forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrendezvous.html
1578 lines (1341 loc) · 112 KB
/
rendezvous.html
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
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Rendezvous — PyTorch 1.12 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/elastic/rendezvous.html"/>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="../_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="../_static/panels-main.c949a650a448cc0ae9fd3441c0e17fb0.css" type="text/css" />
<link rel="stylesheet" href="../_static/panels-variables.06eb56fa6e07937060861dad626602ad.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/jit.css" type="text/css" />
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="Expiration Timers" href="timer.html" />
<link rel="prev" title="Error Propagation" href="errors.html" />
<!-- Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-117752657-2"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-117752657-2');
</script>
<!-- End Google Analytics -->
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active docs-active">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-orange-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/audio/stable/index.html">
<span class="dropdown-title">torchaudio</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/text/stable/index.html">
<span class="dropdown-title">torchtext</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/vision/stable/index.html">
<span class="dropdown-title">torchvision</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchrec">
<span class="dropdown-title">TorchRec</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/data">
<span class="dropdown-title">TorchData</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/serve/">
<span class="dropdown-title">TorchServe</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchx/">
<span class="dropdown-title">TorchX</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/xla">
<span class="dropdown-title">PyTorch on XLA Devices</span>
<p></p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-arrow">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class="dropdown-title">About</span>
<p>Learn about PyTorch’s features and capabilities</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/#community-module">
<span class="dropdown-title">Community</span>
<p>Join the PyTorch developer community to contribute, learn, and get your questions answered.</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/resources">
<span class="dropdown-title">Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://discuss.pytorch.org/" target="_blank">
<span class="dropdown-title">Forums</span>
<p>A place to discuss PyTorch code, issues, install, research</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/hub">
<span class="dropdown-title">Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">GitHub</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='https://pytorch.org/docs/versions.html'>1.12 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="../community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption"><span class="caption-text">Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../notes/amp_examples.html">CUDA Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="../deploy.html">torch::deploy</a></li>
</ul>
<p class="caption"><span class="caption-text">Python API</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="../nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="../amp.html">torch.amp</a></li>
<li class="toctree-l1"><a class="reference internal" href="../autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="../library.html">torch.library</a></li>
<li class="toctree-l1"><a class="reference internal" href="../cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="../backends.html">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="../distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="../distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1 current"><a class="reference internal" href="../distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="../fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="../distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="../futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="../fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="../jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="../linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="../monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="../torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="../package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="../profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="../nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="../onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="../pipeline.html">Pipeline Parallelism</a></li>
<li class="toctree-l1"><a class="reference internal" href="../quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="../nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="../sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="../storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="../bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="../checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="../data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="../dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="../mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="../model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="../type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="../named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../config_mod.html">torch.__config__</a></li>
</ul>
<p class="caption"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="http://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../index.html">
Docs
</a> >
</li>
<li><a href="../distributed.elastic.html">Torch Distributed Elastic</a> ></li>
<li>Rendezvous</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/elastic/rendezvous.rst.txt" rel="nofollow"><img src="../_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="section" id="module-torch.distributed.elastic.rendezvous">
<span id="rendezvous"></span><span id="rendezvous-api"></span><h1>Rendezvous<a class="headerlink" href="#module-torch.distributed.elastic.rendezvous" title="Permalink to this headline">¶</a></h1>
<p>In the context of Torch Distributed Elastic we use the term <em>rendezvous</em> to
refer to a particular functionality that combines a <strong>distributed
synchronization</strong> primitive with <strong>peer discovery</strong>.</p>
<p>It is used by Torch Distributed Elastic to gather participants of a training
job (i.e. nodes) such that they all agree on the same list of participants and
everyone’s roles, as well as make a consistent collective decision on when
training can begin/resume.</p>
<p>Torch Distributed Elastic rendezvous provides the following critical
functionalities:</p>
<p><strong>Barrier</strong>:</p>
<p>Nodes performing rendezvous will all block until the rendezvous is considered
complete - this happens when at least <code class="docutils literal notranslate"><span class="pre">min</span></code> total number of nodes have joined
the rendezvous barrier (for the same job). This also implies the barrier is not
necessarily of fixed size.</p>
<p>There’s an additional small waiting time after reaching <code class="docutils literal notranslate"><span class="pre">min</span></code> number of
nodes - this is used to ensure the rendezvous is not completed “too quickly”
(which could potentially exclude additional nodes attempting to join at
approximately the same time).</p>
<p>If <code class="docutils literal notranslate"><span class="pre">max</span></code> number of nodes is gathered at the barrier, the rendezvous is
completed immediately.</p>
<p>There’s also an overall timeout which causes the rendezvous to fail if <code class="docutils literal notranslate"><span class="pre">min</span></code>
number of nodes is never reached - this is meant to be a simple fail-safe to
help release partially allocated job resources, in case there’s a problem with
the resource manager, and is meant to be interpreted as non-retryable.</p>
<p><strong>Exclusivity</strong>:</p>
<p>A simple distributed barrier would not be sufficient, as we also need to ensure
that only one group of nodes exists at any given time (for a given job). In
other words, new nodes (i.e. joining late) should not be able to form a parallel
independent group of workers for the same job.</p>
<p>Torch Distributed Elastic rendezvous ensures that if a group of nodes has
already completed a rendezvous (and hence might already be training), then
additional “late” nodes attempting to rendezvous will only announce themselves
as waiting, and will have to wait until the (previously completed) existing
rendezvous is destroyed first.</p>
<p><strong>Consistency</strong>:</p>
<p>When a rendezvous is completed, all its members will agree on the job membership
and everyone’s role in it. This role is represented using an integer, called
rank, that is between between 0 and world size.</p>
<p>Note that ranks are <em>not stable</em>, in the sense that the same node can be
assigned a different rank in the next (re-)rendezvous.</p>
<p><strong>Fault-tolerance</strong>:</p>
<p>Torch Distributed Elastic rendezvous is designed to tolerate node failures
during the rendezvous process. Should a process crash (or lose network
connectivity, etc), between joining the rendezvous and it being completed, then
a re-rendezvous with remaining healthy nodes will happen automatically.</p>
<p>A node can also fail <em>after</em> it has completed (or <em>has been observered</em> by other
nodes to have completed) the rendezvous - this scenario will be handled by the
Torch Distributed Elastic <code class="docutils literal notranslate"><span class="pre">train_loop</span></code> instead (where it will also trigger a
re-rendezvous).</p>
<p><strong>Shared key-value store</strong>:</p>
<p>When the rendezvous is completed, a shared key-value store is created and
returned. This store implements a <code class="docutils literal notranslate"><span class="pre">torch.distributed.Store</span></code> API (see
<a class="reference external" href="https://pytorch.org/docs/stable/distributed.html">distributed communication docs</a>).</p>
<p>This store is only shared by the members of the completed rendezvous. It
is intended to be used by Torch Distributed Elastic to exchange information
necessary to initialize job control and data-planes.</p>
<p><strong>Waiting workers and rendezvous closing</strong>:</p>
<p>Torch Distributed Elastic rendezvous handler object provides additional
functionalities, which are technically not part of the rendezvous process:</p>
<ol class="arabic simple">
<li><p>Querying how many workers arrived late at the barrier, who can participate in
<em>next</em> rendezvous.</p></li>
<li><p>Setting the rendezvous <em>closed</em> to signal all nodes not to participate in
next rendezvous.</p></li>
</ol>
<p><strong>DynamicRendezvousHandler</strong>:</p>
<p>Torch Distributed Elastic comes with the <a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">DynamicRendezvousHandler</span></code></a>
class that implements the rendezvous mechanism described above. It is a backend-
agnostic type that expects a particular <a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">RendezvousBackend</span></code></a> instance
to be specified during construction.</p>
<p>Torch distributed users can either implement their own backend type or use one
of the following implementations that come with PyTorch:</p>
<ul>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend" title="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">C10dRendezvousBackend</span></code></a>: Uses a C10d store (by default
<code class="docutils literal notranslate"><span class="pre">TCPStore</span></code>) as the rendezvous backend. The main advantage of using a C10d
store is that it requires no 3rd-party dependency (such as etcd) to establish
a rendezvous.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend" title="torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">EtcdRendezvousBackend</span></code></a>: Supersedes the legacy
<a class="reference internal" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous.EtcdRendezvousHandler" title="torch.distributed.elastic.rendezvous.etcd_rendezvous.EtcdRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">EtcdRendezvousHandler</span></code></a> class. Passing an
<a class="reference internal" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend" title="torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">EtcdRendezvousBackend</span></code></a> instance to
<a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">DynamicRendezvousHandler</span></code></a> is functionally equivalent to
instantiating an <a class="reference internal" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous.EtcdRendezvousHandler" title="torch.distributed.elastic.rendezvous.etcd_rendezvous.EtcdRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">EtcdRendezvousHandler</span></code></a>.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">store</span> <span class="o">=</span> <span class="n">TCPStore</span><span class="p">(</span><span class="s2">"localhost"</span><span class="p">)</span>
<span class="n">backend</span> <span class="o">=</span> <span class="n">C10dRendezvousBackend</span><span class="p">(</span><span class="n">store</span><span class="p">,</span> <span class="s2">"my_run_id"</span><span class="p">)</span>
<span class="n">rdzv_handler</span> <span class="o">=</span> <span class="n">DynamicRendezvousHandler</span><span class="o">.</span><span class="n">from_backend</span><span class="p">(</span>
<span class="n">run_id</span><span class="o">=</span><span class="s2">"my_run_id"</span><span class="p">,</span>
<span class="n">store</span><span class="o">=</span><span class="n">store</span><span class="p">,</span>
<span class="n">backend</span><span class="o">=</span><span class="n">backend</span><span class="p">,</span>
<span class="n">min_nodes</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span>
<span class="n">max_nodes</span><span class="o">=</span><span class="mi">4</span>
<span class="p">)</span>
</pre></div>
</div>
</li>
</ul>
<p>Below is a state diagram describing how rendezvous works.</p>
<img alt="../_images/etcd_rdzv_diagram.png" src="../_images/etcd_rdzv_diagram.png" />
<div class="section" id="registry">
<h2>Registry<a class="headerlink" href="#registry" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousParameters">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousParameters</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">backend</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">endpoint</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">run_id</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">min_nodes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">max_nodes</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousParameters"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousParameters" title="Permalink to this definition">¶</a></dt>
<dd><p>Holds the parameters to construct a <a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler" title="torch.distributed.elastic.rendezvous.RendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">RendezvousHandler</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>backend</strong> – The name of the backend to use to handle the rendezvous.</p></li>
<li><p><strong>endpoint</strong> – The endpoint of the rendezvous, usually in form <hostname>[:<port>].</p></li>
<li><p><strong>run_id</strong> – The id of the rendezvous.</p></li>
<li><p><strong>min_nodes</strong> – The minimum number of nodes to admit to the rendezvous.</p></li>
<li><p><strong>max_nodes</strong> – The maximum number of nodes to admit to the rendezvous.</p></li>
<li><p><strong>**kwargs</strong> – Additional parameters for the specified backend.</p></li>
</ul>
</dd>
</dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousParameters.get">
<code class="sig-name descname"><span class="pre">get</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">key</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">default</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousParameters.get"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousParameters.get" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the value for <code class="docutils literal notranslate"><span class="pre">key</span></code> if <code class="docutils literal notranslate"><span class="pre">key</span></code> exists, else <code class="docutils literal notranslate"><span class="pre">default</span></code>.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousParameters.get_as_bool">
<code class="sig-name descname"><span class="pre">get_as_bool</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">key</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">default</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousParameters.get_as_bool"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousParameters.get_as_bool" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the value for <code class="docutils literal notranslate"><span class="pre">key</span></code> as a <code class="docutils literal notranslate"><span class="pre">bool</span></code>.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousParameters.get_as_int">
<code class="sig-name descname"><span class="pre">get_as_int</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">key</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">default</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousParameters.get_as_int"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousParameters.get_as_int" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the value for <code class="docutils literal notranslate"><span class="pre">key</span></code> as an <code class="docutils literal notranslate"><span class="pre">int</span></code>.</p>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousHandlerRegistry</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandlerRegistry"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry" title="Permalink to this definition">¶</a></dt>
<dd><p>Represents a registry of <a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler" title="torch.distributed.elastic.rendezvous.RendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">RendezvousHandler</span></code></a> backends.</p>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry.create_handler">
<code class="sig-name descname"><span class="pre">create_handler</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandlerRegistry.create_handler"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry.create_handler" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates a new <a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler" title="torch.distributed.elastic.rendezvous.RendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">RendezvousHandler</span></code></a>.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry.register">
<code class="sig-name descname"><span class="pre">register</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">backend</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">creator</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandlerRegistry.register"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandlerRegistry.register" title="Permalink to this definition">¶</a></dt>
<dd><p>Registers a new rendezvous backend.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>backend</strong> – The name of the backend.</p></li>
<li><p><strong>creater</strong> – The callback to invoke to construct the
<a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler" title="torch.distributed.elastic.rendezvous.RendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">RendezvousHandler</span></code></a>.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
</dd></dl>
<span class="target" id="module-torch.distributed.elastic.rendezvous.registry"></span></div>
<div class="section" id="handler">
<h2>Handler<a class="headerlink" href="#handler" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousHandler</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler" title="Permalink to this definition">¶</a></dt>
<dd><p>Main rendezvous interface.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Distributed Torch users normally <strong>do not</strong> need to implement their own
<code class="docutils literal notranslate"><span class="pre">RendezvousHandler</span></code>. An implementation based on C10d Store is already
provided, and is recommended for most users.</p>
</div>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.get_backend">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">get_backend</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.get_backend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.get_backend" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the name of the rendezvous backend.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.get_run_id">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">get_run_id</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.get_run_id"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.get_run_id" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the run id of the rendezvous.</p>
<p>The run id is a user-defined id that uniquely identifies an instance of
a distributed application. It typically maps to a job id and is used to
allow nodes to join the correct distributed application.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.is_closed">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">is_closed</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.is_closed"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.is_closed" title="Permalink to this definition">¶</a></dt>
<dd><p>Checks whether the rendezvous has been closed.</p>
<p>A closed rendezvous means all future attempts to re-rendezvous within
same job will fail.</p>
<p><code class="docutils literal notranslate"><span class="pre">is_closed()</span></code> and <a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.set_closed" title="torch.distributed.elastic.rendezvous.RendezvousHandler.set_closed"><code class="xref py py-meth docutils literal notranslate"><span class="pre">set_closed()</span></code></a> have semantics of eventual
propagation and should not be used for synchronization. The intention is
that if at least one node decides the job is finished, it will close the
rendezvous, and other nodes will soon observe this and stop running as
well.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.next_rendezvous">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">next_rendezvous</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.next_rendezvous"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.next_rendezvous" title="Permalink to this definition">¶</a></dt>
<dd><p>Main entry-point into the rendezvous barrier.</p>
<p>Blocks until the rendezvous is complete and the current process is
included in the formed worker group, or a timeout occurs, or the
rendezvous was marked closed.</p>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>A tuple of <a class="reference internal" href="../distributed.html#torch.distributed.Store" title="torch.distributed.Store"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.Store</span></code></a>, <code class="docutils literal notranslate"><span class="pre">rank</span></code>, and
<code class="docutils literal notranslate"><span class="pre">world</span> <span class="pre">size</span></code>.</p>
</dd>
<dt class="field-even">Raises</dt>
<dd class="field-even"><ul class="simple">
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousClosedError" title="torch.distributed.elastic.rendezvous.RendezvousClosedError"><strong>RendezvousClosedError</strong></a> – The rendezvous is closed.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousConnectionError" title="torch.distributed.elastic.rendezvous.RendezvousConnectionError"><strong>RendezvousConnectionError</strong></a> – The connection to the rendezvous backend has failed.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousStateError" title="torch.distributed.elastic.rendezvous.RendezvousStateError"><strong>RendezvousStateError</strong></a> – The rendezvous state is corrupt.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousTimeoutError" title="torch.distributed.elastic.rendezvous.RendezvousTimeoutError"><strong>RendezvousTimeoutError</strong></a> – The rendezvous did not complete on time.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.num_nodes_waiting">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">num_nodes_waiting</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.num_nodes_waiting"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.num_nodes_waiting" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the number of nodes who arrived late at the rendezvous
barrier, hence were not included in the current worker group.</p>
<p>Callers should periodically call this method to check whether new
nodes are waiting to join the job and if so admit them by calling
<a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.next_rendezvous" title="torch.distributed.elastic.rendezvous.RendezvousHandler.next_rendezvous"><code class="xref py py-meth docutils literal notranslate"><span class="pre">next_rendezvous()</span></code></a> (re-rendezvous).</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.set_closed">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">set_closed</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.set_closed"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.set_closed" title="Permalink to this definition">¶</a></dt>
<dd><p>Marks the rendezvous as closed.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.RendezvousHandler.shutdown">
<code class="sig-name descname"><span class="pre">shutdown</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousHandler.shutdown"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousHandler.shutdown" title="Permalink to this definition">¶</a></dt>
<dd><p>Closes all resources that were open for the rendezvous.</p>
<p>Example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">rdzv_handler</span> <span class="o">=</span> <span class="o">...</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">store</span><span class="p">,</span> <span class="n">rank</span><span class="p">,</span> <span class="n">world_size</span> <span class="o">=</span> <span class="n">rdzv_handler</span><span class="o">.</span><span class="n">next_rendezvous</span><span class="p">()</span>
<span class="k">finally</span><span class="p">:</span>
<span class="n">rdzv_handler</span><span class="o">.</span><span class="n">shutdown</span><span class="p">()</span>
</pre></div>
</div>
</dd></dl>
</dd></dl>
</div>
<div class="section" id="exceptions">
<h2>Exceptions<a class="headerlink" href="#exceptions" title="Permalink to this headline">¶</a></h2>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousError">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousError</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousError"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousError" title="Permalink to this definition">¶</a></dt>
<dd><p>Represents the base type for rendezvous errors.</p>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousClosedError">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousClosedError</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousClosedError"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousClosedError" title="Permalink to this definition">¶</a></dt>
<dd><p>Raised when a rendezvous is closed.</p>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousTimeoutError">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousTimeoutError</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousTimeoutError"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousTimeoutError" title="Permalink to this definition">¶</a></dt>
<dd><p>Raised when a rendezvous did not complete on time.</p>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousConnectionError">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousConnectionError</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousConnectionError"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousConnectionError" title="Permalink to this definition">¶</a></dt>
<dd><p>Raised when the connection to a rendezvous backend has failed.</p>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.RendezvousStateError">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousStateError</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/api.html#RendezvousStateError"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.RendezvousStateError" title="Permalink to this definition">¶</a></dt>
<dd><p>Raised when the state of a rendezvous is corrupt.</p>
</dd></dl>
</div>
<div class="section" id="implementations">
<h2>Implementations<a class="headerlink" href="#implementations" title="Permalink to this headline">¶</a></h2>
<div class="section" id="dynamic-rendezvous">
<h3>Dynamic Rendezvous<a class="headerlink" href="#dynamic-rendezvous" title="Permalink to this headline">¶</a></h3>
<dl class="py function">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.create_handler">
<code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.dynamic_rendezvous.</span></code><code class="sig-name descname"><span class="pre">create_handler</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">store</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">backend</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">params</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#create_handler"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.create_handler" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates a new <a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">DynamicRendezvousHandler</span></code></a> from the specified
parameters.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>store</strong> – The C10d store to return as part of the rendezvous.</p></li>
<li><p><strong>backend</strong> – The backend to use to hold the rendezvous state.</p></li>
</ul>
</dd>
</dl>
<table class="docutils colwidths-auto align-default">
<thead>
<tr class="row-odd"><th class="head"><p>Parameter</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>join_timeout</p></td>
<td><p>The total time, in seconds, within which the
rendezvous is expected to complete. Defaults to 600
seconds.</p></td>
</tr>
<tr class="row-odd"><td><p>last_call_timeout</p></td>
<td><p>An additional wait amount, in seconds, before
completing the rendezvous once the minimum number of
nodes has been reached. Defaults to 30 seconds.</p></td>
</tr>
<tr class="row-even"><td><p>close_timeout</p></td>
<td><p>The time, in seconds, within which the rendezvous is
expected to close after a call to
<code class="xref py py-meth docutils literal notranslate"><span class="pre">RendezvousHandler.set_closed()</span></code> or
<code class="xref py py-meth docutils literal notranslate"><span class="pre">RendezvousHandler.shutdown()</span></code>. Defaults to
30 seconds.</p></td>
</tr>
</tbody>
</table>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.dynamic_rendezvous.</span></code><code class="sig-name descname"><span class="pre">DynamicRendezvousHandler</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#DynamicRendezvousHandler"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler" title="Permalink to this definition">¶</a></dt>
<dd><p>Represents a handler that sets up a rendezvous among a set of nodes.</p>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler.from_backend">
<em class="property"><span class="pre">classmethod</span> </em><code class="sig-name descname"><span class="pre">from_backend</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">run_id</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">store</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">backend</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">min_nodes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">max_nodes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">timeout</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#DynamicRendezvousHandler.from_backend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler.from_backend" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates a new <a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.DynamicRendezvousHandler"><code class="xref py py-class docutils literal notranslate"><span class="pre">DynamicRendezvousHandler</span></code></a>.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>run_id</strong> – The run id of the rendezvous.</p></li>
<li><p><strong>store</strong> – The C10d store to return as part of the rendezvous.</p></li>
<li><p><strong>backend</strong> – The backend to use to hold the rendezvous state.</p></li>
<li><p><strong>min_nodes</strong> – The minimum number of nodes to admit to the rendezvous.</p></li>
<li><p><strong>max_nodes</strong> – The maximum number of nodes to admit to the rendezvous.</p></li>
<li><p><strong>timeout</strong> – The timeout configuration of the rendezvous.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.dynamic_rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousBackend</span></code><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#RendezvousBackend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend" title="Permalink to this definition">¶</a></dt>
<dd><p>Represents a backend that holds the rendezvous state.</p>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.get_state">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">get_state</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#RendezvousBackend.get_state"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.get_state" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the rendezvous state.</p>
<dl class="field-list simple">
<dt class="field-odd">Returns</dt>
<dd class="field-odd"><p>A tuple of the encoded rendezvous state and its fencing token or
<code class="docutils literal notranslate"><span class="pre">None</span></code> if no state is found in the backend.</p>
</dd>
<dt class="field-even">Raises</dt>
<dd class="field-even"><ul class="simple">
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousConnectionError" title="torch.distributed.elastic.rendezvous.RendezvousConnectionError"><strong>RendezvousConnectionError</strong></a> – The connection to the backend has failed.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousStateError" title="torch.distributed.elastic.rendezvous.RendezvousStateError"><strong>RendezvousStateError</strong></a> – The rendezvous state is corrupt.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.name">
<em class="property"><span class="pre">abstract</span> <span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">name</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.name" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the name of the backend.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.set_state">
<em class="property"><span class="pre">abstract</span> </em><code class="sig-name descname"><span class="pre">set_state</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">state</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">token</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#RendezvousBackend.set_state"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.set_state" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the rendezvous state.</p>
<p>The new rendezvous state is set conditionally:</p>
<blockquote>
<div><ul class="simple">
<li><p>If the specified <code class="docutils literal notranslate"><span class="pre">token</span></code> matches the fencing token stored in the
backend, the state will be updated. The new state will be returned
to the caller along with its fencing token.</p></li>
<li><p>If the specified <code class="docutils literal notranslate"><span class="pre">token</span></code> does not match the fencing token stored
in the backend, the state won’t be updated; instead the existing
state along with its fencing token will be returned to the caller.</p></li>
<li><p>If the specified <code class="docutils literal notranslate"><span class="pre">token</span></code> is <code class="docutils literal notranslate"><span class="pre">None</span></code>, the new state will be set
only if there is no existing state in the backend. Either the new
state or the existing state along with its fencing token will be
returned to the caller.</p></li>
</ul>
</div></blockquote>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>state</strong> – The encoded rendezvous state.</p></li>
<li><p><strong>token</strong> – An optional fencing token that was retrieved by a previous call
to <a class="reference internal" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.get_state" title="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousBackend.get_state"><code class="xref py py-meth docutils literal notranslate"><span class="pre">get_state()</span></code></a> or <code class="docutils literal notranslate"><span class="pre">set_state()</span></code>.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>A tuple of the serialized rendezvous state, its fencing token, and
a boolean value indicating whether our set attempt succeeded.</p>
</dd>
<dt class="field-odd">Raises</dt>
<dd class="field-odd"><ul class="simple">
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousConnectionError" title="torch.distributed.elastic.rendezvous.RendezvousConnectionError"><strong>RendezvousConnectionError</strong></a> – The connection to the backend has failed.</p></li>
<li><p><a class="reference internal" href="#torch.distributed.elastic.rendezvous.RendezvousStateError" title="torch.distributed.elastic.rendezvous.RendezvousStateError"><strong>RendezvousStateError</strong></a> – The rendezvous state is corrupt.</p></li>
</ul>
</dd>
</dl>
</dd></dl>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.dynamic_rendezvous.</span></code><code class="sig-name descname"><span class="pre">RendezvousTimeout</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">join</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">last_call</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">close</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">heartbeat</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/dynamic_rendezvous.html#RendezvousTimeout"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout" title="Permalink to this definition">¶</a></dt>
<dd><p>Holds the timeout configuration of a rendezvous.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>join</strong> – The time within which the rendezvous is expected to complete.</p></li>
<li><p><strong>last_call</strong> – An additional wait amount before completing the rendezvous once the
rendezvous has the minimum number of required participants.</p></li>
<li><p><strong>close</strong> – The time within which the rendezvous is expected to close after a
call to <code class="xref py py-meth docutils literal notranslate"><span class="pre">RendezvousHandler.set_closed()</span></code> or
<code class="xref py py-meth docutils literal notranslate"><span class="pre">RendezvousHandler.shutdown()</span></code>.</p></li>
<li><p><strong>keep_alive</strong> – The time within which a keep-alive heartbeat is expected to
complete.</p></li>
</ul>
</dd>
</dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.close">
<em class="property"><span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">close</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.close" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the close timeout.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.heartbeat">
<em class="property"><span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">heartbeat</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.heartbeat" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the keep-alive heartbeat timeout.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.join">
<em class="property"><span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">join</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.join" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the join timeout.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.last_call">
<em class="property"><span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">last_call</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.dynamic_rendezvous.RendezvousTimeout.last_call" title="Permalink to this definition">¶</a></dt>
<dd><p>Gets the last call timeout.</p>
</dd></dl>
</dd></dl>
<div class="section" id="c10d-backend">
<h4>C10d Backend<a class="headerlink" href="#c10d-backend" title="Permalink to this headline">¶</a></h4>
<dl class="py function">
<dt id="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.create_backend">
<code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.</span></code><code class="sig-name descname"><span class="pre">create_backend</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.html#create_backend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.create_backend" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates a new <a class="reference internal" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend" title="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">C10dRendezvousBackend</span></code></a> from the specified
parameters.</p>
<table class="docutils colwidths-auto align-default">
<thead>
<tr class="row-odd"><th class="head"><p>Parameter</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>store_type</p></td>
<td><p>The type of the C10d store. The currently supported types
are “tcp” and “file” which correspond to
<a class="reference internal" href="../distributed.html#torch.distributed.TCPStore" title="torch.distributed.TCPStore"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.TCPStore</span></code></a> and
<a class="reference internal" href="../distributed.html#torch.distributed.FileStore" title="torch.distributed.FileStore"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.FileStore</span></code></a>, respectively.
Defaults to “tcp”.</p></td>
</tr>
<tr class="row-odd"><td><p>read_timeout</p></td>
<td><p>The read timeout, in seconds, for store operations.
Defaults to 60 seconds.</p>
<p>Note this only applies to
<a class="reference internal" href="../distributed.html#torch.distributed.TCPStore" title="torch.distributed.TCPStore"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.TCPStore</span></code></a>. It is not relevant
to <a class="reference internal" href="../distributed.html#torch.distributed.FileStore" title="torch.distributed.FileStore"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.FileStore</span></code></a> which does not
take in timeout as a parameter.</p>
</td>
</tr>
<tr class="row-even"><td><p>is_host</p></td>
<td><p>A boolean value indicating whether this backend instance
will host the C10d store. If not specified it will be
inferred heuristically by matching the hostname or the IP
address of this machine against the specified rendezvous
endpoint. Defaults to <code class="docutils literal notranslate"><span class="pre">None</span></code>.</p>
<p>Note that this configuration option only applies to
<a class="reference internal" href="../distributed.html#torch.distributed.TCPStore" title="torch.distributed.TCPStore"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.TCPStore</span></code></a>. In normal
circumstances you can safely skip it; the only time when
it is needed is if its value cannot be correctly
determined (e.g. the rendezvous endpoint has a CNAME as
the hostname or does not match the FQDN of the machine).</p>
</td>
</tr>
</tbody>
</table>
</dd></dl>
<dl class="py class">
<dt id="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend">
<em class="property"><span class="pre">class</span> </em><code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.</span></code><code class="sig-name descname"><span class="pre">C10dRendezvousBackend</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">store</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">run_id</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.html#C10dRendezvousBackend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend" title="Permalink to this definition">¶</a></dt>
<dd><p>Represents a C10d-backed rendezvous backend.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>store</strong> – The <a class="reference internal" href="../distributed.html#torch.distributed.Store" title="torch.distributed.Store"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.distributed.Store</span></code></a> instance to use to
communicate with the C10d store.</p></li>
<li><p><strong>run_id</strong> – The run id of the rendezvous.</p></li>
</ul>
</dd>
</dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.get_state">
<code class="sig-name descname"><span class="pre">get_state</span></code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.html#C10dRendezvousBackend.get_state"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.get_state" title="Permalink to this definition">¶</a></dt>
<dd><p>See base class.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.name">
<em class="property"><span class="pre">property</span> </em><code class="sig-name descname"><span class="pre">name</span></code><a class="headerlink" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.name" title="Permalink to this definition">¶</a></dt>
<dd><p>See base class.</p>
</dd></dl>
<dl class="py method">
<dt id="torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.set_state">
<code class="sig-name descname"><span class="pre">set_state</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">state</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">token</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.html#C10dRendezvousBackend.set_state"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.c10d_rendezvous_backend.C10dRendezvousBackend.set_state" title="Permalink to this definition">¶</a></dt>
<dd><p>See base class.</p>
</dd></dl>
</dd></dl>
</div>
<div class="section" id="etcd-backend">
<h4>Etcd Backend<a class="headerlink" href="#etcd-backend" title="Permalink to this headline">¶</a></h4>
<dl class="py function">
<dt id="torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.create_backend">
<code class="sig-prename descclassname"><span class="pre">torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.</span></code><code class="sig-name descname"><span class="pre">create_backend</span></code><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">params</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="../_modules/torch/distributed/elastic/rendezvous/etcd_rendezvous_backend.html#create_backend"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.create_backend" title="Permalink to this definition">¶</a></dt>
<dd><p>Creates a new <a class="reference internal" href="#torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend" title="torch.distributed.elastic.rendezvous.etcd_rendezvous_backend.EtcdRendezvousBackend"><code class="xref py py-class docutils literal notranslate"><span class="pre">EtcdRendezvousBackend</span></code></a> from the specified
parameters.</p>
<table class="docutils colwidths-auto align-default">
<thead>
<tr class="row-odd"><th class="head"><p>Parameter</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>read_timeout</p></td>
<td><p>The read timeout, in seconds, for etcd operations.
Defaults to 60 seconds.</p></td>
</tr>
<tr class="row-odd"><td><p>protocol</p></td>
<td><p>The protocol to use to communicate with etcd. Valid
values are “http” and “https”. Defaults to “http”.</p></td>
</tr>