forked from grpc/grpc-go
-
Notifications
You must be signed in to change notification settings - Fork 0
/
stream.go
1823 lines (1683 loc) · 55.1 KB
/
stream.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
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
*
* Copyright 2014 gRPC 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.
*
*/
package grpc
import (
"context"
"errors"
"io"
"math"
rand "math/rand/v2"
"strconv"
"sync"
"time"
"google.golang.org/grpc/balancer"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/encoding"
"google.golang.org/grpc/internal"
"google.golang.org/grpc/internal/balancerload"
"google.golang.org/grpc/internal/binarylog"
"google.golang.org/grpc/internal/channelz"
"google.golang.org/grpc/internal/grpcutil"
imetadata "google.golang.org/grpc/internal/metadata"
iresolver "google.golang.org/grpc/internal/resolver"
"google.golang.org/grpc/internal/serviceconfig"
istatus "google.golang.org/grpc/internal/status"
"google.golang.org/grpc/internal/transport"
"google.golang.org/grpc/mem"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/peer"
"google.golang.org/grpc/stats"
"google.golang.org/grpc/status"
)
var metadataFromOutgoingContextRaw = internal.FromOutgoingContextRaw.(func(context.Context) (metadata.MD, [][]string, bool))
// StreamHandler defines the handler called by gRPC server to complete the
// execution of a streaming RPC.
//
// If a StreamHandler returns an error, it should either be produced by the
// status package, or be one of the context errors. Otherwise, gRPC will use
// codes.Unknown as the status code and err.Error() as the status message of the
// RPC.
type StreamHandler func(srv any, stream ServerStream) error
// StreamDesc represents a streaming RPC service's method specification. Used
// on the server when registering services and on the client when initiating
// new streams.
type StreamDesc struct {
// StreamName and Handler are only used when registering handlers on a
// server.
StreamName string // the name of the method excluding the service
Handler StreamHandler // the handler called for the method
// ServerStreams and ClientStreams are used for registering handlers on a
// server as well as defining RPC behavior when passed to NewClientStream
// and ClientConn.NewStream. At least one must be true.
ServerStreams bool // indicates the server can perform streaming sends
ClientStreams bool // indicates the client can perform streaming sends
}
// Stream defines the common interface a client or server stream has to satisfy.
//
// Deprecated: See ClientStream and ServerStream documentation instead.
type Stream interface {
// Deprecated: See ClientStream and ServerStream documentation instead.
Context() context.Context
// Deprecated: See ClientStream and ServerStream documentation instead.
SendMsg(m any) error
// Deprecated: See ClientStream and ServerStream documentation instead.
RecvMsg(m any) error
}
// ClientStream defines the client-side behavior of a streaming RPC.
//
// All errors returned from ClientStream methods are compatible with the
// status package.
type ClientStream interface {
// Header returns the header metadata received from the server if there
// is any. It blocks if the metadata is not ready to read. If the metadata
// is nil and the error is also nil, then the stream was terminated without
// headers, and the status can be discovered by calling RecvMsg.
Header() (metadata.MD, error)
// Trailer returns the trailer metadata from the server, if there is any.
// It must only be called after stream.CloseAndRecv has returned, or
// stream.Recv has returned a non-nil error (including io.EOF).
Trailer() metadata.MD
// CloseSend closes the send direction of the stream. It closes the stream
// when non-nil error is met. It is also not safe to call CloseSend
// concurrently with SendMsg.
CloseSend() error
// Context returns the context for this stream.
//
// It should not be called until after Header or RecvMsg has returned. Once
// called, subsequent client-side retries are disabled.
Context() context.Context
// SendMsg is generally called by generated code. On error, SendMsg aborts
// the stream. If the error was generated by the client, the status is
// returned directly; otherwise, io.EOF is returned and the status of
// the stream may be discovered using RecvMsg. For unary or server-streaming
// RPCs (StreamDesc.ClientStreams is false), a nil error is returned
// unconditionally.
//
// SendMsg blocks until:
// - There is sufficient flow control to schedule m with the transport, or
// - The stream is done, or
// - The stream breaks.
//
// SendMsg does not wait until the message is received by the server. An
// untimely stream closure may result in lost messages. To ensure delivery,
// users should ensure the RPC completed successfully using RecvMsg.
//
// It is safe to have a goroutine calling SendMsg and another goroutine
// calling RecvMsg on the same stream at the same time, but it is not safe
// to call SendMsg on the same stream in different goroutines. It is also
// not safe to call CloseSend concurrently with SendMsg.
//
// It is not safe to modify the message after calling SendMsg. Tracing
// libraries and stats handlers may use the message lazily.
SendMsg(m any) error
// RecvMsg blocks until it receives a message into m or the stream is
// done. It returns io.EOF when the stream completes successfully. On
// any other error, the stream is aborted and the error contains the RPC
// status.
//
// It is safe to have a goroutine calling SendMsg and another goroutine
// calling RecvMsg on the same stream at the same time, but it is not
// safe to call RecvMsg on the same stream in different goroutines.
RecvMsg(m any) error
}
// NewStream creates a new Stream for the client side. This is typically
// called by generated code. ctx is used for the lifetime of the stream.
//
// To ensure resources are not leaked due to the stream returned, one of the following
// actions must be performed:
//
// 1. Call Close on the ClientConn.
// 2. Cancel the context provided.
// 3. Call RecvMsg until a non-nil error is returned. A protobuf-generated
// client-streaming RPC, for instance, might use the helper function
// CloseAndRecv (note that CloseSend does not Recv, therefore is not
// guaranteed to release all resources).
// 4. Receive a non-nil, non-io.EOF error from Header or SendMsg.
//
// If none of the above happen, a goroutine and a context will be leaked, and grpc
// will not call the optionally-configured stats handler with a stats.End message.
func (cc *ClientConn) NewStream(ctx context.Context, desc *StreamDesc, method string, opts ...CallOption) (ClientStream, error) {
// allow interceptor to see all applicable call options, which means those
// configured as defaults from dial option as well as per-call options
opts = combine(cc.dopts.callOptions, opts)
if cc.dopts.streamInt != nil {
return cc.dopts.streamInt(ctx, desc, cc, method, newClientStream, opts...)
}
return newClientStream(ctx, desc, cc, method, opts...)
}
// NewClientStream is a wrapper for ClientConn.NewStream.
func NewClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, opts ...CallOption) (ClientStream, error) {
return cc.NewStream(ctx, desc, method, opts...)
}
func newClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, opts ...CallOption) (_ ClientStream, err error) {
// Start tracking the RPC for idleness purposes. This is where a stream is
// created for both streaming and unary RPCs, and hence is a good place to
// track active RPC count.
if err := cc.idlenessMgr.OnCallBegin(); err != nil {
return nil, err
}
// Add a calloption, to decrement the active call count, that gets executed
// when the RPC completes.
opts = append([]CallOption{OnFinish(func(error) { cc.idlenessMgr.OnCallEnd() })}, opts...)
if md, added, ok := metadataFromOutgoingContextRaw(ctx); ok {
// validate md
if err := imetadata.Validate(md); err != nil {
return nil, status.Error(codes.Internal, err.Error())
}
// validate added
for _, kvs := range added {
for i := 0; i < len(kvs); i += 2 {
if err := imetadata.ValidatePair(kvs[i], kvs[i+1]); err != nil {
return nil, status.Error(codes.Internal, err.Error())
}
}
}
}
if channelz.IsOn() {
cc.incrCallsStarted()
defer func() {
if err != nil {
cc.incrCallsFailed()
}
}()
}
// Provide an opportunity for the first RPC to see the first service config
// provided by the resolver.
if err := cc.waitForResolvedAddrs(ctx); err != nil {
return nil, err
}
var mc serviceconfig.MethodConfig
var onCommit func()
newStream := func(ctx context.Context, done func()) (iresolver.ClientStream, error) {
return newClientStreamWithParams(ctx, desc, cc, method, mc, onCommit, done, opts...)
}
rpcInfo := iresolver.RPCInfo{Context: ctx, Method: method}
rpcConfig, err := cc.safeConfigSelector.SelectConfig(rpcInfo)
if err != nil {
if st, ok := status.FromError(err); ok {
// Restrict the code to the list allowed by gRFC A54.
if istatus.IsRestrictedControlPlaneCode(st) {
err = status.Errorf(codes.Internal, "config selector returned illegal status: %v", err)
}
return nil, err
}
return nil, toRPCErr(err)
}
if rpcConfig != nil {
if rpcConfig.Context != nil {
ctx = rpcConfig.Context
}
mc = rpcConfig.MethodConfig
onCommit = rpcConfig.OnCommitted
if rpcConfig.Interceptor != nil {
rpcInfo.Context = nil
ns := newStream
newStream = func(ctx context.Context, done func()) (iresolver.ClientStream, error) {
cs, err := rpcConfig.Interceptor.NewStream(ctx, rpcInfo, done, ns)
if err != nil {
return nil, toRPCErr(err)
}
return cs, nil
}
}
}
return newStream(ctx, func() {})
}
func newClientStreamWithParams(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, mc serviceconfig.MethodConfig, onCommit, doneFunc func(), opts ...CallOption) (_ iresolver.ClientStream, err error) {
c := defaultCallInfo()
if mc.WaitForReady != nil {
c.failFast = !*mc.WaitForReady
}
// Possible context leak:
// The cancel function for the child context we create will only be called
// when RecvMsg returns a non-nil error, if the ClientConn is closed, or if
// an error is generated by SendMsg.
// https://github.com/grpc/grpc-go/issues/1818.
var cancel context.CancelFunc
if mc.Timeout != nil && *mc.Timeout >= 0 {
ctx, cancel = context.WithTimeout(ctx, *mc.Timeout)
} else {
ctx, cancel = context.WithCancel(ctx)
}
defer func() {
if err != nil {
cancel()
}
}()
for _, o := range opts {
if err := o.before(c); err != nil {
return nil, toRPCErr(err)
}
}
c.maxSendMessageSize = getMaxSize(mc.MaxReqSize, c.maxSendMessageSize, defaultClientMaxSendMessageSize)
c.maxReceiveMessageSize = getMaxSize(mc.MaxRespSize, c.maxReceiveMessageSize, defaultClientMaxReceiveMessageSize)
if err := setCallInfoCodec(c); err != nil {
return nil, err
}
callHdr := &transport.CallHdr{
Host: cc.authority,
Method: method,
ContentSubtype: c.contentSubtype,
DoneFunc: doneFunc,
}
// Set our outgoing compression according to the UseCompressor CallOption, if
// set. In that case, also find the compressor from the encoding package.
// Otherwise, use the compressor configured by the WithCompressor DialOption,
// if set.
var cp Compressor
var comp encoding.Compressor
if ct := c.compressorType; ct != "" {
callHdr.SendCompress = ct
if ct != encoding.Identity {
comp = encoding.GetCompressor(ct)
if comp == nil {
return nil, status.Errorf(codes.Internal, "grpc: Compressor is not installed for requested grpc-encoding %q", ct)
}
}
} else if cc.dopts.cp != nil {
callHdr.SendCompress = cc.dopts.cp.Type()
cp = cc.dopts.cp
}
if c.creds != nil {
callHdr.Creds = c.creds
}
cs := &clientStream{
callHdr: callHdr,
ctx: ctx,
methodConfig: &mc,
opts: opts,
callInfo: c,
cc: cc,
desc: desc,
codec: c.codec,
cp: cp,
comp: comp,
cancel: cancel,
firstAttempt: true,
onCommit: onCommit,
}
if !cc.dopts.disableRetry {
cs.retryThrottler = cc.retryThrottler.Load().(*retryThrottler)
}
if ml := binarylog.GetMethodLogger(method); ml != nil {
cs.binlogs = append(cs.binlogs, ml)
}
if cc.dopts.binaryLogger != nil {
if ml := cc.dopts.binaryLogger.GetMethodLogger(method); ml != nil {
cs.binlogs = append(cs.binlogs, ml)
}
}
// Pick the transport to use and create a new stream on the transport.
// Assign cs.attempt upon success.
op := func(a *csAttempt) error {
if err := a.getTransport(); err != nil {
return err
}
if err := a.newStream(); err != nil {
return err
}
// Because this operation is always called either here (while creating
// the clientStream) or by the retry code while locked when replaying
// the operation, it is safe to access cs.attempt directly.
cs.attempt = a
return nil
}
if err := cs.withRetry(op, func() { cs.bufferForRetryLocked(0, op, nil) }); err != nil {
return nil, err
}
if len(cs.binlogs) != 0 {
md, _ := metadata.FromOutgoingContext(ctx)
logEntry := &binarylog.ClientHeader{
OnClientSide: true,
Header: md,
MethodName: method,
Authority: cs.cc.authority,
}
if deadline, ok := ctx.Deadline(); ok {
logEntry.Timeout = time.Until(deadline)
if logEntry.Timeout < 0 {
logEntry.Timeout = 0
}
}
for _, binlog := range cs.binlogs {
binlog.Log(cs.ctx, logEntry)
}
}
if desc != unaryStreamDesc {
// Listen on cc and stream contexts to cleanup when the user closes the
// ClientConn or cancels the stream context. In all other cases, an error
// should already be injected into the recv buffer by the transport, which
// the client will eventually receive, and then we will cancel the stream's
// context in clientStream.finish.
go func() {
select {
case <-cc.ctx.Done():
cs.finish(ErrClientConnClosing)
case <-ctx.Done():
cs.finish(toRPCErr(ctx.Err()))
}
}()
}
return cs, nil
}
// newAttemptLocked creates a new csAttempt without a transport or stream.
func (cs *clientStream) newAttemptLocked(isTransparent bool) (*csAttempt, error) {
if err := cs.ctx.Err(); err != nil {
return nil, toRPCErr(err)
}
if err := cs.cc.ctx.Err(); err != nil {
return nil, ErrClientConnClosing
}
ctx := newContextWithRPCInfo(cs.ctx, cs.callInfo.failFast, cs.callInfo.codec, cs.cp, cs.comp)
method := cs.callHdr.Method
var beginTime time.Time
shs := cs.cc.dopts.copts.StatsHandlers
for _, sh := range shs {
ctx = sh.TagRPC(ctx, &stats.RPCTagInfo{FullMethodName: method, FailFast: cs.callInfo.failFast})
beginTime = time.Now()
begin := &stats.Begin{
Client: true,
BeginTime: beginTime,
FailFast: cs.callInfo.failFast,
IsClientStream: cs.desc.ClientStreams,
IsServerStream: cs.desc.ServerStreams,
IsTransparentRetryAttempt: isTransparent,
}
sh.HandleRPC(ctx, begin)
}
var trInfo *traceInfo
if EnableTracing {
trInfo = &traceInfo{
tr: newTrace("grpc.Sent."+methodFamily(method), method),
firstLine: firstLine{
client: true,
},
}
if deadline, ok := ctx.Deadline(); ok {
trInfo.firstLine.deadline = time.Until(deadline)
}
trInfo.tr.LazyLog(&trInfo.firstLine, false)
ctx = newTraceContext(ctx, trInfo.tr)
}
if cs.cc.parsedTarget.URL.Scheme == internal.GRPCResolverSchemeExtraMetadata {
// Add extra metadata (metadata that will be added by transport) to context
// so the balancer can see them.
ctx = grpcutil.WithExtraMetadata(ctx, metadata.Pairs(
"content-type", grpcutil.ContentType(cs.callHdr.ContentSubtype),
))
}
return &csAttempt{
ctx: ctx,
beginTime: beginTime,
cs: cs,
dc: cs.cc.dopts.dc,
statsHandlers: shs,
trInfo: trInfo,
}, nil
}
func (a *csAttempt) getTransport() error {
cs := a.cs
var err error
a.t, a.pickResult, err = cs.cc.getTransport(a.ctx, cs.callInfo.failFast, cs.callHdr.Method)
if err != nil {
if de, ok := err.(dropError); ok {
err = de.error
a.drop = true
}
return err
}
if a.trInfo != nil {
a.trInfo.firstLine.SetRemoteAddr(a.t.RemoteAddr())
}
return nil
}
func (a *csAttempt) newStream() error {
cs := a.cs
cs.callHdr.PreviousAttempts = cs.numRetries
// Merge metadata stored in PickResult, if any, with existing call metadata.
// It is safe to overwrite the csAttempt's context here, since all state
// maintained in it are local to the attempt. When the attempt has to be
// retried, a new instance of csAttempt will be created.
if a.pickResult.Metadata != nil {
// We currently do not have a function it the metadata package which
// merges given metadata with existing metadata in a context. Existing
// function `AppendToOutgoingContext()` takes a variadic argument of key
// value pairs.
//
// TODO: Make it possible to retrieve key value pairs from metadata.MD
// in a form passable to AppendToOutgoingContext(), or create a version
// of AppendToOutgoingContext() that accepts a metadata.MD.
md, _ := metadata.FromOutgoingContext(a.ctx)
md = metadata.Join(md, a.pickResult.Metadata)
a.ctx = metadata.NewOutgoingContext(a.ctx, md)
}
s, err := a.t.NewStream(a.ctx, cs.callHdr)
if err != nil {
nse, ok := err.(*transport.NewStreamError)
if !ok {
// Unexpected.
return err
}
if nse.AllowTransparentRetry {
a.allowTransparentRetry = true
}
// Unwrap and convert error.
return toRPCErr(nse.Err)
}
a.s = s
a.ctx = s.Context()
a.p = &parser{r: s, bufferPool: a.cs.cc.dopts.copts.BufferPool}
return nil
}
// clientStream implements a client side Stream.
type clientStream struct {
callHdr *transport.CallHdr
opts []CallOption
callInfo *callInfo
cc *ClientConn
desc *StreamDesc
codec baseCodec
cp Compressor
comp encoding.Compressor
cancel context.CancelFunc // cancels all attempts
sentLast bool // sent an end stream
methodConfig *MethodConfig
ctx context.Context // the application's context, wrapped by stats/tracing
retryThrottler *retryThrottler // The throttler active when the RPC began.
binlogs []binarylog.MethodLogger
// serverHeaderBinlogged is a boolean for whether server header has been
// logged. Server header will be logged when the first time one of those
// happens: stream.Header(), stream.Recv().
//
// It's only read and used by Recv() and Header(), so it doesn't need to be
// synchronized.
serverHeaderBinlogged bool
mu sync.Mutex
firstAttempt bool // if true, transparent retry is valid
numRetries int // exclusive of transparent retry attempt(s)
numRetriesSincePushback int // retries since pushback; to reset backoff
finished bool // TODO: replace with atomic cmpxchg or sync.Once?
// attempt is the active client stream attempt.
// The only place where it is written is the newAttemptLocked method and this method never writes nil.
// So, attempt can be nil only inside newClientStream function when clientStream is first created.
// One of the first things done after clientStream's creation, is to call newAttemptLocked which either
// assigns a non nil value to the attempt or returns an error. If an error is returned from newAttemptLocked,
// then newClientStream calls finish on the clientStream and returns. So, finish method is the only
// place where we need to check if the attempt is nil.
attempt *csAttempt
// TODO(hedging): hedging will have multiple attempts simultaneously.
committed bool // active attempt committed for retry?
onCommit func()
replayBuffer []replayOp // operations to replay on retry
replayBufferSize int // current size of replayBuffer
}
type replayOp struct {
op func(a *csAttempt) error
cleanup func()
}
// csAttempt implements a single transport stream attempt within a
// clientStream.
type csAttempt struct {
ctx context.Context
cs *clientStream
t transport.ClientTransport
s *transport.ClientStream
p *parser
pickResult balancer.PickResult
finished bool
dc Decompressor
decomp encoding.Compressor
decompSet bool
mu sync.Mutex // guards trInfo.tr
// trInfo may be nil (if EnableTracing is false).
// trInfo.tr is set when created (if EnableTracing is true),
// and cleared when the finish method is called.
trInfo *traceInfo
statsHandlers []stats.Handler
beginTime time.Time
// set for newStream errors that may be transparently retried
allowTransparentRetry bool
// set for pick errors that are returned as a status
drop bool
}
func (cs *clientStream) commitAttemptLocked() {
if !cs.committed && cs.onCommit != nil {
cs.onCommit()
}
cs.committed = true
for _, op := range cs.replayBuffer {
if op.cleanup != nil {
op.cleanup()
}
}
cs.replayBuffer = nil
}
func (cs *clientStream) commitAttempt() {
cs.mu.Lock()
cs.commitAttemptLocked()
cs.mu.Unlock()
}
// shouldRetry returns nil if the RPC should be retried; otherwise it returns
// the error that should be returned by the operation. If the RPC should be
// retried, the bool indicates whether it is being retried transparently.
func (a *csAttempt) shouldRetry(err error) (bool, error) {
cs := a.cs
if cs.finished || cs.committed || a.drop {
// RPC is finished or committed or was dropped by the picker; cannot retry.
return false, err
}
if a.s == nil && a.allowTransparentRetry {
return true, nil
}
// Wait for the trailers.
unprocessed := false
if a.s != nil {
<-a.s.Done()
unprocessed = a.s.Unprocessed()
}
if cs.firstAttempt && unprocessed {
// First attempt, stream unprocessed: transparently retry.
return true, nil
}
if cs.cc.dopts.disableRetry {
return false, err
}
pushback := 0
hasPushback := false
if a.s != nil {
if !a.s.TrailersOnly() {
return false, err
}
// TODO(retry): Move down if the spec changes to not check server pushback
// before considering this a failure for throttling.
sps := a.s.Trailer()["grpc-retry-pushback-ms"]
if len(sps) == 1 {
var e error
if pushback, e = strconv.Atoi(sps[0]); e != nil || pushback < 0 {
channelz.Infof(logger, cs.cc.channelz, "Server retry pushback specified to abort (%q).", sps[0])
cs.retryThrottler.throttle() // This counts as a failure for throttling.
return false, err
}
hasPushback = true
} else if len(sps) > 1 {
channelz.Warningf(logger, cs.cc.channelz, "Server retry pushback specified multiple values (%q); not retrying.", sps)
cs.retryThrottler.throttle() // This counts as a failure for throttling.
return false, err
}
}
var code codes.Code
if a.s != nil {
code = a.s.Status().Code()
} else {
code = status.Code(err)
}
rp := cs.methodConfig.RetryPolicy
if rp == nil || !rp.RetryableStatusCodes[code] {
return false, err
}
// Note: the ordering here is important; we count this as a failure
// only if the code matched a retryable code.
if cs.retryThrottler.throttle() {
return false, err
}
if cs.numRetries+1 >= rp.MaxAttempts {
return false, err
}
var dur time.Duration
if hasPushback {
dur = time.Millisecond * time.Duration(pushback)
cs.numRetriesSincePushback = 0
} else {
fact := math.Pow(rp.BackoffMultiplier, float64(cs.numRetriesSincePushback))
cur := min(float64(rp.InitialBackoff)*fact, float64(rp.MaxBackoff))
// Apply jitter by multiplying with a random factor between 0.8 and 1.2
cur *= 0.8 + 0.4*rand.Float64()
dur = time.Duration(int64(cur))
cs.numRetriesSincePushback++
}
// TODO(dfawley): we could eagerly fail here if dur puts us past the
// deadline, but unsure if it is worth doing.
t := time.NewTimer(dur)
select {
case <-t.C:
cs.numRetries++
return false, nil
case <-cs.ctx.Done():
t.Stop()
return false, status.FromContextError(cs.ctx.Err()).Err()
}
}
// Returns nil if a retry was performed and succeeded; error otherwise.
func (cs *clientStream) retryLocked(attempt *csAttempt, lastErr error) error {
for {
attempt.finish(toRPCErr(lastErr))
isTransparent, err := attempt.shouldRetry(lastErr)
if err != nil {
cs.commitAttemptLocked()
return err
}
cs.firstAttempt = false
attempt, err = cs.newAttemptLocked(isTransparent)
if err != nil {
// Only returns error if the clientconn is closed or the context of
// the stream is canceled.
return err
}
// Note that the first op in replayBuffer always sets cs.attempt
// if it is able to pick a transport and create a stream.
if lastErr = cs.replayBufferLocked(attempt); lastErr == nil {
return nil
}
}
}
func (cs *clientStream) Context() context.Context {
cs.commitAttempt()
// No need to lock before using attempt, since we know it is committed and
// cannot change.
if cs.attempt.s != nil {
return cs.attempt.s.Context()
}
return cs.ctx
}
func (cs *clientStream) withRetry(op func(a *csAttempt) error, onSuccess func()) error {
cs.mu.Lock()
for {
if cs.committed {
cs.mu.Unlock()
// toRPCErr is used in case the error from the attempt comes from
// NewClientStream, which intentionally doesn't return a status
// error to allow for further inspection; all other errors should
// already be status errors.
return toRPCErr(op(cs.attempt))
}
if len(cs.replayBuffer) == 0 {
// For the first op, which controls creation of the stream and
// assigns cs.attempt, we need to create a new attempt inline
// before executing the first op. On subsequent ops, the attempt
// is created immediately before replaying the ops.
var err error
if cs.attempt, err = cs.newAttemptLocked(false /* isTransparent */); err != nil {
cs.mu.Unlock()
cs.finish(err)
return err
}
}
a := cs.attempt
cs.mu.Unlock()
err := op(a)
cs.mu.Lock()
if a != cs.attempt {
// We started another attempt already.
continue
}
if err == io.EOF {
<-a.s.Done()
}
if err == nil || (err == io.EOF && a.s.Status().Code() == codes.OK) {
onSuccess()
cs.mu.Unlock()
return err
}
if err := cs.retryLocked(a, err); err != nil {
cs.mu.Unlock()
return err
}
}
}
func (cs *clientStream) Header() (metadata.MD, error) {
var m metadata.MD
err := cs.withRetry(func(a *csAttempt) error {
var err error
m, err = a.s.Header()
return toRPCErr(err)
}, cs.commitAttemptLocked)
if m == nil && err == nil {
// The stream ended with success. Finish the clientStream.
err = io.EOF
}
if err != nil {
cs.finish(err)
// Do not return the error. The user should get it by calling Recv().
return nil, nil
}
if len(cs.binlogs) != 0 && !cs.serverHeaderBinlogged && m != nil {
// Only log if binary log is on and header has not been logged, and
// there is actually headers to log.
logEntry := &binarylog.ServerHeader{
OnClientSide: true,
Header: m,
PeerAddr: nil,
}
if peer, ok := peer.FromContext(cs.Context()); ok {
logEntry.PeerAddr = peer.Addr
}
cs.serverHeaderBinlogged = true
for _, binlog := range cs.binlogs {
binlog.Log(cs.ctx, logEntry)
}
}
return m, nil
}
func (cs *clientStream) Trailer() metadata.MD {
// On RPC failure, we never need to retry, because usage requires that
// RecvMsg() returned a non-nil error before calling this function is valid.
// We would have retried earlier if necessary.
//
// Commit the attempt anyway, just in case users are not following those
// directions -- it will prevent races and should not meaningfully impact
// performance.
cs.commitAttempt()
if cs.attempt.s == nil {
return nil
}
return cs.attempt.s.Trailer()
}
func (cs *clientStream) replayBufferLocked(attempt *csAttempt) error {
for _, f := range cs.replayBuffer {
if err := f.op(attempt); err != nil {
return err
}
}
return nil
}
func (cs *clientStream) bufferForRetryLocked(sz int, op func(a *csAttempt) error, cleanup func()) {
// Note: we still will buffer if retry is disabled (for transparent retries).
if cs.committed {
return
}
cs.replayBufferSize += sz
if cs.replayBufferSize > cs.callInfo.maxRetryRPCBufferSize {
cs.commitAttemptLocked()
cleanup()
return
}
cs.replayBuffer = append(cs.replayBuffer, replayOp{op: op, cleanup: cleanup})
}
func (cs *clientStream) SendMsg(m any) (err error) {
defer func() {
if err != nil && err != io.EOF {
// Call finish on the client stream for errors generated by this SendMsg
// call, as these indicate problems created by this client. (Transport
// errors are converted to an io.EOF error in csAttempt.sendMsg; the real
// error will be returned from RecvMsg eventually in that case, or be
// retried.)
cs.finish(err)
}
}()
if cs.sentLast {
return status.Errorf(codes.Internal, "SendMsg called after CloseSend")
}
if !cs.desc.ClientStreams {
cs.sentLast = true
}
// load hdr, payload, data
hdr, data, payload, pf, err := prepareMsg(m, cs.codec, cs.cp, cs.comp, cs.cc.dopts.copts.BufferPool)
if err != nil {
return err
}
defer func() {
data.Free()
// only free payload if compression was made, and therefore it is a different set
// of buffers from data.
if pf.isCompressed() {
payload.Free()
}
}()
dataLen := data.Len()
payloadLen := payload.Len()
// TODO(dfawley): should we be checking len(data) instead?
if payloadLen > *cs.callInfo.maxSendMessageSize {
return status.Errorf(codes.ResourceExhausted, "trying to send message larger than max (%d vs. %d)", payloadLen, *cs.callInfo.maxSendMessageSize)
}
// always take an extra ref in case data == payload (i.e. when the data isn't
// compressed). The original ref will always be freed by the deferred free above.
payload.Ref()
op := func(a *csAttempt) error {
return a.sendMsg(m, hdr, payload, dataLen, payloadLen)
}
// onSuccess is invoked when the op is captured for a subsequent retry. If the
// stream was established by a previous message and therefore retries are
// disabled, onSuccess will not be invoked, and payloadRef can be freed
// immediately.
onSuccessCalled := false
err = cs.withRetry(op, func() {
cs.bufferForRetryLocked(len(hdr)+payloadLen, op, payload.Free)
onSuccessCalled = true
})
if !onSuccessCalled {
payload.Free()
}
if len(cs.binlogs) != 0 && err == nil {
cm := &binarylog.ClientMessage{
OnClientSide: true,
Message: data.Materialize(),
}
for _, binlog := range cs.binlogs {
binlog.Log(cs.ctx, cm)
}
}
return err
}
func (cs *clientStream) RecvMsg(m any) error {
if len(cs.binlogs) != 0 && !cs.serverHeaderBinlogged {
// Call Header() to binary log header if it's not already logged.
cs.Header()
}
var recvInfo *payloadInfo
if len(cs.binlogs) != 0 {
recvInfo = &payloadInfo{}
defer recvInfo.free()
}
err := cs.withRetry(func(a *csAttempt) error {
return a.recvMsg(m, recvInfo)
}, cs.commitAttemptLocked)
if len(cs.binlogs) != 0 && err == nil {
sm := &binarylog.ServerMessage{
OnClientSide: true,
Message: recvInfo.uncompressedBytes.Materialize(),
}
for _, binlog := range cs.binlogs {
binlog.Log(cs.ctx, sm)
}
}
if err != nil || !cs.desc.ServerStreams {
// err != nil or non-server-streaming indicates end of stream.
cs.finish(err)
}
return err
}
func (cs *clientStream) CloseSend() error {
if cs.sentLast {
// TODO: return an error and finish the stream instead, due to API misuse?
return nil
}
cs.sentLast = true
op := func(a *csAttempt) error {
a.s.Write(nil, nil, &transport.WriteOptions{Last: true})
// Always return nil; io.EOF is the only error that might make sense
// instead, but there is no need to signal the client to call RecvMsg
// as the only use left for the stream after CloseSend is to call
// RecvMsg. This also matches historical behavior.
return nil