrBook) (*dsAddrBookGc, error) {
if ab.opts.GCPurgeInterval < 0 {
return nil, fmt.Errorf("negative GC purge interval provided: %s", ab.opts.GCPurgeInterval)
}
if ab.opts.GCLookaheadInterval < 0 {
return nil, fmt.Errorf("negative GC lookahead interval provided: %s", ab.opts.GCLookaheadInterval)
}
if ab.opts.GCInitialDelay < 0 {
return nil, fmt.Errorf("negative GC initial delay provided: %s", ab.opts.GCInitialDelay)
}
if ab.opts.GCLookaheadInterval > 0 && ab.opts.GCLookaheadInterval < ab.opts.GCPurgeInterval {
return nil, fmt.Errorf("lookahead interval must be larger than purge interval, respectively: %s, %s",
ab.opts.GCLookaheadInterval, ab.opts.GCPurgeInterval)
}
lookaheadEnabled := ab.opts.GCLookaheadInterval > 0
gc := &dsAddrBookGc{
ctx: ctx,
ab: ab,
running: make(chan struct{}, 1),
lookaheadEnabled: lookaheadEnabled,
}
if lookaheadEnabled {
gc.purgeFunc = gc.purgeLookahead
} else {
gc.purgeFunc = gc.purgeStore
}
// do not start GC timers if purge is disabled; this GC can only be triggered manually.
if ab.opts.GCPurgeInterval > 0 {
gc.ab.childrenDone.Add(1)
go gc.background()
}
return gc, nil
}
// gc prunes expired addresses from the datastore at regular intervals. It should be spawned as a goroutine.
func (gc *dsAddrBookGc) background() {
defer gc.ab.childrenDone.Done()
select {
case <-gc.ab.clock.After(gc.ab.opts.GCInitialDelay):
case <-gc.ab.ctx.Done():
// yield if we have been cancelled/closed before the delay elapses.
return
}
purgeTimer := time.NewTicker(gc.ab.opts.GCPurgeInterval)
defer purgeTimer.Stop()
var lookaheadCh <-chan time.Time
if gc.lookaheadEnabled {
lookaheadTimer := time.NewTicker(gc.ab.opts.GCLookaheadInterval)
lookaheadCh = lookaheadTimer.C
gc.populateLookahead() // do a lookahead now
defer lookaheadTimer.Stop()
}
for {
select {
case <-purgeTimer.C:
gc.purgeFunc()
case <-lookaheadCh:
// will never trigger if lookahead is disabled (nil Duration).
gc.populateLookahead()
case <-gc.ctx.Done():
return
}
}
}
// purgeCycle runs a single GC purge cycle. It operates within the lookahead window if lookahead is enabled; else it
// visits all entries in the datastore, deleting the addresses that have expired.
func (gc *dsAddrBookGc) purgeLookahead() {
select {
case gc.running <- struct{}{}:
defer func() { <-gc.running }()
default:
// yield if lookahead is running.
return
}
var id peer.ID
record := &addrsRecord{AddrBookRecord: &pb.AddrBookRecord{}} // empty record to reuse and avoid allocs.
batch, err := newCyclicBatch(gc.ab.ds, defaultOpsPerCyclicBatch)
if err != nil {
log.Warnf("failed while creating batch to purge GC entries: %v", err)
}
// This function drops an unparseable GC entry; this is for safety. It is an escape hatch in case
// we modify the format of keys going forward. If a user runs a new version against an old DB,
// if we don't clean up unparseable entries we'll end up accumulating garbage.
dropInError := func(key ds.Key, err error, msg string) {
if err != nil {
log.Warnf("failed while %s record with GC key: %v, err: %v; deleting", msg, key, err)
}
if err = batch.Delete(context.TODO(), key); err != nil {
log.Warnf("failed to delete corrupt GC lookahead entry: %v, err: %v", key, err)
}
}
// This function drops a GC key if the entry is cleaned correctly. It may reschedule another visit
// if the next earliest expiry falls within the current window again.
dropOrReschedule := func(key ds.Key, ar *addrsRecord) {
if err := batch.Delete(context.TODO(), key); err != nil {
log.Warnf("failed to delete lookahead entry: %v, err: %v", key, err)
}
// re-add the record if it needs to be visited again in this window.
if len(ar.Addrs) != 0 && ar.Addrs[0].Expiry <= gc.currWindowEnd {
gcKey := gcLookaheadBase.ChildString(fmt.Sprintf("%d/%s", ar.Addrs[0].Expiry, key.Name()))
if err := batch.Put(context.TODO(), gcKey, []byte{}); err != nil {
log.Warnf("failed to add new GC key: %v, err: %v", gcKey, err)
}
}
}
results, err := gc.ab.ds.Query(context.TODO(), purgeLookaheadQuery)
if err != nil {
log.Warnf("failed while fetching entries to purge: %v", err)
return
}
defer results.Close()
now := gc.ab.clock.Now().Unix()
// keys: /peers/gc/addrs/<unix timestamp of next visit>/<peer ID b32>
// values: nil
for result := range results.Next() {
gcKey := ds.RawKey(result.Key)
ts, err := strconv.ParseInt(gcKey.Parent().Name(), 10, 64)
if err != nil {
dropInError(gcKey, err, "parsing timestamp")
log.Warnf("failed while parsing timestamp from key: %v, err: %v", result.Key, err)
continue
} else if ts > now {
// this is an ordered cursor; when we hit an entry with a timestamp beyond now, we can break.
break
}
idb32, err := b32.RawStdEncoding.DecodeString(gcKey.Name())
if err != nil {
dropInError(gcKey, err, "parsing peer ID")
log.Warnf("failed while parsing b32 peer ID from key: %v, err: %v", result.Key, err)
continue
}
id, err = peer.IDFromBytes(idb32)
if err != nil {
dropInError(gcKey, err, "decoding peer ID")
log.Warnf("failed while decoding peer ID from key: %v, err: %v", result.Key, err)
continue
}
// if the record is in cache, we clean it and flush it if necessary.
if cached, ok := gc.ab.cache.Peek(id); ok {
cached.Lock()
if cached.clean(gc.ab.clock.Now()) {
if err = cached.flush(batch); err != nil {
log.Warnf("failed to flush entry modified by GC for peer: &v, err: %v", id.Pretty(), err)
}
}
dropOrReschedule(gcKey, cached)
cached.Unlock()
continue
}
record.Reset()
// otherwise, fetch it from the store, clean it and flush it.
entryKey := addrBookBase.ChildString(gcKey.Name())
val, err := gc.ab.ds.Get(context.TODO(), entryKey)
if err != nil {
// captures all errors, including ErrNotFound.
dropInError(gcKey, err, "fetching entry")
continue
}
err = proto.Unmarshal(val, record)
if err != nil {
dropInError(gcKey, err, "unmarshalling entry")
continue
}
if record.clean(gc.ab.clock.Now()) {
err = record.flush(batch)
if err != nil {
log.Warnf("failed to flush entry modified by GC for peer: &v, err: %v", id.Pretty(), err)
}
}
dropOrReschedule(gcKey, record)
}
if err = batch.Commit(context.TODO()); err != nil {
log.Warnf("failed to commit GC purge batch: %v", err)
}
}
func (gc *dsAddrBookGc) purgeStore() {
select {
case gc.running <- struct{}{}:
defer func() { <-gc.running }()
default:
// yield if lookahead is running.
return
}
record := &addrsRecord{AddrBookRecord: &pb.AddrBookRecord{}} // empty record to reuse and avoid allocs.
batch, err := newCyclicBatch(gc.ab.ds, defaultOpsPerCyclicBatch)
if err != nil {
log.Warnf("failed while creating batch to purge GC entries: %v", err)
}
results, err := gc.ab.ds.Query(context.TODO(), purgeStoreQuery)
if err != nil {
log.Warnf("failed while opening iterator: %v", err)
return
}
defer results.Close()
// keys: /peers/addrs/<peer ID b32>
for result := range results.Next() {
record.Reset()
if err = proto.Unmarshal(result.Value, record); err != nil {
// TODO log
continue
}
id := record.Id
if !record.clean(gc.ab.clock.Now()) {
continue
}
if err := record.flush(batch); err != nil {
log.Warnf("failed to flush entry modified by GC for peer: &v, err: %v", id, err)
}
gc.ab.cache.Remove(peer.ID(id))
}
if err = batch.Commit(context.TODO()); err != nil {
log.Warnf("failed to commit GC purge batch: %v", err)
}
}
// populateLookahead populates the lookahead window by scanning the entire store and picking entries whose earliest
// expiration falls within the window period.
//
// Those entries are stored in the lookahead region in the store, indexed by the timestamp when they need to be
// visited, to facilitate temporal range scans.
func (gc *dsAddrBookGc) populateLookahead() {
if gc.ab.opts.GCLookaheadInterval == 0 {
return
}
select {
case gc.running <- struct{}{}:
defer func() { <-gc.running }()
default:
// yield if something's running.
return
}
until := gc.ab.clock.Now().Add(gc.ab.opts.GCLookaheadInterval).Unix()
var id peer.ID
record := &addrsRecord{AddrBookRecord: &pb.AddrBookRecord{}}
results, err := gc.ab.ds.Query(context.TODO(), populateLookaheadQuery)
if err != nil {
log.Warnf("failed while querying to populate lookahead GC window: %v", err)
return
}
defer results.Close()
batch, err := newCyclicBatch(gc.ab.ds, defaultOpsPerCyclicBatch)
if err != nil {
log.Warnf("failed while creating batch to populate lookahead GC window: %v", err)
return
}
for result := range results.Next() {
idb32 := ds.RawKey(result.Key).Name()
k, err := b32.RawStdEncoding.DecodeString(idb32)
if err != nil {
log.Warnf("failed while decoding peer ID from key: %v, err: %v", result.Key, err)
continue
}
if id, err = peer.IDFromBytes(k); err != nil {
log.Warnf("failed while decoding peer ID from key: %v, err: %v", result.Key, err)
}
// if the record is in cache, use the cached version.
if cached, ok := gc.ab.cache.Peek(id); ok {
cached.RLock()
if len(cached.Addrs) == 0 || cached.Addrs[0].Expiry > until {
cached.RUnlock()
continue
}
gcKey := gcLookaheadBase.ChildString(fmt.Sprintf("%d/%s", cached.Addrs[0].Expiry, idb32))
if err = batch.Put(context.TODO(), gcKey, []byte{}); err != nil {
log.Warnf("failed while inserting GC entry for peer: %v, err: %v", id.Pretty(), err)
}
cached.RUnlock()
continue
}
record.Reset()
val, err := gc.ab.ds.Get(context.TODO(), ds.RawKey(result.Key))
if err != nil {
log.Warnf("failed which getting record from store for peer: %v, err: %v", id.Pretty(), err)
continue
}
if err := proto.Unmarshal(val, record); err != nil {
log.Warnf("failed while unmarshalling record from store for peer: %v, err: %v", id.Pretty(), err)
continue
}
if len(record.Addrs) > 0 && record.Addrs[0].Expiry <= until {
gcKey := gcLookaheadBase.ChildString(fmt.Sprintf("%d/%s", record.Addrs[0].Expiry, idb32))
if err = batch.Put(context.TODO(), gcKey, []byte{}); err != nil {
log.Warnf("failed while inserting GC entry for peer: %v, err: %v", id.Pretty(), err)
}
}
}
if err = batch.Commit(context.TODO()); err != nil {
log.Warnf("failed to commit GC lookahead batch: %v", err)
}
gc.currWindowEnd = until
}
// orderByTimestampInKey orders the results by comparing the timestamp in the
// key. A lexiographic sort by itself is wrong since "10" is less than "2", but
// as an int 2 is obviously less than 10.
func orderByTimestampInKey(a, b query.Entry) int {
aKey := ds.RawKey(a.Key)
aInt, err := strconv.ParseInt(aKey.Parent().Name(), 10, 64)
if err != nil {
return -1
}
bKey := ds.RawKey(b.Key)
bInt, err := strconv.ParseInt(bKey.Parent().Name(), 10, 64)
if err != nil {
return -1
}
if aInt < bInt {
return -1
} else if aInt == bInt {
return 0
}
return 1
}