-
Notifications
You must be signed in to change notification settings - Fork 27.4k
Broken test_ops_jit on Windows (CPU, and CUDA) #96858
Copy link
Copy link
Closed
Labels
module: ciRelated to continuous integrationRelated to continuous integrationmodule: windowsWindows support for PyTorchWindows support for PyTorchoncall: jitAdd this issue/PR to JIT oncall triage queueAdd this issue/PR to JIT oncall triage queuetriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Description
🐛 Describe the bug
This test starts to break on Windows. Many tests there are failing, and the jobs are close very close to timing out. At first glance, it doesn’t look like a PR causes it. The error points to opening a temp file:
2023-03-15T16:06:39.6597601Z ================================== FAILURES ===================================
2023-03-15T16:06:39.6597885Z _________ TestJitCPU.test_variant_consistency_jit_diff_cpu_complex64 __________
2023-03-15T16:06:39.6598150Z Traceback (most recent call last):
2023-03-15T16:06:39.6598490Z File "C:\actions-runner\_work\pytorch\pytorch\test\test_ops_jit.py", line 85, in test_variant_consistency_jit
2023-03-15T16:06:39.6598879Z self.indiv_variant_test_jit(device, dtype, op, sample, func_type, variant, has_fake_function)
2023-03-15T16:06:39.6599279Z File "C:\actions-runner\_work\pytorch\pytorch\test\test_ops_jit.py", line 136, in indiv_variant_test_jit
2023-03-15T16:06:39.6599585Z check_against_reference(self,
2023-03-15T16:06:39.6599950Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\common_jit.py", line 108, in check_against_reference
2023-03-15T16:06:39.6600379Z outputs_test = output_func(self.runAndSaveRNG(func, recording_inputs, kwargs))
2023-03-15T16:06:39.6600910Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\common_jit.py", line 160, in runAndSaveRNG
2023-03-15T16:06:39.6601259Z results = func(*inputs, **kwargs)
2023-03-15T16:06:39.6601687Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\jit_metaprogramming_utils.py", line 471, in traced_fn
2023-03-15T16:06:39.6602114Z self.assertExportImport(traced.graph, split_inputs.all_tensors)
2023-03-15T16:06:39.6602550Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\common_jit.py", line 148, in assertExportImport
2023-03-15T16:06:39.6602944Z self.assertExportImportModule(m, inputs)
2023-03-15T16:06:39.6603348Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\common_jit.py", line 151, in assertExportImportModule
2023-03-15T16:06:39.6603740Z m_import = self.getExportImportCopy(m)
2023-03-15T16:06:39.6604141Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\testing\_internal\common_jit.py", line 174, in getExportImportCopy
2023-03-15T16:06:39.6604525Z return torch.jit.load(fname, map_location=map_location)
2023-03-15T16:06:39.6604900Z File "C:\actions-runner\_work\pytorch\pytorch\build\win_tmp\build\torch\jit\_serialization.py", line 162, in load
2023-03-15T16:06:39.6605388Z cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files, _restore_shapes) # type: ignore[call-arg]
2023-03-15T16:06:39.6605886Z RuntimeError: open file failed because of errno 22 on fopen: Invalid argument, file path: C:\Users\RUNNER~1\AppData\Local\Temp\tmpbg1wa8z3
I will disable the test on Windows while we investigate.
Versions
Windows, PyTorch CI, CPU and CUDA
cc @EikanWang @jgong5 @wenzhe-nrv @sanchitintel @peterjc123 @mszhanyi @skyline75489 @nbcsm @seemethere @malfet @pytorch/pytorch-dev-infra @kit1980
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
module: ciRelated to continuous integrationRelated to continuous integrationmodule: windowsWindows support for PyTorchWindows support for PyTorchoncall: jitAdd this issue/PR to JIT oncall triage queueAdd this issue/PR to JIT oncall triage queuetriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Type
Projects
Status
Done