(
rank: int, device_uuid: str, named_tensors: dict[str, torch.Tensor], queue: Queue
)
| 50 | |
| 51 | |
| 52 | def checker_proc_with_error( |
| 53 | rank: int, device_uuid: str, named_tensors: dict[str, torch.Tensor], queue: Queue |
| 54 | ): |
| 55 | device_manager.device_module.set_device(rank) |
| 56 | named_tensors = { |
| 57 | name: tensor.to(device_manager.device_type) for name, tensor in named_tensors.items() |
| 58 | } |
| 59 | _ = named_tensors |
| 60 | _zmq_ctx = zmq.Context() |
| 61 | |
| 62 | def trigger_error(socket_paths: list[tuple[str, str]]): |
| 63 | socket_paths = dict(socket_paths) |
| 64 | update_weights_from_ipc( |
| 65 | _zmq_ctx, |
| 66 | socket_paths[device_uuid], |
| 67 | device_id=rank, |
| 68 | run=error_run, |
| 69 | post_hook=lambda: device_manager.device_module.synchronize(), |
| 70 | ) |
| 71 | |
| 72 | def error_run(weights: list[tuple[str, torch.Tensor]]): |
| 73 | _ = weights # Do some fake processing |
| 74 | time.sleep(random.uniform(0.1, 0.5)) |
| 75 | if rank == 0: |
| 76 | raise RuntimeError("Intentional Error for testing.") |
| 77 | |
| 78 | while True: |
| 79 | socket_paths: list[tuple[str, str]] = queue.get() |
| 80 | if socket_paths is None: |
| 81 | break |
| 82 | try: |
| 83 | trigger_error(socket_paths) |
| 84 | except RuntimeError as e: |
| 85 | assert str(e) == "Some workers failed to update weights" |
| 86 | |
| 87 | |
| 88 | def checker_proc(rank: int, device_uuid: str, named_tensors: dict[str, torch.Tensor], queue: Queue): |
nothing calls this directly
no test coverage detected