forked from pytorch/pytorch.github.io
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmultiprocessing.html
860 lines (671 loc) · 43.8 KB
/
multiprocessing.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
<!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 charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Multiprocessing package - torch.multiprocessing — PyTorch master documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/multiprocessing.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="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="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<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">
</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>
<div class="ecosystem-dropdown">
<a id="dropdownMenuButton" data-toggle="ecosystem-dropdown">
Ecosystem
</a>
<div class="ecosystem-dropdown-menu">
<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>
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class=dropdown-title>Tools & Libraries</span>
<p>Explore the ecosystem of tools and libraries</p>
</a>
</div>
</div>
</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">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<div class="resources-dropdown">
<a id="resourcesDropdownButton" data-toggle="resources-dropdown">
Resources
</a>
<div class="resources-dropdown-menu">
<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://pytorch.org/features">
<span class=dropdown-title>About</span>
<p>Learn about PyTorch’s features and capabilities</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='http://pytorch.org/docs/versions.html'>1.4.0 ▼</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"><span class="caption-text">Notes</span></p>
<ul>
<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/distributed_autograd.html">Distributed Autograd Design</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/large_scale_deployments.html">Features for large-scale deployments</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/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="notes/rref.html">Remote Reference Protocol</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>
<li class="toctree-l1"><a class="reference external" href="http://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
<p class="caption"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/cppdocs/">C++ API</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
</ul>
<p class="caption"><span class="caption-text">Python API</span></p>
<ul>
<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="autograd.html">torch.autograd</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="distributed.html">torch.distributed</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="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="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="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="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="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="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__.html">torch.__config__</a></li>
</ul>
<p class="caption"><span class="caption-text">torchvision Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchvision/index.html">torchvision</a></li>
</ul>
<p class="caption"><span class="caption-text">torchaudio Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio">torchaudio</a></li>
</ul>
<p class="caption"><span class="caption-text">torchtext Reference</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text">torchtext</a></li>
</ul>
<p class="caption"><span class="caption-text">Community</span></p>
<ul>
<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/governance.html">PyTorch Governance</a></li>
<li class="toctree-l1"><a class="reference internal" href="community/persons_of_interest.html">PyTorch Governance | Persons of Interest</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>Multiprocessing package - torch.multiprocessing</li>
<li class="pytorch-breadcrumbs-aside">
<a href="_sources/multiprocessing.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.multiprocessing">
<span id="multiprocessing-package-torch-multiprocessing"></span><span id="multiprocessing-doc"></span><h1>Multiprocessing package - torch.multiprocessing<a class="headerlink" href="#module-torch.multiprocessing" title="Permalink to this headline">¶</a></h1>
<p>torch.multiprocessing is a wrapper around the native <code class="xref py py-mod docutils literal notranslate"><span class="pre">multiprocessing</span></code>
module. It registers custom reducers, that use shared memory to provide shared
views on the same data in different processes. Once the tensor/storage is moved
to shared_memory (see <a class="reference internal" href="tensors.html#torch.Tensor.share_memory_" title="torch.Tensor.share_memory_"><code class="xref py py-func docutils literal notranslate"><span class="pre">share_memory_()</span></code></a>), it will be possible
to send it to other processes without making any copies.</p>
<p>The API is 100% compatible with the original module - it’s enough to change
<code class="docutils literal notranslate"><span class="pre">import</span> <span class="pre">multiprocessing</span></code> to <code class="docutils literal notranslate"><span class="pre">import</span> <span class="pre">torch.multiprocessing</span></code> to have all the
tensors sent through the queues or shared via other mechanisms, moved to shared
memory.</p>
<p>Because of the similarity of APIs we do not document most of this package
contents, and we recommend referring to very good docs of the original module.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If the main process exits abruptly (e.g. because of an incoming signal),
Python’s <code class="docutils literal notranslate"><span class="pre">multiprocessing</span></code> sometimes fails to clean up its children.
It’s a known caveat, so if you’re seeing any resource leaks after
interrupting the interpreter, it probably means that this has just happened
to you.</p>
</div>
<div class="section" id="strategy-management">
<h2>Strategy management<a class="headerlink" href="#strategy-management" title="Permalink to this headline">¶</a></h2>
<dl class="function">
<dt id="torch.multiprocessing.get_all_sharing_strategies">
<code class="sig-prename descclassname">torch.multiprocessing.</code><code class="sig-name descname">get_all_sharing_strategies</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/multiprocessing.html#get_all_sharing_strategies"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.get_all_sharing_strategies" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns a set of sharing strategies supported on a current system.</p>
</dd></dl>
<dl class="function">
<dt id="torch.multiprocessing.get_sharing_strategy">
<code class="sig-prename descclassname">torch.multiprocessing.</code><code class="sig-name descname">get_sharing_strategy</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/multiprocessing.html#get_sharing_strategy"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.get_sharing_strategy" title="Permalink to this definition">¶</a></dt>
<dd><p>Returns the current strategy for sharing CPU tensors.</p>
</dd></dl>
<dl class="function">
<dt id="torch.multiprocessing.set_sharing_strategy">
<code class="sig-prename descclassname">torch.multiprocessing.</code><code class="sig-name descname">set_sharing_strategy</code><span class="sig-paren">(</span><em class="sig-param">new_strategy</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/multiprocessing.html#set_sharing_strategy"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.set_sharing_strategy" title="Permalink to this definition">¶</a></dt>
<dd><p>Sets the strategy for sharing CPU tensors.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>new_strategy</strong> (<em>str</em>) – Name of the selected strategy. Should be one of
the values returned by <a class="reference internal" href="#torch.multiprocessing.get_all_sharing_strategies" title="torch.multiprocessing.get_all_sharing_strategies"><code class="xref py py-func docutils literal notranslate"><span class="pre">get_all_sharing_strategies()</span></code></a>.</p>
</dd>
</dl>
</dd></dl>
</div>
<div class="section" id="sharing-cuda-tensors">
<span id="multiprocessing-cuda-sharing-details"></span><h2>Sharing CUDA tensors<a class="headerlink" href="#sharing-cuda-tensors" title="Permalink to this headline">¶</a></h2>
<p>Sharing CUDA tensors between processes is supported only in Python 3, using
a <code class="docutils literal notranslate"><span class="pre">spawn</span></code> or <code class="docutils literal notranslate"><span class="pre">forkserver</span></code> start methods. <code class="xref py py-mod docutils literal notranslate"><span class="pre">python:multiprocessing</span></code> in
Python 2 can only create subprocesses using <code class="docutils literal notranslate"><span class="pre">fork</span></code>, and it’s not supported
by the CUDA runtime.</p>
<p>Unlike CPU tensors, the sending process is required to keep the original tensor
as long as the receiving process retains a copy of the tensor. The refcounting is
implemented under the hood but requires users to follow the next best practices.</p>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>If the consumer process dies abnormally to a fatal signal, the shared tensor
could be forever kept in memory as long as the sending process is running.</p>
</div>
<ol class="arabic simple">
<li><p>Release memory ASAP in the consumer.</p></li>
</ol>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1">## Good</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="c1"># do somethings with x</span>
<span class="k">del</span> <span class="n">x</span>
</pre></div>
</div>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1">## Bad</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="c1"># do somethings with x</span>
<span class="c1"># do everything else (producer have to keep x in memory)</span>
</pre></div>
</div>
<p>2. Keep producer process running until all consumers exits. This will prevent
the situation when the producer process releasing memory which is still in use
by the consumer.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1">## producer</span>
<span class="c1"># send tensors, do something</span>
<span class="n">event</span><span class="o">.</span><span class="n">wait</span><span class="p">()</span>
</pre></div>
</div>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1">## consumer</span>
<span class="c1"># receive tensors and use them</span>
<span class="n">event</span><span class="o">.</span><span class="n">set</span><span class="p">()</span>
</pre></div>
</div>
<ol class="arabic simple" start="3">
<li><p>Don’t pass received tensors.</p></li>
</ol>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># not going to work</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="n">queue_2</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># you need to create a process-local copy</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
<span class="n">x_clone</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span>
<span class="n">queue_2</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">x_clone</span><span class="p">)</span>
</pre></div>
</div>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># putting and getting from the same queue in the same process will likely end up with segfault</span>
<span class="n">queue</span><span class="o">.</span><span class="n">put</span><span class="p">(</span><span class="n">tensor</span><span class="p">)</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">queue</span><span class="o">.</span><span class="n">get</span><span class="p">()</span>
</pre></div>
</div>
</div>
<div class="section" id="sharing-strategies">
<h2>Sharing strategies<a class="headerlink" href="#sharing-strategies" title="Permalink to this headline">¶</a></h2>
<p>This section provides a brief overview into how different sharing strategies
work. Note that it applies only to CPU tensor - CUDA tensors will always use
the CUDA API, as that’s the only way they can be shared.</p>
<div class="section" id="file-descriptor-file-descriptor">
<h3>File descriptor - <code class="docutils literal notranslate"><span class="pre">file_descriptor</span></code><a class="headerlink" href="#file-descriptor-file-descriptor" title="Permalink to this headline">¶</a></h3>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>This is the default strategy (except for macOS and OS X where it’s not
supported).</p>
</div>
<p>This strategy will use file descriptors as shared memory handles. Whenever a
storage is moved to shared memory, a file descriptor obtained from <code class="docutils literal notranslate"><span class="pre">shm_open</span></code>
is cached with the object, and when it’s going to be sent to other processes,
the file descriptor will be transferred (e.g. via UNIX sockets) to it. The
receiver will also cache the file descriptor and <code class="docutils literal notranslate"><span class="pre">mmap</span></code> it, to obtain a shared
view onto the storage data.</p>
<p>Note that if there will be a lot of tensors shared, this strategy will keep a
large number of file descriptors open most of the time. If your system has low
limits for the number of open file descriptors, and you can’t raise them, you
should use the <code class="docutils literal notranslate"><span class="pre">file_system</span></code> strategy.</p>
</div>
<div class="section" id="file-system-file-system">
<h3>File system - <code class="docutils literal notranslate"><span class="pre">file_system</span></code><a class="headerlink" href="#file-system-file-system" title="Permalink to this headline">¶</a></h3>
<p>This strategy will use file names given to <code class="docutils literal notranslate"><span class="pre">shm_open</span></code> to identify the shared
memory regions. This has a benefit of not requiring the implementation to cache
the file descriptors obtained from it, but at the same time is prone to shared
memory leaks. The file can’t be deleted right after its creation, because other
processes need to access it to open their views. If the processes fatally
crash, or are killed, and don’t call the storage destructors, the files will
remain in the system. This is very serious, because they keep using up the
memory until the system is restarted, or they’re freed manually.</p>
<p>To counter the problem of shared memory file leaks, <a class="reference internal" href="#module-torch.multiprocessing" title="torch.multiprocessing"><code class="xref py py-mod docutils literal notranslate"><span class="pre">torch.multiprocessing</span></code></a>
will spawn a daemon named <code class="docutils literal notranslate"><span class="pre">torch_shm_manager</span></code> that will isolate itself from
the current process group, and will keep track of all shared memory allocations.
Once all processes connected to it exit, it will wait a moment to ensure there
will be no new connections, and will iterate over all shared memory files
allocated by the group. If it finds that any of them still exist, they will be
deallocated. We’ve tested this method and it proved to be robust to various
failures. Still, if your system has high enough limits, and <code class="docutils literal notranslate"><span class="pre">file_descriptor</span></code>
is a supported strategy, we do not recommend switching to this one.</p>
</div>
</div>
<div class="section" id="spawning-subprocesses">
<h2>Spawning subprocesses<a class="headerlink" href="#spawning-subprocesses" title="Permalink to this headline">¶</a></h2>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>Available for Python >= 3.4.</p>
<p>This depends on the <code class="docutils literal notranslate"><span class="pre">spawn</span></code> start method in Python’s
<code class="docutils literal notranslate"><span class="pre">multiprocessing</span></code> package.</p>
</div>
<p>Spawning a number of subprocesses to perform some function can be done
by creating <code class="docutils literal notranslate"><span class="pre">Process</span></code> instances and calling <code class="docutils literal notranslate"><span class="pre">join</span></code> to wait for
their completion. This approach works fine when dealing with a single
subprocess but presents potential issues when dealing with multiple
processes.</p>
<p>Namely, joining processes sequentially implies they will terminate
sequentially. If they don’t, and the first process does not terminate,
the process termination will go unnoticed. Also, there are no native
facilities for error propagation.</p>
<p>The <code class="docutils literal notranslate"><span class="pre">spawn</span></code> function below addresses these concerns and takes care
of error propagation, out of order termination, and will actively
terminate processes upon detecting an error in one of them.</p>
<dl class="function">
<dt id="torch.multiprocessing.spawn">
<code class="sig-prename descclassname">torch.multiprocessing.</code><code class="sig-name descname">spawn</code><span class="sig-paren">(</span><em class="sig-param">fn</em>, <em class="sig-param">args=()</em>, <em class="sig-param">nprocs=1</em>, <em class="sig-param">join=True</em>, <em class="sig-param">daemon=False</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/multiprocessing/spawn.html#spawn"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.spawn" title="Permalink to this definition">¶</a></dt>
<dd><p>Spawns <code class="docutils literal notranslate"><span class="pre">nprocs</span></code> processes that run <code class="docutils literal notranslate"><span class="pre">fn</span></code> with <code class="docutils literal notranslate"><span class="pre">args</span></code>.</p>
<p>If one of the processes exits with a non-zero exit status, the
remaining processes are killed and an exception is raised with the
cause of termination. In the case an exception was caught in the
child process, it is forwarded and its traceback is included in
the exception raised in the parent process.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><ul class="simple">
<li><p><strong>fn</strong> (<em>function</em>) – <p>Function is called as the entrypoint of the
spawned process. This function must be defined at the top
level of a module so it can be pickled and spawned. This
is a requirement imposed by multiprocessing.</p>
<p>The function is called as <code class="docutils literal notranslate"><span class="pre">fn(i,</span> <span class="pre">*args)</span></code>, where <code class="docutils literal notranslate"><span class="pre">i</span></code> is
the process index and <code class="docutils literal notranslate"><span class="pre">args</span></code> is the passed through tuple
of arguments.</p>
</p></li>
<li><p><strong>args</strong> (<em>tuple</em>) – Arguments passed to <code class="docutils literal notranslate"><span class="pre">fn</span></code>.</p></li>
<li><p><strong>nprocs</strong> (<em>python:int</em>) – Number of processes to spawn.</p></li>
<li><p><strong>join</strong> (<em>bool</em>) – Perform a blocking join on all processes.</p></li>
<li><p><strong>daemon</strong> (<em>bool</em>) – The spawned processes’ daemon flag. If set to True,
daemonic processes will be created.</p></li>
</ul>
</dd>
<dt class="field-even">Returns</dt>
<dd class="field-even"><p>None if <code class="docutils literal notranslate"><span class="pre">join</span></code> is <code class="docutils literal notranslate"><span class="pre">True</span></code>,
<a class="reference internal" href="#torch.multiprocessing.SpawnContext" title="torch.multiprocessing.SpawnContext"><code class="xref py py-class docutils literal notranslate"><span class="pre">SpawnContext</span></code></a> if <code class="docutils literal notranslate"><span class="pre">join</span></code> is <code class="docutils literal notranslate"><span class="pre">False</span></code></p>
</dd>
</dl>
</dd></dl>
<dl class="class">
<dt id="torch.multiprocessing.SpawnContext">
<em class="property">class </em><code class="sig-prename descclassname">torch.multiprocessing.</code><code class="sig-name descname">SpawnContext</code><a class="reference internal" href="_modules/torch/multiprocessing/spawn.html#SpawnContext"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.SpawnContext" title="Permalink to this definition">¶</a></dt>
<dd><p>Returned by <a class="reference internal" href="#torch.multiprocessing.spawn" title="torch.multiprocessing.spawn"><code class="xref py py-func docutils literal notranslate"><span class="pre">spawn()</span></code></a> when called with <code class="docutils literal notranslate"><span class="pre">join=False</span></code>.</p>
<dl class="method">
<dt id="torch.multiprocessing.SpawnContext.join">
<code class="sig-name descname">join</code><span class="sig-paren">(</span><em class="sig-param">timeout=None</em><span class="sig-paren">)</span><a class="reference internal" href="_modules/torch/multiprocessing/spawn.html#SpawnContext.join"><span class="viewcode-link">[source]</span></a><a class="headerlink" href="#torch.multiprocessing.SpawnContext.join" title="Permalink to this definition">¶</a></dt>
<dd><p>Tries to join one or more processes in this spawn context.
If one of them exited with a non-zero exit status, this function
kills the remaining processes and raises an exception with the cause
of the first process exiting.</p>
<p>Returns <code class="docutils literal notranslate"><span class="pre">True</span></code> if all processes have been joined successfully,
<code class="docutils literal notranslate"><span class="pre">False</span></code> if there are more processes that need to be joined.</p>
<dl class="field-list simple">
<dt class="field-odd">Parameters</dt>
<dd class="field-odd"><p><strong>timeout</strong> (<em>python:float</em>) – Wait this long before giving up on waiting.</p>
</dd>
</dl>
</dd></dl>
</dd></dl>
</div>
</div>
</article>
</div>
<footer>
<hr>
<div role="contentinfo">
<p>
© Copyright 2019, Torch Contributors.
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
</div>
<div class="pytorch-content-right" id="pytorch-content-right">
<div class="pytorch-right-menu" id="pytorch-right-menu">
<div class="pytorch-side-scroll" id="pytorch-side-scroll-right">
<ul>
<li><a class="reference internal" href="#">Multiprocessing package - torch.multiprocessing</a><ul>
<li><a class="reference internal" href="#strategy-management">Strategy management</a></li>
<li><a class="reference internal" href="#sharing-cuda-tensors">Sharing CUDA tensors</a></li>
<li><a class="reference internal" href="#sharing-strategies">Sharing strategies</a><ul>
<li><a class="reference internal" href="#file-descriptor-file-descriptor">File descriptor - <code class="docutils literal notranslate"><span class="pre">file_descriptor</span></code></a></li>
<li><a class="reference internal" href="#file-system-file-system">File system - <code class="docutils literal notranslate"><span class="pre">file_system</span></code></a></li>
</ul>
</li>
<li><a class="reference internal" href="#spawning-subprocesses">Spawning subprocesses</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</section>
</div>
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script src="_static/language_data.js"></script>
<script type="text/javascript" src="_static/js/vendor/popper.min.js"></script>
<script type="text/javascript" src="_static/js/vendor/bootstrap.min.js"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-90545585-1', 'auto');
ga('send', 'pageview');
</script>
<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>
<img height="1" width="1" style="border-style:none;" alt="" src="https://www.googleadservices.com/pagead/conversion/795629140/?label=txkmCPmdtosBENSssfsC&guid=ON&script=0"/>
<!-- Begin Footer -->
<div class="container-fluid docs-tutorials-resources" id="docs-tutorials-resources">
<div class="container">
<div class="row">
<div class="col-md-4 text-center">
<h2>Docs</h2>
<p>Access comprehensive developer documentation for PyTorch</p>
<a class="with-right-arrow" href="https://pytorch.org/docs/stable/index.html">View Docs</a>
</div>
<div class="col-md-4 text-center">
<h2>Tutorials</h2>
<p>Get in-depth tutorials for beginners and advanced developers</p>
<a class="with-right-arrow" href="https://pytorch.org/tutorials">View Tutorials</a>
</div>
<div class="col-md-4 text-center">
<h2>Resources</h2>
<p>Find development resources and get your questions answered</p>
<a class="with-right-arrow" href="https://pytorch.org/resources">View Resources</a>
</div>
</div>
</div>
</div>
<footer class="site-footer">
<div class="container footer-container">
<div class="footer-logo-wrapper">
<a href="https://pytorch.org/" class="footer-logo"></a>
</div>
<div class="footer-links-wrapper">
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/">PyTorch</a></li>
<li><a href="https://pytorch.org/get-started">Get Started</a></li>
<li><a href="https://pytorch.org/features">Features</a></li>
<li><a href="https://pytorch.org/ecosystem">Ecosystem</a></li>
<li><a href="https://pytorch.org/blog/">Blog</a></li>
<li><a href="https://pytorch.org/resources">Resources</a></li>
</ul>
</div>
<div class="footer-links-col">
<ul>
<li class="list-title"><a href="https://pytorch.org/support">Support</a></li>
<li><a href="https://pytorch.org/tutorials">Tutorials</a></li>
<li><a href="https://pytorch.org/docs/stable/index.html">Docs</a></li>
<li><a href="https://discuss.pytorch.org" target="_blank">Discuss</a></li>
<li><a href="https://github.com/pytorch/pytorch/issues" target="_blank">Github Issues</a></li>
<li><a href="https://pytorch.slack.com" target="_blank">Slack</a></li>
<li><a href="https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md" target="_blank">Contributing</a></li>
</ul>
</div>
<div class="footer-links-col follow-us-col">
<ul>
<li class="list-title">Follow Us</li>
<li>
<div id="mc_embed_signup">
<form
action="https://twitter.us14.list-manage.com/subscribe/post?u=75419c71fe0a935e53dfa4a3f&id=91d0dccd39"
method="post"
id="mc-embedded-subscribe-form"
name="mc-embedded-subscribe-form"
class="email-subscribe-form validate"
target="_blank"
novalidate>
<div id="mc_embed_signup_scroll" class="email-subscribe-form-fields-wrapper">
<div class="mc-field-group">
<label for="mce-EMAIL" style="display:none;">Email Address</label>
<input type="email" value="" name="EMAIL" class="required email" id="mce-EMAIL" placeholder="Email Address">
</div>
<div id="mce-responses" class="clear">
<div class="response" id="mce-error-response" style="display:none"></div>
<div class="response" id="mce-success-response" style="display:none"></div>
</div> <!-- real people should not fill this in and expect good things - do not remove this or risk form bot signups-->
<div style="position: absolute; left: -5000px;" aria-hidden="true"><input type="text" name="b_75419c71fe0a935e53dfa4a3f_91d0dccd39" tabindex="-1" value=""></div>
<div class="clear">
<input type="submit" value="" name="subscribe" id="mc-embedded-subscribe" class="button email-subscribe-button">
</div>
</div>
</form>
</div>
</li>
</ul>
<div class="footer-social-icons">
<a href="https://www.facebook.com/pytorch" target="_blank" class="facebook"></a>
<a href="https://twitter.com/pytorch" target="_blank" class="twitter"></a>
</div>
</div>
</div>
</div>
</footer>
<div class="cookie-banner-wrapper">
<div class="container">
<p class="gdpr-notice">To analyze traffic and optimize your experience, we serve cookies on this site. By clicking or navigating, you agree to allow our usage of cookies. As the current maintainers of this site, Facebook’s Cookies Policy applies. Learn more, including about available controls: <a href="https://www.facebook.com/policies/cookies/">Cookies Policy</a>.</p>
<img class="close-button" src="_static/images/pytorch-x.svg">
</div>
</div>
<!-- End Footer -->
<!-- Begin Mobile Menu -->
<div class="mobile-main-menu">
<div class="container-fluid">
<div class="container">
<div class="mobile-main-menu-header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<a class="main-menu-close-button" href="#" data-behavior="close-mobile-menu"></a>
</div>
</div>
</div>
<div class="mobile-main-menu-links-container">
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/features">Features</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/hub">PyTorch Hub</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">
<a href="https://pytorch.org/docs/stable/index.html">Docs</a>
</li>
<li>
<a href="https://pytorch.org/resources">Resources</a>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">Github</a>
</li>
</ul>
</div>
</div>
</div>
<!-- End Mobile Menu -->
<script type="text/javascript" src="_static/js/vendor/anchor.min.js"></script>
<script type="text/javascript">
$(document).ready(function() {
mobileMenu.bind();
mobileTOC.bind();
pytorchAnchors.bind();
sideMenus.bind();
scrollToAnchor.bind();
highlightNavigation.bind();
mainMenuDropdown.bind();
// Add class to links that have code blocks, since we cannot create links in code blocks
$("article.pytorch-article a span.pre").each(function(e) {
$(this).closest("a").addClass("has-code");
});
})
</script>
</body>
</html>