| """Physical PP4 admission contract. Solo/compile receipts can never satisfy it.""" |
| from __future__ import annotations |
|
|
| import fcntl |
| import os |
| import re |
| from pathlib import Path |
|
|
| from offload.launch.windows import up2_checks as ck |
| from offload.prefill_maxlen.evidence import Json, record |
|
|
| SCHEMA = 'pp4-physical-pair-v1' |
| CASES = tuple((batch, depth, mode) for batch in range(1, 5) for depth in (4, 5) |
| for mode in ('sampled', 'mixed-even', 'mixed-odd', 'greedy')) |
|
|
|
|
| def guard() -> None: |
| ck.require(os.environ.get('WINDOW_COORDINATOR') == 'Deepseek/coordinator' |
| and os.environ.get('PP4_PAIR_QUALIFICATION') == '1', 'coordinator-only physical pair qualification') |
| |
| from offload.launch.windows.pf1_window import WORKER_GPUS |
| ck.require(os.environ.get('CUDA_VISIBLE_DEVICES') == WORKER_GPUS, 'physical production NVLink pair required') |
| for name in ('PP4_PAIR_LOCK_FD0', 'PP4_PAIR_LOCK_FD1'): |
| fcntl.flock(int(os.environ[name]), fcntl.LOCK_EX | fcntl.LOCK_NB) |
| import torch |
| ck.require(torch.cuda.device_count() == 2, 'two physical devices required; emulation refuses') |
| for rank, uuid in enumerate(WORKER_GPUS.split(',')): |
| ck.require(str(torch.cuda.get_device_properties(rank).uuid).removeprefix('GPU-') == uuid.removeprefix('GPU-'), |
| 'physical pair UUID differs') |
|
|
|
|
| def validate(data: dict[str, Json], commit: str, run: str) -> None: |
| from offload.launch.windows.pf1_window import WORKER_GPUS |
| ck.require(data.get('schema') == SCHEMA and data.get('source_commit') == commit and data.get('run') == run |
| and data.get('complete') is True and data.get('classification') == 'MEASURED physical pair', |
| 'physical-peer receipt identity/incomplete; solo is not qualification') |
| ck.require(data.get('gpus') == WORKER_GPUS, 'physical-peer GPU receipt mismatch') |
| rows = ck.rows(data['cases']) |
| ck.require([(r.get('batch'), r.get('depth'), r.get('mode')) for r in rows] == list(CASES), 'physical-peer case coverage differs') |
| for row in rows: |
| ck.require(row.get('rank0_exact') is True and row.get('rank1_exact') is True |
| and row.get('changed_input_graph_replays') == 10 |
| and re.fullmatch('[0-9a-f]{64}', str(row.get('solo_sha256'))) is not None, |
| 'physical-peer q/ID/graph identity failed') |
| for rank in (0, 1): |
| check = record(data[f'timeout{rank}']) |
| ck.require(check.get('missing_rank') == 1-rank and check.get('timeout_refused') is True |
| and check.get('sticky_retry_refused') is True and check.get('process_exited') is True, |
| 'physical-peer timeout/sticky-error teardown unproved') |
| worker = record(data['whole_worker']) |
| ck.require(worker.get('complete') is True and worker.get('physical_pair') is True |
| and worker.get('identity_checks') == 120 and worker.get('identity_exact') == 120 |
| and worker.get('archived_ids_exact') == 48 and worker.get('archived_confidence_exact') == 48, |
| 'physical whole-K2/N4 identity differs from solo receipt') |
| sources = record(data['source_sha256']) |
| ck.require(bool(sources) and all(re.fullmatch('[0-9a-f]{64}', str(v)) for v in sources.values()), 'physical-peer source hashes absent') |
|
|
|
|
| def checked_output(path: Path) -> None: |
| ck.require(path.resolve().is_relative_to(Path('/opt/dsv41/x86host-data/pp4')), |
| 'pair artifacts must stay in PP4 scratch') |
|
|
