forked from dgraph-io/dgraph
/
backup.go
312 lines (283 loc) · 7.56 KB
/
backup.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
package worker
import (
"bufio"
"bytes"
"compress/gzip"
"fmt"
"math/rand"
"os"
"path"
"sync"
"time"
"github.com/dgraph-io/dgraph/group"
"github.com/dgraph-io/dgraph/types"
"github.com/dgraph-io/dgraph/x"
"golang.org/x/net/context"
"google.golang.org/grpc"
)
const numBackupRoutines = 10
type kv struct {
prefix string
list *types.PostingList
}
func toRDF(buf *bytes.Buffer, item kv) {
pl := item.list
for _, p := range pl.Postings {
x.Check2(buf.WriteString(item.prefix))
if !bytes.Equal(p.Value, nil) {
// Value posting
// Convert to appropriate type
vID := types.TypeID(p.ValType)
src := types.ValueForType(vID)
src.Value = p.Value
str, err := types.Convert(src, types.StringID)
x.Check(err)
x.Check2(buf.WriteString(fmt.Sprintf("\"%s\"", str.Value)))
if types.TypeID(p.ValType) == types.GeoID {
x.Check2(buf.WriteString(fmt.Sprintf("^^<geo:geojson> ")))
// TODO(tzdybal) - uncomment
/* } else if len(p.Lang) > 0 {
x.Check2(buf.WriteString(fmt.Sprintf("@%s", p.Lang))) */
} else if types.TypeID(p.ValType) == types.PasswordID {
x.Check2(buf.WriteString(fmt.Sprintf("^^<pwd:%s>", vID.Name())))
} else if types.TypeID(p.ValType) != types.BinaryID &&
types.TypeID(p.ValType) != types.DefaultID {
x.Check2(buf.WriteString(fmt.Sprintf("^^<xs:%s>", vID.Name())))
}
x.Check2(buf.WriteString(" .\n"))
return
}
x.Check2(buf.WriteString(fmt.Sprintf("<%#x> .\n", p.Uid)))
}
}
func writeToFile(fpath string, ch chan []byte) error {
f, err := os.Create(fpath)
if err != nil {
return err
}
defer f.Close()
x.Check(err)
w := bufio.NewWriterSize(f, 1000000)
gw, err := gzip.NewWriterLevel(w, gzip.BestCompression)
if err != nil {
return err
}
for buf := range ch {
if _, err := gw.Write(buf); err != nil {
return err
}
}
if err := gw.Flush(); err != nil {
return err
}
if err := gw.Close(); err != nil {
return err
}
return w.Flush()
}
// Backup creates a backup of data by exporting it as an RDF gzip.
func backup(gid uint32, bdir string) error {
// Use a goroutine to write to file.
err := os.MkdirAll(bdir, 0700)
if err != nil {
return err
}
fpath := path.Join(bdir, fmt.Sprintf("dgraph-%d-%s.rdf.gz", gid,
time.Now().Format("2006-01-02-15-04")))
fmt.Printf("Backing up at: %v\n", fpath)
chb := make(chan []byte, 1000)
errChan := make(chan error, 1)
go func() {
errChan <- writeToFile(fpath, chb)
}()
// Use a bunch of goroutines to convert to RDF format.
chkv := make(chan kv, 1000)
var wg sync.WaitGroup
wg.Add(numBackupRoutines)
for i := 0; i < numBackupRoutines; i++ {
go func() {
buf := new(bytes.Buffer)
buf.Grow(50000)
for item := range chkv {
toRDF(buf, item)
if buf.Len() >= 40000 {
tmp := make([]byte, buf.Len())
copy(tmp, buf.Bytes())
chb <- tmp
buf.Reset()
}
}
if buf.Len() > 0 {
tmp := make([]byte, buf.Len())
copy(tmp, buf.Bytes())
chb <- tmp
}
wg.Done()
}()
}
// Iterate over rocksdb.
it := pstore.NewIterator()
defer it.Close()
var lastPred string
for it.SeekToFirst(); it.Valid(); {
key := it.Key().Data()
pk := x.Parse(key)
if pk.IsIndex() {
// Seek to the end of index keys.
it.Seek(pk.SkipRangeOfSameType())
continue
}
if pk.IsReverse() {
// Seek to the end of reverse keys.
it.Seek(pk.SkipRangeOfSameType())
continue
}
if pk.Attr == "_uid_" {
// Skip the UID mappings.
it.Seek(pk.SkipPredicate())
continue
}
if pk.IsSchema() {
// index values are the last keys currently
it.Seek(pk.SkipSchema())
continue
}
x.AssertTrue(pk.IsData())
pred, uid := pk.Attr, pk.Uid
if pred != lastPred && group.BelongsTo(pred) != gid {
it.Seek(pk.SkipPredicate())
continue
}
prefix := fmt.Sprintf("<%#x> <%s> ", uid, pred)
pl := &types.PostingList{}
x.Check(pl.Unmarshal(it.Value().Data()))
chkv <- kv{
prefix: prefix,
list: pl,
}
lastPred = pred
it.Next()
}
close(chkv) // We have stopped output to chkv.
wg.Wait() // Wait for numBackupRoutines to finish.
close(chb) // We have stopped output to chb.
err = <-errChan
return err
}
func handleBackupForGroup(ctx context.Context, reqId uint64, gid uint32) *BackupPayload {
n := groups().Node(gid)
if n.AmLeader() {
x.Trace(ctx, "Leader of group: %d. Running backup.", gid)
if err := backup(gid, *backupPath); err != nil {
x.TraceError(ctx, err)
return &BackupPayload{
ReqId: reqId,
Status: BackupPayload_FAILED,
}
}
x.Trace(ctx, "Backup done for group: %d.", gid)
return &BackupPayload{
ReqId: reqId,
Status: BackupPayload_SUCCESS,
GroupId: gid,
}
}
// I'm not the leader. Relay to someone who I think is.
var addrs []string
{
// Try in order: leader of given group, any server from given group, leader of group zero.
_, addr := groups().Leader(gid)
addrs = append(addrs, addr)
addrs = append(addrs, groups().AnyServer(gid))
_, addr = groups().Leader(0)
addrs = append(addrs, addr)
}
var conn *grpc.ClientConn
for _, addr := range addrs {
pl := pools().get(addr)
var err error
conn, err = pl.Get()
if err == nil {
x.Trace(ctx, "Relaying backup request for group %d to %q", gid, pl.Addr)
defer pl.Put(conn)
break
}
x.TraceError(ctx, err)
}
// Unable to find any connection to any of these servers. This should be exceedingly rare.
// But probably not worthy of crashing the server. We can just skip the backup.
if conn == nil {
x.Trace(ctx, fmt.Sprintf("Unable to find a server to backup group: %d", gid))
return &BackupPayload{
ReqId: reqId,
Status: BackupPayload_FAILED,
GroupId: gid,
}
}
c := NewWorkerClient(conn)
nr := &BackupPayload{
ReqId: reqId,
GroupId: gid,
}
nrep, err := c.Backup(ctx, nr)
if err != nil {
x.TraceError(ctx, err)
return &BackupPayload{
ReqId: reqId,
Status: BackupPayload_FAILED,
GroupId: gid,
}
}
return nrep
}
// Backup request is used to trigger backups for the request list of groups.
// If a server receives request to backup a group that it doesn't handle, it would
// automatically relay that request to the server that it thinks should handle the request.
func (w *grpcWorker) Backup(ctx context.Context, req *BackupPayload) (*BackupPayload, error) {
reply := &BackupPayload{ReqId: req.ReqId}
reply.Status = BackupPayload_FAILED // Set by default.
if ctx.Err() != nil {
return reply, ctx.Err()
}
if !w.addIfNotPresent(req.ReqId) {
reply.Status = BackupPayload_DUPLICATE
return reply, nil
}
chb := make(chan *BackupPayload, 1)
go func() {
chb <- handleBackupForGroup(ctx, req.ReqId, req.GroupId)
}()
select {
case rep := <-chb:
return rep, nil
case <-ctx.Done():
return reply, ctx.Err()
}
}
func BackupOverNetwork(ctx context.Context) error {
// If we haven't even had a single membership update, don't run backup.
if len(*peerAddr) > 0 && groups().LastUpdate() == 0 {
x.Trace(ctx, "This server hasn't yet been fully initiated. Please retry later.")
return x.Errorf("Uninitiated server. Please retry later")
}
// Let's first collect all groups.
gids := groups().KnownGroups()
ch := make(chan *BackupPayload, len(gids))
for _, gid := range gids {
go func(group uint32) {
reqId := uint64(rand.Int63())
ch <- handleBackupForGroup(ctx, reqId, group)
}(gid)
}
for i := 0; i < len(gids); i++ {
bp := <-ch
if bp.Status != BackupPayload_SUCCESS {
x.Trace(ctx, "Backup status: %v for group id: %d", bp.Status, bp.GroupId)
return fmt.Errorf("Backup status: %v for group id: %d", bp.Status, bp.GroupId)
} else {
x.Trace(ctx, "Backup successful for group: %v", bp.GroupId)
}
}
x.Trace(ctx, "DONE backup")
return nil
}