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
|
package shell
import (
"bytes"
"context"
"flag"
"fmt"
"github.com/chrislusf/seaweedfs/weed/operation"
"github.com/chrislusf/seaweedfs/weed/pb"
"github.com/chrislusf/seaweedfs/weed/pb/volume_server_pb"
"github.com/chrislusf/seaweedfs/weed/storage/needle_map"
"golang.org/x/exp/slices"
"io"
"math"
)
func init() {
Commands = append(Commands, &commandVolumeCheckDisk{})
}
type commandVolumeCheckDisk struct {
env *CommandEnv
}
func (c *commandVolumeCheckDisk) Name() string {
return "volume.check.disk"
}
func (c *commandVolumeCheckDisk) Help() string {
return `check all replicated volumes to find and fix inconsistencies. It is optional and resource intensive.
How it works:
find all volumes that are replicated
for each volume id, if there are more than 2 replicas, find one pair with the largest 2 in file count.
for the pair volume A and B
append entries in A and not in B to B
append entries in B and not in A to A
`
}
func (c *commandVolumeCheckDisk) Do(args []string, commandEnv *CommandEnv, writer io.Writer) (err error) {
fsckCommand := flag.NewFlagSet(c.Name(), flag.ContinueOnError)
slowMode := fsckCommand.Bool("slow", false, "slow mode checks all replicas even file counts are the same")
verbose := fsckCommand.Bool("v", false, "verbose mode")
volumeId := fsckCommand.Uint("volumeId", 0, "the volume id")
applyChanges := fsckCommand.Bool("force", false, "apply the fix")
nonRepairThreshold := fsckCommand.Float64("nonRepairThreshold", 0.3, "repair when missing keys is not more than this limit")
if err = fsckCommand.Parse(args); err != nil {
return nil
}
infoAboutSimulationMode(writer, *applyChanges, "-force")
if err = commandEnv.confirmIsLocked(args); err != nil {
return
}
c.env = commandEnv
// collect topology information
topologyInfo, _, err := collectTopologyInfo(commandEnv, 0)
if err != nil {
return err
}
volumeReplicas, _ := collectVolumeReplicaLocations(topologyInfo)
// pick 1 pairs of volume replica
fileCount := func(replica *VolumeReplica) uint64 {
return replica.info.FileCount - replica.info.DeleteCount
}
for _, replicas := range volumeReplicas {
if *volumeId > 0 && replicas[0].info.Id != uint32(*volumeId) {
continue
}
slices.SortFunc(replicas, func(a, b *VolumeReplica) bool {
return fileCount(a) > fileCount(b)
})
for len(replicas) >= 2 {
a, b := replicas[0], replicas[1]
if !*slowMode {
if fileCount(a) == fileCount(b) {
replicas = replicas[1:]
continue
}
}
if a.info.ReadOnly || b.info.ReadOnly {
fmt.Fprintf(writer, "skipping readonly volume %d on %s and %s\n", a.info.Id, a.location.dataNode.Id, b.location.dataNode.Id)
replicas = replicas[1:]
continue
}
if err := c.syncTwoReplicas(a, b, *applyChanges, *nonRepairThreshold, *verbose, writer); err != nil {
fmt.Fprintf(writer, "sync volume %d on %s and %s: %v\n", a.info.Id, a.location.dataNode.Id, b.location.dataNode.Id, err)
}
replicas = replicas[1:]
}
}
return nil
}
func (c *commandVolumeCheckDisk) syncTwoReplicas(a *VolumeReplica, b *VolumeReplica, applyChanges bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (err error) {
aHasChanges, bHasChanges := true, true
for aHasChanges || bHasChanges {
if aHasChanges, bHasChanges, err = c.checkBoth(a, b, applyChanges, nonRepairThreshold, verbose, writer); err != nil {
return err
}
}
return nil
}
func (c *commandVolumeCheckDisk) checkBoth(a *VolumeReplica, b *VolumeReplica, applyChanges bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (aHasChanges bool, bHasChanges bool, err error) {
aDB, bDB := needle_map.NewMemDb(), needle_map.NewMemDb()
defer func() {
aDB.Close()
bDB.Close()
}()
// read index db
if err = c.readIndexDatabase(aDB, a.info.Collection, a.info.Id, pb.NewServerAddressFromDataNode(a.location.dataNode), verbose, writer); err != nil {
return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", a.location.dataNode, a.info.Id, err)
}
if err := c.readIndexDatabase(bDB, b.info.Collection, b.info.Id, pb.NewServerAddressFromDataNode(b.location.dataNode), verbose, writer); err != nil {
return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", b.location.dataNode, b.info.Id, err)
}
// find and make up the differences
if aHasChanges, err = c.doVolumeCheckDisk(bDB, aDB, b, a, verbose, writer, applyChanges, nonRepairThreshold); err != nil {
return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", b.location.dataNode, a.location.dataNode, b.info.Id, err)
}
if bHasChanges, err = c.doVolumeCheckDisk(aDB, bDB, a, b, verbose, writer, applyChanges, nonRepairThreshold); err != nil {
return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", a.location.dataNode, b.location.dataNode, a.info.Id, err)
}
return
}
func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_map.MemDb, source, target *VolumeReplica, verbose bool, writer io.Writer, applyChanges bool, nonRepairThreshold float64) (hasChanges bool, err error) {
// find missing keys
// hash join, can be more efficient
var missingNeedles []needle_map.NeedleValue
var counter int
minuend.AscendingVisit(func(value needle_map.NeedleValue) error {
counter++
if _, found := subtrahend.Get(value.Key); !found {
missingNeedles = append(missingNeedles, value)
}
return nil
})
fmt.Fprintf(writer, "volume %d %s has %d entries, %s missed %d entries\n", source.info.Id, source.location.dataNode.Id, counter, target.location.dataNode.Id, len(missingNeedles))
if counter == 0 || len(missingNeedles) == 0 {
return false, nil
}
missingNeedlesFraction := float64(len(missingNeedles)) / float64(counter)
if missingNeedlesFraction > nonRepairThreshold {
return false, fmt.Errorf(
"failed to start repair volume %d, percentage of missing keys is greater than the threshold: %.2f > %.2f",
source.info.Id, missingNeedlesFraction, nonRepairThreshold)
}
for _, needleValue := range missingNeedles {
needleBlob, err := c.readSourceNeedleBlob(pb.NewServerAddressFromDataNode(source.location.dataNode), source.info.Id, needleValue)
if err != nil {
return hasChanges, err
}
if !applyChanges {
continue
}
if verbose {
fmt.Fprintf(writer, "read %d,%x %s => %s \n", source.info.Id, needleValue.Key, source.location.dataNode.Id, target.location.dataNode.Id)
}
hasChanges = true
if err = c.writeNeedleBlobToTarget(pb.NewServerAddressFromDataNode(target.location.dataNode), source.info.Id, needleValue, needleBlob); err != nil {
return hasChanges, err
}
}
return
}
func (c *commandVolumeCheckDisk) readSourceNeedleBlob(sourceVolumeServer pb.ServerAddress, volumeId uint32, needleValue needle_map.NeedleValue) (needleBlob []byte, err error) {
err = operation.WithVolumeServerClient(false, sourceVolumeServer, c.env.option.GrpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
resp, err := client.ReadNeedleBlob(context.Background(), &volume_server_pb.ReadNeedleBlobRequest{
VolumeId: volumeId,
NeedleId: uint64(needleValue.Key),
Offset: needleValue.Offset.ToActualOffset(),
Size: int32(needleValue.Size),
})
if err != nil {
return err
}
needleBlob = resp.NeedleBlob
return nil
})
return
}
func (c *commandVolumeCheckDisk) writeNeedleBlobToTarget(targetVolumeServer pb.ServerAddress, volumeId uint32, needleValue needle_map.NeedleValue, needleBlob []byte) error {
return operation.WithVolumeServerClient(false, targetVolumeServer, c.env.option.GrpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
_, err := client.WriteNeedleBlob(context.Background(), &volume_server_pb.WriteNeedleBlobRequest{
VolumeId: volumeId,
NeedleId: uint64(needleValue.Key),
Size: int32(needleValue.Size),
NeedleBlob: needleBlob,
})
return err
})
}
func (c *commandVolumeCheckDisk) readIndexDatabase(db *needle_map.MemDb, collection string, volumeId uint32, volumeServer pb.ServerAddress, verbose bool, writer io.Writer) error {
var buf bytes.Buffer
if err := c.copyVolumeIndexFile(collection, volumeId, volumeServer, &buf, verbose, writer); err != nil {
return err
}
if verbose {
fmt.Fprintf(writer, "load collection %s volume %d index size %d from %s ...\n", collection, volumeId, buf.Len(), volumeServer)
}
return db.LoadFromReaderAt(bytes.NewReader(buf.Bytes()))
}
func (c *commandVolumeCheckDisk) copyVolumeIndexFile(collection string, volumeId uint32, volumeServer pb.ServerAddress, buf *bytes.Buffer, verbose bool, writer io.Writer) error {
return operation.WithVolumeServerClient(true, volumeServer, c.env.option.GrpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
ext := ".idx"
copyFileClient, err := volumeServerClient.CopyFile(context.Background(), &volume_server_pb.CopyFileRequest{
VolumeId: volumeId,
Ext: ".idx",
CompactionRevision: math.MaxUint32,
StopOffset: math.MaxInt64,
Collection: collection,
IsEcVolume: false,
IgnoreSourceFileNotFound: false,
})
if err != nil {
return fmt.Errorf("failed to start copying volume %d%s: %v", volumeId, ext, err)
}
err = writeToBuffer(copyFileClient, buf)
if err != nil {
return fmt.Errorf("failed to copy %d%s from %s: %v", volumeId, ext, volumeServer, err)
}
return nil
})
}
func writeToBuffer(client volume_server_pb.VolumeServer_CopyFileClient, buf *bytes.Buffer) error {
for {
resp, receiveErr := client.Recv()
if receiveErr == io.EOF {
break
}
if receiveErr != nil {
return fmt.Errorf("receiving: %v", receiveErr)
}
buf.Write(resp.FileContent)
}
return nil
}
|
