[Distributed] Fix extra context on device 0 #135273
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/135273
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 1 Unrelated FailureAs of commit c76c13c with merge base a1899b5 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
oncall: distributed
This is a partial fix to #135279 It moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. But it doesn't fully fix #135279 -- there seems to be extra context creation in destruction of Work objects too. cc wconstab fduwjj cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
fduwjj
approved these changes
Sep 6, 2024
Contributor
fduwjj
left a comment
fduwjj
left a comment
There was a problem hiding this comment.
LGTM, nice fix. For my own learning purpose, why do we also create CudaContext when removing Work?
wconstab
approved these changes
Sep 6, 2024
Contributor
wconstab
left a comment
wconstab
left a comment
There was a problem hiding this comment.
Was this extra context always happening or is it a regression after we fixed the issue for nan checker?
Collaborator
Author
This is a partial fix to #135279 It moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. But it doesn't fully fix #135279 -- there seems to be extra context creation in destruction of Work objects too. cc wconstab fduwjj cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This is a partial fix to #135279 It moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. But it doesn't fully fix #135279 -- there seems to be extra context creation in destruction of Work objects too. cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
kwen2501
changed the title
Collaborator
Author
This PR contains multiple fixes for issue #135279: ## Fix 1 Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. ## Fix 2 Additional context seems to be also created during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
Collaborator
|
That's great debugging ~~ , would we consider Fix 1 to be less invasive than Fix 2? It seems a bit more intuitive to think that we should have the current device set correctly at event creation time rather than assuming explicit EDIT: I misunderstood and it seems these fixes are for two separate causes. If there is something counterintuitive like a destruction of an event created when another device is set initializes a context on device 0 then that is something that we might want to follow up with the CUDA team on |
Collaborator
Author
|
@eqy Thanks for the review! Nothing counterintuitive so far :) |
yf225
added
ciflow/periodic
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
This PR contains multiple fixes for issue #135279: ## First part: Moves the GPU guard (`cudaSetDevice`) before the `currentStreamCaptureStatusMayInitCtx` call. As its name suggests, it May Init Ctx. ## Second part: Even with the above fix, additional contexts are still observed during Work object destruction, e.g. ``` work = dist.all_reduce(tensor, async_op=True) time.sleep(5) <-- no additional context yet del work <-- additional context shows up ``` ### Debug process Chasing it down to destruction of a `Future` object -- a member variable of `Work`. Then further down to the following member of `Future`: ``` std::vector<c10::Event> events_; ``` When the `events_` are destroyed, we hit the road down to: https://github.com/pytorch/pytorch/blob/1f3a79379012b408e0375e81fe9205dcba5e34ba/c10/cuda/impl/CUDAGuardImpl.h#L106-L121 When there is no "preset" CUDA context (**which is the case for python garbage collector**), line 112: `c10::cuda::GetDevice(&orig_device)` will set `orig_device` to 0. Then, at line 120, `c10::cuda::SetDevice(orig_device)` will "officially" set the context to device 0 -- **that's where rank 1, 2, ... can create extra context on device 0!** ### Solution This PR adds an explicit destructor to `Future`. In this destructor, destroy each event with a device guard. ## Test Added test_extra_cuda_context, implemented via - `pynvml` (if available), or - memory consumption check. `python test/distributed/test_c10d_nccl.py -k test_extra_cuda_context` cc XilunWu H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
Collaborator
Author
|
@pytorchbot merge -f "Failures are unrelated (1. test_mkl_verbose; 2. compile_time_instruction_count)" |
Collaborator
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
async_op=Truecollective if underallow_inflight_collective_as_graph_input_ctx()context manager #137763This PR contains multiple fixes for issue #135279:
First part:
Moves the GPU guard (
cudaSetDevice) before thecurrentStreamCaptureStatusMayInitCtxcall.As its name suggests, it May Init Ctx.
Second part:
Even with the above fix, additional contexts are still observed during Work object destruction, e.g.
Debug process
Chasing it down to destruction of a
Futureobject -- a member variable ofWork.Then further down to the following member of
Future:When the
events_are destroyed, we hit the road down to:pytorch/c10/cuda/impl/CUDAGuardImpl.h
Lines 106 to 121 in 1f3a793
When there is no "preset" CUDA context (which is the case for python garbage collector), line 112:
c10::cuda::GetDevice(&orig_device)will setorig_deviceto 0. Then, at line 120,c10::cuda::SetDevice(orig_device)will "officially" set the context to device 0 --that's where rank 1, 2, ... can create extra context on device 0!
Solution
This PR adds an explicit destructor to
Future. In this destructor, destroy each event with a device guard.Test
Added test_extra_cuda_context, implemented via
pynvml(if available), orpython test/distributed/test_c10d_nccl.py -k test_extra_cuda_contextcc @XilunWu @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o