forked from cockroachdb/cockroach
/
start.go
975 lines (880 loc) · 31.3 KB
/
start.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
// Copyright 2015 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.
package cli
import (
"bytes"
"flag"
"fmt"
"io"
"io/ioutil"
"math"
"net"
"net/url"
"os"
"os/signal"
"path/filepath"
"runtime"
"runtime/pprof"
"strconv"
"strings"
"syscall"
"text/tabwriter"
"time"
humanize "github.com/dustin/go-humanize"
"github.com/elastic/gosigar"
opentracing "github.com/opentracing/opentracing-go"
"github.com/pkg/errors"
"github.com/spf13/cobra"
"golang.org/x/net/context"
"google.golang.org/grpc"
"github.com/cockroachdb/cockroach/pkg/build"
"github.com/cockroachdb/cockroach/pkg/cli/cliflags"
"github.com/cockroachdb/cockroach/pkg/rpc"
"github.com/cockroachdb/cockroach/pkg/security"
"github.com/cockroachdb/cockroach/pkg/server"
"github.com/cockroachdb/cockroach/pkg/server/serverpb"
"github.com/cockroachdb/cockroach/pkg/util/envutil"
"github.com/cockroachdb/cockroach/pkg/util/grpcutil"
"github.com/cockroachdb/cockroach/pkg/util/hlc"
"github.com/cockroachdb/cockroach/pkg/util/humanizeutil"
"github.com/cockroachdb/cockroach/pkg/util/log"
"github.com/cockroachdb/cockroach/pkg/util/log/logflags"
"github.com/cockroachdb/cockroach/pkg/util/stop"
"github.com/cockroachdb/cockroach/pkg/util/syncutil"
"github.com/cockroachdb/cockroach/pkg/util/timeutil"
)
// jemallocHeapDump is an optional function to be called at heap dump time.
// This will be non-nil when jemalloc is linked in with profiling enabled.
// The function takes a filename to write the profile to.
var jemallocHeapDump func(string) error
// startCmd starts a node by initializing the stores and joining
// the cluster.
var startCmd = &cobra.Command{
Use: "start",
Short: "start a node",
Long: `
Start a CockroachDB node, which will export data from one or more
storage devices, specified via --store flags.
If no cluster exists yet and this is the first node, no additional
flags are required. If the cluster already exists, and this node is
uninitialized, specify the --join flag to point to any healthy node
(or list of nodes) already part of the cluster.
`,
Example: ` cockroach start --insecure --store=attrs=ssd,path=/mnt/ssd1 [--join=host:port,[host:port]]`,
RunE: MaybeShoutError(MaybeDecorateGRPCError(runStart)),
}
// maxSizePerProfile is the maximum total size in bytes for profiles per
// profile type.
var maxSizePerProfile = envutil.EnvOrDefaultInt64(
"COCKROACH_MAX_SIZE_PER_PROFILE", 100<<20 /* 100 MB */)
// gcProfiles removes old profiles matching the specified prefix when the sum
// of newer profiles is larger than maxSize. Requires that the suffix used for
// the profiles indicates age (e.g. by using a date/timestamp suffix) such that
// sorting the filenames corresponds to ordering the profiles from oldest to
// newest.
func gcProfiles(dir, prefix string, maxSize int64) {
files, err := ioutil.ReadDir(dir)
if err != nil {
log.Warning(context.Background(), err)
return
}
var sum int64
var found int
for i := len(files) - 1; i >= 0; i-- {
f := files[i]
if !f.Mode().IsRegular() {
continue
}
if !strings.HasPrefix(f.Name(), prefix) {
continue
}
found++
sum += f.Size()
if found == 1 {
// Always keep the most recent profile.
continue
}
if sum <= maxSize {
continue
}
if err := os.Remove(filepath.Join(dir, f.Name())); err != nil {
log.Info(context.Background(), err)
}
}
}
func initMemProfile(ctx context.Context, dir string) {
const jeprof = "jeprof."
const memprof = "memprof."
gcProfiles(dir, jeprof, maxSizePerProfile)
gcProfiles(dir, memprof, maxSizePerProfile)
memProfileInterval := envutil.EnvOrDefaultDuration("COCKROACH_MEMPROF_INTERVAL", -1)
if memProfileInterval <= 0 {
return
}
if min := time.Second; memProfileInterval < min {
log.Infof(ctx, "fixing excessively short memory profiling interval: %s -> %s",
memProfileInterval, min)
memProfileInterval = min
}
if jemallocHeapDump != nil {
log.Infof(ctx, "writing go and jemalloc memory profiles to %s every %s", dir, memProfileInterval)
} else {
log.Infof(ctx, "writing go only memory profiles to %s every %s", dir, memProfileInterval)
log.Infof(ctx, `to enable jmalloc profiling: "export MALLOC_CONF=prof:true" or "ln -s prof:true /etc/malloc.conf"`)
}
go func() {
ctx := context.Background()
t := time.NewTicker(memProfileInterval)
defer t.Stop()
for {
<-t.C
func() {
const format = "2006-01-02T15_04_05.999"
suffix := timeutil.Now().Format(format)
// Try jemalloc heap profile first, we only log errors.
if jemallocHeapDump != nil {
jepath := filepath.Join(dir, jeprof+suffix)
if err := jemallocHeapDump(jepath); err != nil {
log.Warningf(ctx, "error writing jemalloc heap %s: %s", jepath, err)
}
gcProfiles(dir, jeprof, maxSizePerProfile)
}
path := filepath.Join(dir, memprof+suffix)
// Try writing a go heap profile.
f, err := os.Create(path)
if err != nil {
log.Warningf(ctx, "error creating go heap file %s", err)
return
}
defer f.Close()
if err = pprof.WriteHeapProfile(f); err != nil {
log.Warningf(ctx, "error writing go heap %s: %s", path, err)
return
}
gcProfiles(dir, memprof, maxSizePerProfile)
}()
}
}()
}
func initCPUProfile(ctx context.Context, dir string) {
const cpuprof = "cpuprof."
gcProfiles(dir, cpuprof, maxSizePerProfile)
cpuProfileInterval := envutil.EnvOrDefaultDuration("COCKROACH_CPUPROF_INTERVAL", -1)
if cpuProfileInterval <= 0 {
return
}
if min := time.Second; cpuProfileInterval < min {
log.Infof(ctx, "fixing excessively short cpu profiling interval: %s -> %s",
cpuProfileInterval, min)
cpuProfileInterval = min
}
go func() {
defer log.RecoverAndReportPanic(ctx, &serverCfg.Settings.SV)
ctx := context.Background()
t := time.NewTicker(cpuProfileInterval)
defer t.Stop()
var currentProfile *os.File
defer func() {
if currentProfile != nil {
pprof.StopCPUProfile()
currentProfile.Close()
}
}()
for {
func() {
const format = "2006-01-02T15_04_05.999"
suffix := timeutil.Now().Add(cpuProfileInterval).Format(format)
f, err := os.Create(filepath.Join(dir, cpuprof+suffix))
if err != nil {
log.Warningf(ctx, "error creating go cpu file %s", err)
return
}
// Stop the current profile if it exists.
if currentProfile != nil {
pprof.StopCPUProfile()
currentProfile.Close()
currentProfile = nil
gcProfiles(dir, cpuprof, maxSizePerProfile)
}
// Start the new profile.
if err := pprof.StartCPUProfile(f); err != nil {
log.Warningf(ctx, "unable to start cpu profile: %v", err)
f.Close()
return
}
currentProfile = f
}()
<-t.C
}
}()
}
func initBlockProfile() {
// Enable the block profile for a sample of mutex and channel operations.
// Smaller values provide more accurate profiles but are more
// expensive. 0 and 1 are special: 0 disables the block profile and
// 1 captures 100% of block events. For other values, the profiler
// will sample one event per X nanoseconds spent blocking.
//
// The block profile can be viewed with `pprof http://HOST:PORT/debug/pprof/block`
d := envutil.EnvOrDefaultInt64("COCKROACH_BLOCK_PROFILE_RATE",
10000000 /* 1 sample per 10 milliseconds spent blocking */)
runtime.SetBlockProfileRate(int(d))
}
// ErrorCode is the value to be used by main() as exit code in case of
// error. For most errors 1 is appropriate, but a signal termination
// can change this.
var ErrorCode = 1
type percentResolverFunc func(percent int) (int64, error)
// bytesOrPercentageValue is a flag that accepts an integer value, an integer
// plus a unit (e.g. 32GB or 32GiB) or a percentage (e.g. 32%). In all these
// cases, it transforms the string flag input into an int64 value.
//
// Since it accepts a percentage, instances need to be configured with
// instructions on how to resolve a percentage to a number (i.e. the answer to
// the question "a percentage of what?"). This is done by taking in a
// percentResolverFunc. There are predefined ones: memoryPercentResolver and
// diskPercentResolverFactory.
//
// bytesOrPercentageValue can be used in two ways:
// 1. Upon flag parsing, it can write an int64 value through a pointer specified
// by the caller.
// 2. It can store the flag value as a string and only convert it to an int64 on
// a subsequent Resolve() call. Input validation still happens at flag parsing
// time.
//
// Option 2 is useful when percentages cannot be resolved at flag parsing time.
// For example, we have flags that can be expressed as percentages of the
// capacity of storage device. Which storage device is in question might only be
// known once other flags are parsed (e.g. --max-disk-temp-storage=10% depends
// on --store).
type bytesOrPercentageValue struct {
val *int64
bval *humanizeutil.BytesValue
origVal string
// percentResolver is used to turn a percent string into a value. See
// memoryPercentResolver() and diskPercentResolverFactory().
percentResolver percentResolverFunc
}
// memoryPercentResolver turns a percent into the respective fraction of the
// system's internal memory.
func memoryPercentResolver(percent int) (int64, error) {
sizeBytes, err := server.GetTotalMemory(context.TODO())
if err != nil {
return 0, err
}
return (sizeBytes * int64(percent)) / 100, nil
}
// diskPercentResolverFactory takes in a path and produces a percentResolverFunc
// bound to the respective storage device.
//
// An error is returned if dir does not exist.
func diskPercentResolverFactory(dir string) (percentResolverFunc, error) {
fileSystemUsage := gosigar.FileSystemUsage{}
if err := fileSystemUsage.Get(dir); err != nil {
return nil, err
}
if fileSystemUsage.Total > math.MaxInt64 {
return nil, fmt.Errorf("unsupported disk size %s, max supported size is %s",
humanize.IBytes(fileSystemUsage.Total), humanizeutil.IBytes(math.MaxInt64))
}
deviceCapacity := int64(fileSystemUsage.Total)
return func(percent int) (int64, error) {
return (deviceCapacity * int64(percent)) / 100, nil
}, nil
}
// newBytesOrPercentageValue creates a bytesOrPercentageValue.
//
// v and percentResolver can be nil (either they're both specified or they're
// both nil). If they're nil, then Resolve() has to be called later to get the
// passed-in value.
func newBytesOrPercentageValue(
v *int64, percentResolver func(percent int) (int64, error),
) *bytesOrPercentageValue {
if v == nil {
v = new(int64)
}
return &bytesOrPercentageValue{
val: v,
bval: humanizeutil.NewBytesValue(v),
percentResolver: percentResolver,
}
}
func (b *bytesOrPercentageValue) Set(s string) error {
b.origVal = s
if strings.HasSuffix(s, "%") {
percent, err := strconv.Atoi(s[:len(s)-1])
if err != nil {
return err
}
if percent < 0 || percent > 99 {
return fmt.Errorf("percentage %s out of range 0%% - 99%%", s)
}
if b.percentResolver == nil {
// percentResolver not set means that this flag is not yet supposed to set
// any value.
return nil
}
absVal, err := b.percentResolver(percent)
if err != nil {
return err
}
s = fmt.Sprint(absVal)
}
return b.bval.Set(s)
}
// Resolve can be called to get the flag's value (if any). If the flag had been
// previously set, *v will be written.
func (b *bytesOrPercentageValue) Resolve(v *int64, percentResolver percentResolverFunc) error {
// The flag was not passed on the command line.
if b.origVal == "" {
return nil
}
b.percentResolver = percentResolver
b.val = v
b.bval = humanizeutil.NewBytesValue(v)
return b.Set(b.origVal)
}
func (b *bytesOrPercentageValue) Type() string {
return b.bval.Type()
}
func (b *bytesOrPercentageValue) String() string {
return b.bval.String()
}
func (b *bytesOrPercentageValue) IsSet() bool {
return b.bval.IsSet()
}
var cacheSizeValue = newBytesOrPercentageValue(&serverCfg.CacheSize, memoryPercentResolver)
var sqlSizeValue = newBytesOrPercentageValue(&serverCfg.SQLMemoryPoolSize, memoryPercentResolver)
var diskTempStorageSizeValue = newBytesOrPercentageValue(nil /* v */, nil /* percentResolver */)
// runStart starts the cockroach node using --store as the list of
// storage devices ("stores") on this machine and --join as the list
// of other active nodes used to join this node to the cockroach
// cluster, if this is its first time connecting.
func runStart(cmd *cobra.Command, args []string) error {
if len(args) > 0 {
return usageAndError(cmd)
}
tBegin := timeutil.Now()
if ok, err := maybeRerunBackground(); ok {
return err
}
// Deal with flags that may depend on other flags.
// The temp store size can depend on the location of the first regular store
// (if it's expressed as a percentage), so we resolve that flag here.
var tempStorePercentageResolver percentResolverFunc
if !serverCfg.Stores.Specs[0].InMemory {
dir := serverCfg.Stores.Specs[0].Path
// Create the store dir, if it doesn't exist. The dir is required to exist
// by diskPercentResolverFactory.
if err := os.MkdirAll(dir, 0755); err != nil {
return errors.Wrapf(err, "failed to create dir for first store: %s", dir)
}
var err error
tempStorePercentageResolver, err = diskPercentResolverFactory(dir)
if err != nil {
return errors.Wrapf(err, "failed to create resolver for: %s", dir)
}
} else {
tempStorePercentageResolver = memoryPercentResolver
}
if err := diskTempStorageSizeValue.Resolve(
&serverCfg.TempStoreMaxSizeBytes, tempStorePercentageResolver,
); err != nil {
return err
}
if serverCfg.Stores.Specs[0].InMemory && !diskTempStorageSizeValue.IsSet() {
// The default temp store size is different when the first store (and thus
// also the temp storage) is in memory.
serverCfg.TempStoreMaxSizeBytes = server.DefaultTempStoreMaxSizeBytesInMemStore
}
// Use the server-specific values for some flags and settings.
serverCfg.Insecure = startCtx.serverInsecure
serverCfg.SSLCertsDir = startCtx.serverSSLCertsDir
serverCfg.User = security.NodeUser
serverCfg.TempStoreSpec = server.MakeTempStoreSpecFromStoreSpec(serverCfg.Stores.Specs[0])
signalCh := make(chan os.Signal, 1)
signal.Notify(signalCh, syscall.SIGINT, syscall.SIGTERM, syscall.SIGQUIT)
tracer := serverCfg.Settings.Tracer
sp := tracer.StartSpan("server start")
startCtx := opentracing.ContextWithSpan(context.Background(), sp)
// Set up the logging and profiling output.
// It is important that no logging occurs before this point or the log files
// will be created in $TMPDIR instead of their expected location.
stopper, err := setupAndInitializeLoggingAndProfiling(startCtx)
if err != nil {
return err
}
serverCfg.Report(startCtx)
// Run the rest of the startup process in the background to avoid preventing
// proper handling of signals if we get stuck on something during
// initialization (#10138).
var serverStatusMu struct {
syncutil.Mutex
// Used to synchronize server startup with server shutdown if something
// interrupts the process during initialization (it isn't safe to try to
// drain a server that doesn't exist, or to start a server after draining
// has begun).
created, draining bool
}
var s *server.Server
errChan := make(chan error, 1)
go func() {
// Ensure that the log files see the startup messages immediately.
defer log.Flush()
defer func() {
if s != nil {
log.RecoverAndReportPanic(startCtx, &s.ClusterSettings().SV)
}
}()
defer sp.Finish()
if err := func() error {
if err := serverCfg.InitNode(); err != nil {
return errors.Wrap(err, "failed to initialize node")
}
log.Info(startCtx, "starting cockroach node")
if envVarsUsed := envutil.GetEnvVarsUsed(); len(envVarsUsed) > 0 {
log.Infof(startCtx, "using local environment variables: %s", strings.Join(envVarsUsed, ", "))
}
var err error
s, err = server.NewServer(serverCfg, stopper)
if err != nil {
return errors.Wrap(err, "failed to start server")
}
serverStatusMu.Lock()
serverStatusMu.created = true
draining := serverStatusMu.draining
serverStatusMu.Unlock()
if draining {
return nil
}
if err := s.Start(startCtx); err != nil {
if le, ok := err.(server.ListenError); ok {
const errorPrefix = "consider changing the port via --"
if le.Addr == serverCfg.Addr {
err = errors.Wrap(err, errorPrefix+cliflags.ServerPort.Name)
} else if le.Addr == serverCfg.HTTPAddr {
err = errors.Wrap(err, errorPrefix+cliflags.ServerHTTPPort.Name)
}
}
return errors.Wrap(err, "cockroach server exited with error")
}
// We don't do this in (*server.Server).Start() because we don't want it
// in tests.
if !envutil.EnvOrDefaultBool("COCKROACH_SKIP_UPDATE_CHECK", false) {
s.PeriodicallyCheckForUpdates()
}
pgURL, err := serverCfg.PGURL(url.User(sqlConnUser))
if err != nil {
return err
}
var buf bytes.Buffer
info := build.GetInfo()
tw := tabwriter.NewWriter(&buf, 2, 1, 2, ' ', 0)
fmt.Fprintf(tw, "CockroachDB node starting at %s (took %0.1fs)\n", timeutil.Now(), timeutil.Since(tBegin).Seconds())
fmt.Fprintf(tw, "build:\t%s %s @ %s (%s)\n", info.Distribution, info.Tag, info.Time, info.GoVersion)
fmt.Fprintf(tw, "admin:\t%s\n", serverCfg.AdminURL())
fmt.Fprintf(tw, "sql:\t%s\n", pgURL)
if len(serverCfg.SocketFile) != 0 {
fmt.Fprintf(tw, "socket:\t%s\n", serverCfg.SocketFile)
}
fmt.Fprintf(tw, "logs:\t%s\n", flag.Lookup("log-dir").Value)
if serverCfg.Attrs != "" {
fmt.Fprintf(tw, "attrs:\t%s\n", serverCfg.Attrs)
}
if len(serverCfg.Locality.Tiers) > 0 {
fmt.Fprintf(tw, "locality:\t%s\n", serverCfg.Locality)
}
for i, spec := range serverCfg.Stores.Specs {
fmt.Fprintf(tw, "store[%d]:\t%s\n", i, spec)
}
initialBoot := s.InitialBoot()
nodeID := s.NodeID()
if initialBoot {
if nodeID == server.FirstNodeID {
fmt.Fprintf(tw, "status:\tinitialized new cluster\n")
} else {
fmt.Fprintf(tw, "status:\tinitialized new node, joined pre-existing cluster\n")
}
} else {
fmt.Fprintf(tw, "status:\trestarted pre-existing node\n")
}
fmt.Fprintf(tw, "clusterID:\t%s\n", s.ClusterID())
fmt.Fprintf(tw, "nodeID:\t%d\n", nodeID)
if err := tw.Flush(); err != nil {
return err
}
msg := buf.String()
log.Infof(startCtx, "node startup completed:\n%s", msg)
if !log.LoggingToStderr(log.Severity_INFO) {
fmt.Print(msg)
}
return nil
}(); err != nil {
errChan <- err
}
}()
shutdownSpan := tracer.StartSpan("server shutdown")
defer shutdownSpan.Finish()
shutdownCtx := opentracing.ContextWithSpan(context.Background(), shutdownSpan)
var returnErr error
// Block until one of the signals above is received or the stopper
// is stopped externally (for example, via the quit endpoint).
select {
case err := <-errChan:
// SetSync both flushes and ensures that subsequent log writes are flushed too.
log.SetSync(true)
return err
case <-stopper.ShouldStop():
// Server is being stopped externally and our job is finished
// here since we don't know if it's a graceful shutdown or not.
<-stopper.IsStopped()
// SetSync both flushes and ensures that subsequent log writes are flushed too.
log.SetSync(true)
return nil
case sig := <-signalCh:
// We start synchronizing log writes from here, because if a
// signal was received there is a non-zero chance the sender of
// this signal will follow up with SIGKILL if the shutdown is not
// timely, and we don't want logs to be lost.
log.SetSync(true)
log.Infof(shutdownCtx, "received signal '%s'", sig)
if sig == os.Interrupt {
// Graceful shutdown after an interrupt should cause the process
// to terminate with a non-zero exit code; however SIGTERM is
// "legitimate" and should be acknowledged with a success exit
// code. So we keep the error state here for later.
returnErr = errors.New("interrupted")
msgDouble := "Note: a second interrupt will skip graceful shutdown and terminate forcefully"
fmt.Fprintln(os.Stdout, msgDouble)
}
go func() {
serverStatusMu.Lock()
serverStatusMu.draining = true
needToDrain := serverStatusMu.created
serverStatusMu.Unlock()
if needToDrain {
if _, err := s.Drain(server.GracefulDrainModes); err != nil {
// Don't use shutdownCtx because this is in a goroutine that may
// still be running after shutdownCtx's span has been finished.
log.Warning(context.Background(), err)
}
}
stopper.Stop(context.Background())
}()
}
const msgDrain = "initiating graceful shutdown of server"
log.Info(shutdownCtx, msgDrain)
fmt.Fprintln(os.Stdout, msgDrain)
go func() {
ticker := time.NewTicker(5 * time.Second)
defer ticker.Stop()
for {
select {
case <-ticker.C:
log.Infof(context.Background(), "%d running tasks", stopper.NumTasks())
case <-stopper.ShouldStop():
return
}
}
}()
select {
case sig := <-signalCh:
returnErr = fmt.Errorf("received signal '%s' during shutdown, initiating hard shutdown", sig)
// This new signal is not welcome, as it interferes with the graceful
// shutdown process. On Unix, a signal that was not handled gracefully by
// the application should be visible to other processes as an exit code
// encoded as 128+signal number.
//
// Also, on Unix, os.Signal is syscall.Signal and it's convertible to int.
ErrorCode = 128 + int(sig.(syscall.Signal))
// NB: we do not return here to go through log.Flush below.
case <-time.After(time.Minute):
returnErr = errors.New("time limit reached, initiating hard shutdown")
// NB: we do not return here to go through log.Flush below.
case <-stopper.IsStopped():
const msgDone = "server drained and shutdown completed"
log.Infof(shutdownCtx, msgDone)
fmt.Fprintln(os.Stdout, msgDone)
}
return returnErr
}
func maybeWarnCacheSize() {
if cacheSizeValue.IsSet() {
return
}
var buf bytes.Buffer
fmt.Fprintf(&buf, "Using the default setting for --cache (%s).\n", cacheSizeValue)
fmt.Fprintf(&buf, " A significantly larger value is usually needed for good performance.\n")
if size, err := server.GetTotalMemory(context.Background()); err == nil {
fmt.Fprintf(&buf, " If you have a dedicated server a reasonable setting is --cache=25%% (%s).",
humanizeutil.IBytes(size/4))
} else {
fmt.Fprintf(&buf, " If you have a dedicated server a reasonable setting is 25%% of physical memory.")
}
log.Warning(context.Background(), buf.String())
}
// setupAndInitializeLoggingAndProfiling does what it says on the label.
// Prior to this however it determines suitable defaults for the
// logging output directory and the verbosity level of stderr logging.
// We only do this for the "start" command which is why this work
// occurs here and not in an OnInitialize function.
func setupAndInitializeLoggingAndProfiling(ctx context.Context) (*stop.Stopper, error) {
// Default the log directory to the "logs" subdirectory of the first
// non-memory store. If more than one non-memory stores is detected,
// print a warning.
ambiguousLogDirs := false
pf := cockroachCmd.PersistentFlags()
f := pf.Lookup(logflags.LogDirName)
if !log.DirSet() && !f.Changed {
// We only override the log directory if the user has not explicitly
// disabled file logging using --log-dir="".
newDir := ""
for _, spec := range serverCfg.Stores.Specs {
if spec.InMemory {
continue
}
if newDir != "" {
ambiguousLogDirs = true
break
}
newDir = filepath.Join(spec.Path, "logs")
}
if err := f.Value.Set(newDir); err != nil {
return nil, err
}
}
// We need an output directory below. We use the current directory unless
// the log directory is configured, in which case we use that.
outputDirectory := "."
if logDir := f.Value.String(); logDir != "" {
outputDirectory = logDir
ls := pf.Lookup(logflags.LogToStderrName)
if !ls.Changed {
// Unless the settings were overridden by the user, silence
// logging to stderr because the messages will go to a log file.
if err := ls.Value.Set(log.Severity_NONE.String()); err != nil {
return nil, err
}
}
// Make sure the path exists.
if err := os.MkdirAll(logDir, 0755); err != nil {
return nil, err
}
log.Eventf(ctx, "created log directory %s", logDir)
// Start the log file GC daemon to remove files that make the log
// directory too large.
log.StartGCDaemon()
}
if ambiguousLogDirs {
// Note that we can't report this message earlier, because the log directory
// may not have been ready before the call to MkdirAll() above.
log.Shout(ctx, log.Severity_WARNING, "multiple stores configured"+
" and --log-dir not specified, you may want to specify --log-dir to disambiguate.")
}
if startCtx.serverInsecure {
// Use a non-annotated context here since the annotation just looks funny,
// particularly to new users (made worse by it always printing as [n?]).
addr := serverConnHost
if addr == "" {
addr = "<all your IP addresses>"
}
log.Shout(context.Background(), log.Severity_WARNING,
"RUNNING IN INSECURE MODE!\n\n"+
"- Your cluster is open for any client that can access "+addr+".\n"+
"- Any user, even root, can log in without providing a password.\n"+
"- Any user, connecting as root, can read or write any data in your cluster.\n"+
"- There is no network encryption nor authentication, and thus no confidentiality.\n\n"+
"Check out how to secure your cluster: https://www.cockroachlabs.com/docs/stable/secure-a-cluster.html")
}
maybeWarnCacheSize()
// We log build information to stdout (for the short summary), but also
// to stderr to coincide with the full logs.
info := build.GetInfo()
log.Infof(ctx, info.Short())
initMemProfile(ctx, outputDirectory)
initCPUProfile(ctx, outputDirectory)
initBlockProfile()
// Disable Stopper task tracking as performing that call site tracking is
// moderately expensive (certainly outweighing the infrequent benefit it
// provides).
stopper := initBacktrace(outputDirectory)
log.Event(ctx, "initialized profiles")
return stopper, nil
}
func addrWithDefaultHost(addr string) (string, error) {
host, port, err := net.SplitHostPort(addr)
if err != nil {
return "", err
}
if host == "" {
host = "localhost"
}
return net.JoinHostPort(host, port), nil
}
func getClientGRPCConn() (*grpc.ClientConn, *hlc.Clock, *stop.Stopper, error) {
// 0 to disable max offset checks; this RPC context is not a member of the
// cluster, so there's no need to enforce that its max offset is the same
// as that of nodes in the cluster.
clock := hlc.NewClock(hlc.UnixNano, 0)
stopper := stop.NewStopper()
rpcContext := rpc.NewContext(
log.AmbientContext{Tracer: serverCfg.Settings.Tracer},
serverCfg.Config,
clock,
stopper,
)
addr, err := addrWithDefaultHost(serverCfg.AdvertiseAddr)
if err != nil {
return nil, nil, nil, err
}
conn, err := rpcContext.GRPCDial(addr)
if err != nil {
return nil, nil, nil, err
}
return conn, clock, stopper, nil
}
func getAdminClient() (serverpb.AdminClient, *stop.Stopper, error) {
conn, _, stopper, err := getClientGRPCConn()
if err != nil {
return nil, nil, err
}
return serverpb.NewAdminClient(conn), stopper, nil
}
func stopperContext(stopper *stop.Stopper) context.Context {
ctx, cancel := context.WithCancel(context.Background())
stopper.AddCloser(stop.CloserFn(cancel))
return ctx
}
// quitCmd command shuts down the node server.
var quitCmd = &cobra.Command{
Use: "quit",
Short: "drain and shutdown node\n",
Long: `
Shutdown the server. The first stage is drain, where any new requests
will be ignored by the server. When all extant requests have been
completed, the server exits.
`,
RunE: MaybeDecorateGRPCError(runQuit),
}
// checkNodeRunning performs a no-op RPC and returns an error if it failed to
// connect to the server.
func checkNodeRunning(ctx context.Context, c serverpb.AdminClient) error {
// Send a no-op Drain request.
stream, err := c.Drain(ctx, &serverpb.DrainRequest{
On: nil,
Shutdown: false,
})
if err != nil {
return errors.Wrap(err, "Failed to connect to the node: error sending drain request")
}
// Ignore errors from the stream. We've managed to connect to the node above,
// and that's all that this function is interested in.
for {
if _, err := stream.Recv(); err != nil {
if err != io.EOF {
log.Warningf(ctx, "unexpected error from no-op Drain request: %s", err)
}
break
}
}
return nil
}
// doShutdown attempts to trigger a server shutdown. When given an empty
// onModes slice, it's a hard shutdown.
//
// errTryHardShutdown is returned if the caller should do a hard-shutdown.
func doShutdown(ctx context.Context, c serverpb.AdminClient, onModes []int32) error {
// We want to distinguish between the case in which we can't even connect to
// the server (in which case we don't want our caller to try to come back with
// a hard retry) and the case in which an attempt to shut down fails (times
// out, or perhaps drops the connection while waiting). To that end, we first
// run a noop DrainRequest. If that fails, we give up.
if err := checkNodeRunning(ctx, c); err != nil {
return err
}
// Send a drain request and continue reading until the connection drops (which
// then counts as a success, for the connection dropping is likely the result
// of the Stopper having reached the final stages of shutdown).
stream, err := c.Drain(ctx, &serverpb.DrainRequest{
On: onModes,
Shutdown: true,
})
if err != nil {
// This most likely means that we shut down successfully. Note that
// sometimes the connection can be shut down even before a DrainResponse gets
// sent back to us, so we don't require a response on the stream (see
// #14184).
if grpcutil.IsClosedConnection(err) {
return nil
}
return errors.Wrap(err, "Error sending drain request")
}
for {
if _, err := stream.Recv(); err != nil {
if grpcutil.IsClosedConnection(err) {
return nil
}
// Unexpected error; the caller should try again (and harder).
return errTryHardShutdown{err}
}
}
}
type errTryHardShutdown struct{ error }
// runQuit accesses the quit shutdown path.
func runQuit(cmd *cobra.Command, args []string) (err error) {
if len(args) != 0 {
return usageAndError(cmd)
}
defer func() {
if err == nil {
fmt.Println("ok")
}
}()
onModes := make([]int32, len(server.GracefulDrainModes))
for i, m := range server.GracefulDrainModes {
onModes[i] = int32(m)
}
c, stopper, err := getAdminClient()
if err != nil {
return err
}
ctx := stopperContext(stopper)
defer stopper.Stop(ctx)
if quitCtx.serverDecommission {
var myself []string // will remain empty, which means target yourself
if err := runDecommissionNodeImpl(ctx, c, nodeDecommissionWaitAll, myself); err != nil {
return err
}
}
errChan := make(chan error, 1)
go func() {
errChan <- doShutdown(ctx, c, onModes)
}()
select {
case err := <-errChan:
if err != nil {
if _, ok := err.(errTryHardShutdown); ok {
fmt.Printf("graceful shutdown failed: %s; proceeding with hard shutdown\n", err)
break
}
return err
}
return nil
case <-time.After(time.Minute):
fmt.Println("timed out; proceeding with hard shutdown")
}
// Not passing drain modes tells the server to not bother and go
// straight to shutdown.
return errors.Wrap(doShutdown(ctx, c, nil), "hard shutdown failed")
}