-
Notifications
You must be signed in to change notification settings - Fork 4.2k
/
snapshot.go
288 lines (241 loc) · 7.54 KB
/
snapshot.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
package raft
import (
"context"
"errors"
"fmt"
"io"
"io/ioutil"
"sync"
log "github.com/hashicorp/go-hclog"
"github.com/hashicorp/raft"
)
const (
// boltSnapshotID is the stable ID for any boltDB snapshot. Keeping the ID
// stable means there is only ever one bolt snapshot in the system
boltSnapshotID = "bolt-snapshot"
)
// BoltSnapshotStore implements the SnapshotStore interface and allows
// snapshots to be made on the local disk. The main difference between this
// store and the file store is we make the distinction between snapshots that
// have been written by the FSM and by internal Raft operations. The former are
// treated as noop snapshots on Persist and are read in full from the FSM on
// Open. The latter are treated like normal file snapshots and are able to be
// opened and applied as usual.
type BoltSnapshotStore struct {
// path is the directory in which to store file based snapshots
path string
// retain is the number of file based snapshots to keep
retain int
// We hold a copy of the FSM so we can stream snapshots straight out of the
// database.
fsm *FSM
// fileSnapStore is used to fall back to file snapshots when the data is
// being written from the raft library. This currently only happens on a
// follower during a snapshot install RPC.
fileSnapStore *raft.FileSnapshotStore
logger log.Logger
}
// BoltSnapshotSink implements SnapshotSink optionally choosing to write to a
// file.
type BoltSnapshotSink struct {
store *BoltSnapshotStore
logger log.Logger
meta raft.SnapshotMeta
trans raft.Transport
fileSink raft.SnapshotSink
l sync.Mutex
closed bool
}
// NewBoltSnapshotStore creates a new BoltSnapshotStore based
// on a base directory. The `retain` parameter controls how many
// snapshots are retained. Must be at least 1.
func NewBoltSnapshotStore(base string, retain int, logger log.Logger, fsm *FSM) (*BoltSnapshotStore, error) {
if retain < 1 {
return nil, fmt.Errorf("must retain at least one snapshot")
}
if logger == nil {
return nil, fmt.Errorf("no logger provided")
}
fileStore, err := raft.NewFileSnapshotStore(base, retain, nil)
if err != nil {
return nil, err
}
// Setup the store
store := &BoltSnapshotStore{
logger: logger,
fsm: fsm,
fileSnapStore: fileStore,
}
{
// TODO: I think this needs to be done before every NewRaft and
// RecoverCluster call. Not just on Factory method.
// Here we delete all the existing file based snapshots. This is necessary
// because we do not issue a restore on NewRaft. If a previous file snapshot
// had failed to apply we will be incorrectly setting the indexes. It's
// safer to simply delete all file snapshots on startup and rely on Raft to
// reconcile the FSM state.
if err := store.ReapSnapshots(); err != nil {
return nil, err
}
}
return store, nil
}
// Create is used to start a new snapshot
func (f *BoltSnapshotStore) Create(version raft.SnapshotVersion, index, term uint64,
configuration raft.Configuration, configurationIndex uint64, trans raft.Transport) (raft.SnapshotSink, error) {
// We only support version 1 snapshots at this time.
if version != 1 {
return nil, fmt.Errorf("unsupported snapshot version %d", version)
}
// We are processing a snapshot, fastforward the index, term, and
// configuration to the latest seen by the raft system. This could include
// log indexes for operation types that are never sent to the FSM.
if err := f.fsm.witnessSnapshot(index, term, configurationIndex, configuration); err != nil {
return nil, err
}
// Create the sink
sink := &BoltSnapshotSink{
store: f,
logger: f.logger,
meta: raft.SnapshotMeta{
Version: version,
ID: boltSnapshotID,
Index: index,
Term: term,
Configuration: configuration,
ConfigurationIndex: configurationIndex,
},
trans: trans,
}
// Done
return sink, nil
}
// List returns available snapshots in the store. It only returns bolt
// snapshots. No snapshot will be returned if there are no indexes in the
// FSM.
func (f *BoltSnapshotStore) List() ([]*raft.SnapshotMeta, error) {
meta, err := f.getBoltSnapshotMeta()
if err != nil {
return nil, err
}
// If we haven't seen any data yet do not return a snapshot
if meta.Index == 0 {
return nil, nil
}
return []*raft.SnapshotMeta{meta}, nil
}
// getBoltSnapshotMeta returns the fsm's latest state and configuration.
func (f *BoltSnapshotStore) getBoltSnapshotMeta() (*raft.SnapshotMeta, error) {
latestIndex, latestConfig := f.fsm.LatestState()
meta := &raft.SnapshotMeta{
Version: 1,
ID: boltSnapshotID,
Index: latestIndex.Index,
Term: latestIndex.Term,
}
if latestConfig != nil {
index, configuration := protoConfigurationToRaftConfiguration(latestConfig)
meta.Configuration = configuration
meta.ConfigurationIndex = index
}
return meta, nil
}
// Open takes a snapshot ID and returns a ReadCloser for that snapshot.
func (f *BoltSnapshotStore) Open(id string) (*raft.SnapshotMeta, io.ReadCloser, error) {
var readCloser io.ReadCloser
var meta *raft.SnapshotMeta
switch id {
case boltSnapshotID:
var err error
meta, err = f.getBoltSnapshotMeta()
if err != nil {
return nil, nil, err
}
// If we don't have any data return an error
if meta.Index == 0 {
return nil, nil, errors.New("no snapshot data")
}
// Stream data out of the FSM to calculate the size
var writeCloser *io.PipeWriter
readCloser, writeCloser = io.Pipe()
metaReadCloser, metaWriteCloser := io.Pipe()
go func() {
f.fsm.writeTo(context.Background(), metaWriteCloser, writeCloser)
}()
// Compute the size
n, err := io.Copy(ioutil.Discard, metaReadCloser)
if err != nil {
f.logger.Error("failed to read state file", "error", err)
metaReadCloser.Close()
readCloser.Close()
return nil, nil, err
}
meta.Size = n
default:
var err error
meta, readCloser, err = f.fileSnapStore.Open(id)
if err != nil {
return nil, nil, err
}
}
return meta, readCloser, nil
}
// ReapSnapshots reaps any snapshots beyond the retain count.
func (f *BoltSnapshotStore) ReapSnapshots() error {
return f.fileSnapStore.ReapSnapshots()
}
// ID returns the ID of the snapshot, can be used with Open()
// after the snapshot is finalized.
func (s *BoltSnapshotSink) ID() string {
s.l.Lock()
defer s.l.Unlock()
if s.fileSink != nil {
return s.fileSink.ID()
}
return s.meta.ID
}
// Write is used to append to the state file. We write to the
// buffered IO object to reduce the amount of context switches.
func (s *BoltSnapshotSink) Write(b []byte) (int, error) {
s.l.Lock()
defer s.l.Unlock()
// If someone is writing to this sink then we need to create a file sink to
// capture the data. This currently only happens when a follower is being
// sent a snapshot.
if s.fileSink == nil {
fileSink, err := s.store.fileSnapStore.Create(s.meta.Version, s.meta.Index, s.meta.Term, s.meta.Configuration, s.meta.ConfigurationIndex, s.trans)
if err != nil {
return 0, err
}
s.fileSink = fileSink
}
return s.fileSink.Write(b)
}
// Close is used to indicate a successful end.
func (s *BoltSnapshotSink) Close() error {
s.l.Lock()
defer s.l.Unlock()
// Make sure close is idempotent
if s.closed {
return nil
}
s.closed = true
if s.fileSink != nil {
return s.fileSink.Close()
}
return nil
}
// Cancel is used to indicate an unsuccessful end.
func (s *BoltSnapshotSink) Cancel() error {
s.l.Lock()
defer s.l.Unlock()
// Make sure close is idempotent
if s.closed {
return nil
}
s.closed = true
if s.fileSink != nil {
return s.fileSink.Cancel()
}
return nil
}