-
Notifications
You must be signed in to change notification settings - Fork 23
/
conn.go
474 lines (436 loc) · 14.8 KB
/
conn.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
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
// Copyright 2023 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.
//
// SPDX-License-Identifier: Apache-2.0
// Package pglogical contains support for reading a PostgreSQL logical
// replication feed.
package pglogical
import (
"context"
"database/sql"
"encoding/json"
"fmt"
"time"
"github.com/cockroachdb/replicator/internal/script"
"github.com/cockroachdb/replicator/internal/types"
"github.com/cockroachdb/replicator/internal/util/hlc"
"github.com/cockroachdb/replicator/internal/util/ident"
"github.com/cockroachdb/replicator/internal/util/notify"
"github.com/cockroachdb/replicator/internal/util/stopper"
"github.com/cockroachdb/replicator/internal/util/stopvar"
"github.com/google/uuid"
"github.com/jackc/pglogrepl"
"github.com/jackc/pgx/v5/pgconn"
"github.com/jackc/pgx/v5/pgproto3"
"github.com/pkg/errors"
log "github.com/sirupsen/logrus"
)
// A Conn encapsulates all wire-connection behavior. It is
// responsible for receiving replication messages and replying with
// status updates.
type Conn struct {
// The destination for writes.
acceptor types.MultiAcceptor
// Columns, as ordered by the source database.
columns *ident.TableMap[[]types.ColData]
// The time of the transaction being received.
nextCommitTime time.Time
// Persistent storage for WAL data.
memo types.Memo
// The pg publication name to subscribe to.
publicationName string
// Map source ids to target tables.
relations map[uint32]ident.Table
// The name of the slot within the publication.
slotName string
// The configuration for opening replication connections.
sourceConfig *pgconn.Config
// How ofter to commit the consistent point
standbyTimeout time.Duration
// Access to the staging cluster.
stagingDB *types.StagingPool
// The destination for writes.
target ident.Schema
// Access to the target database.
targetDB *types.TargetPool
// Support for toasted columns
toastedColumns bool
// Managed by persistWALOffset.
walOffset notify.Var[pglogrepl.LSN]
}
// Start launches goroutines into the context.
func (c *Conn) Start(ctx *stopper.Context) error {
// Call this first to load the previous offset. We want to reset our
// state before starting the main copier routine.
if err := c.persistWALOffset(ctx); err != nil {
return err
}
// Start a process to copy data to the target.
ctx.Go(func() error {
for !ctx.IsStopping() {
if err := c.copyMessages(ctx); err != nil {
log.WithError(err).Warn("error while copying messages; will retry")
select {
case <-ctx.Stopping():
case <-time.After(100 * time.Millisecond):
}
}
}
return nil
})
return nil
}
// accumulateBatch folds replication messages into the batch and sends it to
// the acceptor when a complete transaction has been read. The returned
// batch should be passed to the next invocation of accumulateBatch.
func (c *Conn) accumulateBatch(
ctx *stopper.Context, msg pglogrepl.Message, batch *types.MultiBatch,
) (*types.MultiBatch, error) {
log.Tracef("message %T", msg)
switch msg := msg.(type) {
case *pglogrepl.RelationMessage:
// The replication protocol says that we'll see these
// descriptors before any use of the relation id in the
// stream. We'll map the int value to our table identifiers.
c.onRelation(msg)
return batch, nil
case *pglogrepl.BeginMessage:
c.nextCommitTime = msg.CommitTime
// Create a new batch to accumulate into. It may be discarded
// later if the timestamp precedes the latest commit.
return &types.MultiBatch{}, nil
case *pglogrepl.CommitMessage:
// We rely on the upstream database to replay events in the case of
// errors, so we may receive events that we've already processed.
// We use the COMMIT LSN offset as the consistent point, which is
// written in the WAL synchronously with the upstream transaction.
ignoreLSN, _ := c.walOffset.Get()
if msg.CommitLSN <= ignoreLSN {
// Just discard the entire batch in this case.
log.Tracef("ignoring CommitMessage at %s before %s",
msg.CommitLSN, ignoreLSN)
return nil, nil
}
// In Postgres version < v15, the stream might contain empty transactions.
// See https://github.com/postgres/postgres/commit/d5a9d86d8f
// We will skip them to avoid unnecessary writes to the memo table.
if batch.Count() == 0 {
emptyTransactionCount.Inc()
log.Trace("skipping empty transaction")
} else {
tx, err := c.targetDB.BeginTx(ctx, &sql.TxOptions{})
if err != nil {
return nil, errors.WithStack(err)
}
defer tx.Rollback()
if err := c.acceptor.AcceptMultiBatch(ctx, batch, &types.AcceptOptions{
TargetQuerier: tx,
}); err != nil {
return nil, err
}
if err := tx.Commit(); err != nil {
return nil, errors.WithStack(err)
}
}
c.walOffset.Set(msg.CommitLSN)
return nil, nil
case *pglogrepl.DeleteMessage:
return batch, c.onDataTuple(batch, msg.RelationID, msg.OldTuple, true /* isDelete */)
case *pglogrepl.InsertMessage:
return batch, c.onDataTuple(batch, msg.RelationID, msg.Tuple, false /* isDelete */)
case *pglogrepl.UpdateMessage:
return batch, c.onDataTuple(batch, msg.RelationID, msg.NewTuple, false /* isDelete */)
case *pglogrepl.TruncateMessage:
return nil, errors.Errorf("the TRUNCATE operation cannot be supported on table %d", msg.RelationNum)
case *pglogrepl.TypeMessage:
// This type is intentionally discarded. We interpret the
// type of the data based on the target table, not the
// source.
return batch, nil
default:
return nil, errors.Errorf("unimplemented logical replication message %T", msg)
}
}
// copyMessages is the main replication loop. It will open a connection
// to the source, accumulate messages, and commit data to the target.
func (c *Conn) copyMessages(ctx *stopper.Context) error {
replConn, err := pgconn.ConnectConfig(ctx, c.sourceConfig)
if err != nil {
return errors.WithStack(err)
}
defer replConn.Close(context.Background())
startLogPos, _ := c.walOffset.Get()
if err := pglogrepl.StartReplication(ctx,
replConn, c.slotName, startLogPos,
pglogrepl.StartReplicationOptions{
PluginArgs: []string{
"proto_version '1'",
fmt.Sprintf("publication_names '%s'", c.publicationName)},
},
); err != nil {
dialFailureCount.Inc()
return errors.WithStack(err)
}
dialSuccessCount.Inc()
var batch *types.MultiBatch
standbyDeadline := time.Now().Add(c.standbyTimeout)
for !ctx.IsStopping() {
// Occasionally send updates back to the server so it will
// remember our WAL offset.
if time.Now().After(standbyDeadline) {
standbyDeadline = time.Now().Add(c.standbyTimeout)
lsn, _ := c.walOffset.Get()
if err := pglogrepl.SendStandbyStatusUpdate(ctx, replConn,
pglogrepl.StandbyStatusUpdate{
WALWritePosition: lsn,
},
); err != nil {
return errors.WithStack(err)
}
log.WithField("WALWritePosition", lsn).Trace("sent Standby status message")
}
// Receive one message, with a timeout. In a low-traffic
// situation, we want to ensure that we're sending heartbeats
// back to the source server.
receiveCtx, cancel := context.WithDeadline(ctx, standbyDeadline)
msg, err := replConn.ReceiveMessage(receiveCtx)
cancel()
if err != nil {
if pgconn.Timeout(err) {
continue
}
return errors.WithStack(err)
}
log.Tracef("received %T", msg)
switch msg := msg.(type) {
case *pgproto3.CopyData:
switch msg.Data[0] {
case pglogrepl.PrimaryKeepaliveMessageByteID:
// The server is sending us a keepalive message. This is
// informational, except in the case where an immediate
// acknowledgement is requested. In that case, we'll
// reset the standby deadline to zero, so we kick back a
// message at the top of the loop.
pkm, err := pglogrepl.ParsePrimaryKeepaliveMessage(msg.Data[1:])
if err != nil {
return errors.WithStack(err)
}
log.WithFields(log.Fields{
"ServerWALEnd": pkm.ServerWALEnd,
"ServerTime": pkm.ServerTime,
"ReplyRequested": pkm.ReplyRequested,
}).Debug("primary keepalive received")
if pkm.ReplyRequested {
standbyDeadline = time.Time{}
}
case pglogrepl.XLogDataByteID:
// This is where things get interesting. We have actual
// transaction log data to parse into messages. These
// messages get handed off to the consumer via the
// channel passed in.
xld, err := pglogrepl.ParseXLogData(msg.Data[1:])
if err != nil {
return errors.WithStack(err)
}
log.WithFields(log.Fields{
"ByteCount": len(xld.WALData),
"ServerWALEnd": xld.ServerWALEnd,
"ServerTime": xld.ServerTime,
"WALStart": xld.WALStart,
}).Debug("xlog data")
logicalMsg, err := pglogrepl.Parse(xld.WALData)
if err != nil {
return errors.WithStack(err)
}
log.WithFields(log.Fields{
"logicalMsg": logicalMsg.Type().String(),
}).Debug("xlog data")
// Update our accumulator with the received message.
batch, err = c.accumulateBatch(ctx, logicalMsg, batch)
if err != nil {
return err
}
}
case *pgproto3.NotificationResponse:
log.Debugf("notification from server: %s", msg.Payload)
default:
log.Debugf("unexpected payload message: %T", msg)
}
}
return nil
}
// decodeMutation converts the incoming tuple data into a Mutation.
func (c *Conn) decodeMutation(
tbl ident.Table, data *pglogrepl.TupleData, isDelete bool,
) (types.Mutation, error) {
var mut types.Mutation
var key []string
enc := make(map[string]any)
targetCols, ok := c.columns.Get(tbl)
if !ok {
return mut, errors.Errorf("no column data for %s", tbl)
}
if len(targetCols) != len(data.Columns) {
return mut, errors.Errorf("column count mismatch is %s: %d vs %d",
tbl, len(targetCols), len(data.Columns))
}
for idx, sourceCol := range data.Columns {
targetCol := targetCols[idx]
switch sourceCol.DataType {
case pglogrepl.TupleDataTypeNull:
enc[targetCol.Name.Raw()] = nil
case pglogrepl.TupleDataTypeText:
// The incoming data is in a textual format.
enc[targetCol.Name.Raw()] = string(sourceCol.Data)
if targetCol.Primary {
key = append(key, string(sourceCol.Data))
}
case pglogrepl.TupleDataTypeToast:
if c.toastedColumns {
// TupleDataTypeToast is just a marker that tells us
// that a TOASTed column has not changed.
// Putting a placeholders for downstream apply handlers,
// and a custom template that will be able to handle it.
unchangedToastedColumns.Inc()
if mut.Meta == nil {
mut.Meta = make(map[string]any)
}
mut.Meta[types.CustomUpsert] = "toasted"
enc[targetCol.Name.Raw()] = types.ToastedColumnPlaceholder
continue
}
return mut, errors.Errorf(
"TOASTed columns are not supported in %s.%s", tbl, targetCol.Name)
default:
return mut, errors.Errorf(
"unimplemented tuple data type %q", string(sourceCol.DataType))
}
}
// In the pathological case where a table has no primary key, we'll
// generate a random uuid value to use as the staging key. This is
// fine, because the underlying data has no particular identity to
// update. In fact, it's not possible to issue an UPDATE to Postgres
// when a row has no replication identity.
if len(key) == 0 {
key = []string{uuid.New().String()}
}
var err error
mut.Key, err = json.Marshal(key)
if err != nil {
return mut, errors.WithStack(err)
}
// We don't need the actual column data for delete operations.
if !isDelete {
mut.Data, err = json.Marshal(enc)
if err != nil {
return mut, errors.WithStack(err)
}
}
return mut, errors.WithStack(err)
}
// onDataTuple will add an incoming row tuple to the in-memory slice,
// possibly flushing it when the batch size limit is reached.
func (c *Conn) onDataTuple(
batch *types.MultiBatch, relation uint32, tuple *pglogrepl.TupleData, isDelete bool,
) error {
if batch == nil {
log.Trace("ignoring replayed message")
return nil
}
traceTuple(tuple)
tbl, ok := c.relations[relation]
if !ok {
return errors.Errorf("unknown relation id %d", relation)
}
mut, err := c.decodeMutation(tbl, tuple, isDelete)
if err != nil {
return err
}
// Set an approximate timestamp.
mut.Time = hlc.New(c.nextCommitTime.UnixNano(), 0)
// Set script metadata, which will be acted on by the acceptor.
script.AddMeta("pglogical", tbl, &mut)
return batch.Accumulate(tbl, mut)
}
// learn updates the source database namespace mappings.
func (c *Conn) onRelation(msg *pglogrepl.RelationMessage) {
// The replication protocol says that we'll see these
// descriptors before any use of the relation id in the
// stream. We'll map the int value to our table identifiers.
tbl := ident.NewTable(c.target, ident.New(msg.RelationName))
c.relations[msg.RelationID] = tbl
colNames := make([]types.ColData, len(msg.Columns))
for idx, col := range msg.Columns {
colNames[idx] = types.ColData{
Name: ident.New(col.Name),
Primary: col.Flags == 1,
// This could be made textual if we used the
// ConnInfo metadata methods.
Type: fmt.Sprintf("%d", col.DataType),
}
}
c.columns.Put(tbl, colNames)
log.WithFields(log.Fields{
"Columns": colNames,
"RelationID": msg.RelationID,
"Table": tbl,
}).Trace("learned relation")
}
// persistWALOffset loads an existing value from memo into walOffset. It
// will also start a goroutine in the stopper to occasionally write an
// updated value back to the memo.
func (c *Conn) persistWALOffset(ctx *stopper.Context) error {
key := fmt.Sprintf("pglogical-wal-offset-%s", c.target.Raw())
found, err := c.memo.Get(ctx, c.stagingDB, key)
if err != nil {
return err
}
var lsn pglogrepl.LSN
if len(found) > 0 {
if err := lsn.Scan(found); err != nil {
return errors.WithStack(err)
}
c.walOffset.Set(lsn)
}
ctx.Go(func() error {
_, err := stopvar.DoWhenChanged(ctx, lsn, &c.walOffset,
func(ctx *stopper.Context, _, lsn pglogrepl.LSN) error {
if err := c.memo.Put(ctx, c.stagingDB, key, []byte(lsn.String())); err == nil {
log.Tracef("stored WAL offset %s: %s", key, lsn)
} else {
log.WithError(err).Warn("could not persist LSN offset")
}
return nil
})
return err
})
return nil
}
// traceTuple emits log messages if tracing is enabled.
func traceTuple(t *pglogrepl.TupleData) {
if !log.IsLevelEnabled(log.TraceLevel) {
return
}
if t == nil {
log.Trace("NIL TUPLE")
return
}
s := make([]string, len(t.Columns))
for idx, data := range t.Columns {
if data.DataType == pglogrepl.TupleDataTypeText {
s[idx] = string(data.Data)
}
}
log.WithField("data", s).Trace("values")
}