| """Lease validation shared by GPU entrypoints. No service/power mutation.""" |
| from __future__ import annotations |
|
|
| import fcntl |
| import os |
| from pathlib import Path |
| import subprocess |
|
|
| GPU = "GPU-240876ad-ea42-47de-1b8d-f6153810b7e1" |
| SOLO_B = "GPU-bc62df2a-59cf-9828-7d43-558d1355bab7" |
|
|
|
|
| def leased_gpu() -> str: |
| selected = os.environ.get("CUDA_VISIBLE_DEVICES", "") |
| allowed = (GPU, SOLO_B) if os.environ.get("D2_PILOT_LEASE") == "1" else (GPU,) |
| if selected not in allowed: |
| raise RuntimeError(f"CUDA_VISIBLE_DEVICES must equal a permitted solo UUID: {allowed}") |
| return selected |
|
|
|
|
| def assert_gpu_lease() -> None: |
| selected = leased_gpu() |
| fd = int(os.environ.get("D2_LOCK_FD", "-1")) |
| expected = Path.home()/f"dsv41-offload/locks/gpu-{selected.split('-')[1]}.lock" |
| if fd < 0 or (os.fstat(fd).st_dev, os.fstat(fd).st_ino) != (expected.stat().st_dev, expected.stat().st_ino): |
| raise RuntimeError("missing inherited D2 exclusive GPU lock") |
| |
| with expected.open("a") as probe: |
| try: |
| fcntl.flock(probe.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB) |
| except BlockingIOError: |
| try: |
| fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB) |
| except BlockingIOError as exc: |
| raise RuntimeError("inherited descriptor does not own the GPU lock") from exc |
| return |
| fcntl.flock(probe.fileno(), fcntl.LOCK_UN) |
| raise RuntimeError("GPU lock descriptor is open but not held") |
|
|
|
|
| def power_logger_running() -> bool: |
| return subprocess.run(["tmux", "has-session", "-t", "=dsv41-powerlog"], capture_output=True).returncode == 0 |
|
|
