Add read-only normal-run progress audit and verified startup observation
This commit is contained in:
@@ -7,6 +7,11 @@
|
|||||||
4,080 份固定 CS 输入已生成并由真实进程退出记录及账本确认。
|
4,080 份固定 CS 输入已生成并由真实进程退出记录及账本确认。
|
||||||
独立后台执行器已启动;尚未据此宣称训练完成或学习成功。
|
独立后台执行器已启动;尚未据此宣称训练完成或学习成功。
|
||||||
|
|
||||||
|
启动后观察到 4 个训练 epoch 已完成原更新及独立 CPU 校验、6 次独立
|
||||||
|
probe 已提交;后续仍在运行。这是有时间边界的启动快照,不是当前实时
|
||||||
|
累计数,也不是科学 gate 结果。首轮实际 CPU 校验确认使用封存 spike events
|
||||||
|
重算原 U 更新,父进程记录退出码为 0。
|
||||||
|
|
||||||
- 模型:`/data/digital-drosophila/models/male-cns-lif/normal-conditioning-v1/`
|
- 模型:`/data/digital-drosophila/models/male-cns-lif/normal-conditioning-v1/`
|
||||||
- 实验:`/data/digital-drosophila/experiments/phase4ar-normal-conditioning/`
|
- 实验:`/data/digital-drosophila/experiments/phase4ar-normal-conditioning/`
|
||||||
- 管理日志:`/data/digital-drosophila/logs/phase4ar-normal-conditioning/`
|
- 管理日志:`/data/digital-drosophila/logs/phase4ar-normal-conditioning/`
|
||||||
@@ -50,3 +55,14 @@ delayed reward 或任何游戏、动作解码器、RL、视觉、Web/3D。
|
|||||||
完整性核验与 CPU 测试是工程证据,不是新数值回归,更不是学习证据。
|
完整性核验与 CPU 测试是工程证据,不是新数值回归,更不是学习证据。
|
||||||
最终必须另行进行全产物审计、固定学习 gate 分析及旧阶段完整性复核。
|
最终必须另行进行全产物审计、固定学习 gate 分析及旧阶段完整性复核。
|
||||||
仓库同步源码/文档/小型证据,不等于大体量实验数据已经异机备份。
|
仓库同步源码/文档/小型证据,不等于大体量实验数据已经异机备份。
|
||||||
|
|
||||||
|
## 后续只读检查
|
||||||
|
|
||||||
|
已安排本线程每小时检查,仅在实质进展、异常、完成或需人工操作时提醒。
|
||||||
|
不自动重试、修补已注册源码、进入后续 assay 或声明学习成功。
|
||||||
|
|
||||||
|
独立只读工具位于 `tools/normal-conditioning-audit/progress_normal.py`,
|
||||||
|
8 项临时数据测试在服务器通过。它读取实际提交账本与最新退出记录,
|
||||||
|
不读取大体量脑输出;其进度结果不是完整性报告或科学结论。
|
||||||
|
正在执行的 child 在没有真实退出记录时明确保持 UNKNOWN,不能由 PID
|
||||||
|
或一个提前写出的 results 文件推定成功。完整 run 完成后仍需最终独立审计。
|
||||||
|
|||||||
@@ -0,0 +1,29 @@
|
|||||||
|
{
|
||||||
|
"schema": "phase4ar-startup-observation-v1",
|
||||||
|
"status": "RUNNING-NOT-LEARNING-PASS",
|
||||||
|
"registration_sha256": "b63e6281b6505ca3b60acb467dd0b3fae274cf513dcbfd653649ba52e9ba9740",
|
||||||
|
"scientific_source_commit": "80aeb2093db655329774eff7b32fb6e417b8b40b",
|
||||||
|
"historical_integrity_before_complete": true,
|
||||||
|
"CPU_source_tests": {"tests": 73, "failures": 0, "errors": 0, "skips": 0},
|
||||||
|
"metadata_reader_tests": {"tests": 8, "failures": 0, "errors": 0, "skips": 0, "wall_seconds": 0.032},
|
||||||
|
"progress_tool_sha256": "081d5de098c9a249a13394f9669c76d79dd421ca28765397d3a9d8bbbde051d0",
|
||||||
|
"progress_tool_test_sha256": "f1a34c8d0933cb7226cbeea8afed7321df771e6ad8b1979072af13a5e5bb3c9c",
|
||||||
|
"observed_committed_process_entries": 19,
|
||||||
|
"observed_completed_training_epochs": 4,
|
||||||
|
"observed_completed_probes": 6,
|
||||||
|
"sampler_committed": true,
|
||||||
|
"first_epoch_independent_CPU_validation": {
|
||||||
|
"task_id": "physiological__PAM15__reinforce_A__rep01/normal/paired/000",
|
||||||
|
"actual_parent_exit_code": 0,
|
||||||
|
"frozen_U_verify_receipt_recomputed": true,
|
||||||
|
"actual_sealed_events_used": true,
|
||||||
|
"brain_finite": true
|
||||||
|
},
|
||||||
|
"learning_success_claimed": false,
|
||||||
|
"candidate_generated": false,
|
||||||
|
"snapshot_scope": "startup metadata observations only, not complete final integrity or current real-time progress",
|
||||||
|
"experiment_snapshot_is_final_seal": false,
|
||||||
|
"automatic_retry_authorized": false,
|
||||||
|
"heartbeat_id": "phase-4ar",
|
||||||
|
"heartbeat_scope": "hourly read-only progress, meaningful changes only; never execute or retry experiments"
|
||||||
|
}
|
||||||
@@ -0,0 +1,271 @@
|
|||||||
|
"""Read-only metadata progress, NOT integrity, a scientific gate, or liveness.
|
||||||
|
|
||||||
|
One CLI invocation reads registration, small ledger-entry JSON files and file
|
||||||
|
names. Only the newest committed entry and pending child's small process/result
|
||||||
|
receipts are inspected. No model imports, array loads, neural execution, full
|
||||||
|
artifact SHA walk, PID inference, retries, writes or persistent cache.
|
||||||
|
"""
|
||||||
|
from __future__ import annotations
|
||||||
|
|
||||||
|
import argparse
|
||||||
|
from collections import Counter
|
||||||
|
import hashlib
|
||||||
|
import json
|
||||||
|
from pathlib import Path
|
||||||
|
import re
|
||||||
|
|
||||||
|
|
||||||
|
ENTRY_ID = re.compile(r'(?:input-generation|task-[0-9a-f]{64}--(?:brain|update|verify))\Z')
|
||||||
|
RESULT_NAME = re.compile(r'task-[0-9a-f]{64}\.json\Z')
|
||||||
|
SEQUENCE_NAME = re.compile(r'[0-9]{16}\.json\Z')
|
||||||
|
SHA = re.compile(r'[0-9a-f]{64}\Z')
|
||||||
|
MAX_METADATA_BYTES = 64 * 1024 * 1024
|
||||||
|
|
||||||
|
|
||||||
|
def require(ok, message):
|
||||||
|
if not ok:
|
||||||
|
raise ValueError(message)
|
||||||
|
|
||||||
|
|
||||||
|
def safe(path):
|
||||||
|
path = Path(path)
|
||||||
|
require(path.is_absolute() and '..' not in path.parts and not any(p.is_symlink() or
|
||||||
|
(hasattr(p, 'is_junction') and p.is_junction()) for p in (path, *path.parents)), 'Unsafe linked/nonabsolute path')
|
||||||
|
return path
|
||||||
|
|
||||||
|
|
||||||
|
def read_json(path):
|
||||||
|
path = safe(path)
|
||||||
|
require(path.is_file() and path.stat().st_size <= MAX_METADATA_BYTES, 'Missing/oversized metadata: ' + str(path))
|
||||||
|
raw = path.read_bytes()
|
||||||
|
def unique(items):
|
||||||
|
result = {}
|
||||||
|
for key, value in items:
|
||||||
|
require(key not in result, 'Duplicate JSON field'); result[key] = value
|
||||||
|
return result
|
||||||
|
value = json.loads(raw.decode('utf-8'), object_pairs_hook=unique,
|
||||||
|
parse_constant=lambda _: (_ for _ in ()).throw(ValueError('Nonfinite JSON metadata')))
|
||||||
|
require(type(value) is dict, 'Metadata object required')
|
||||||
|
return value, raw
|
||||||
|
|
||||||
|
|
||||||
|
def directory_files(path):
|
||||||
|
path = safe(path)
|
||||||
|
if not path.exists():
|
||||||
|
return []
|
||||||
|
require(path.is_dir(), 'Expected metadata directory')
|
||||||
|
rows = sorted(path.iterdir())
|
||||||
|
for item in rows:
|
||||||
|
require(safe(item).is_file(), 'Unexpected directory/nonregular metadata child: ' + str(item))
|
||||||
|
return rows
|
||||||
|
|
||||||
|
|
||||||
|
def named_record(path, regsha):
|
||||||
|
if not path.exists():
|
||||||
|
return None
|
||||||
|
value, _ = read_json(path)
|
||||||
|
require(value.get('registration_sha256') == regsha, 'Record registration differs: ' + str(path))
|
||||||
|
return value
|
||||||
|
|
||||||
|
|
||||||
|
def process_record(exp, entry_id, regsha, *, expected_sha=None, job_sha=None):
|
||||||
|
path = exp / 'receipts' / entry_id / 'process.json'
|
||||||
|
if not path.exists():
|
||||||
|
return dict(path=str(path), present=False, actual_exit='UNKNOWN')
|
||||||
|
value, raw = read_json(path)
|
||||||
|
require(value.get('entry_id') == entry_id and value.get('registration_sha256') == regsha and
|
||||||
|
type(value.get('complete')) is bool and type(value.get('timed_out')) is bool and
|
||||||
|
type(value.get('exit_observed_by_parent')) is bool and
|
||||||
|
(value.get('exit_code') is None or type(value.get('exit_code')) is int), 'Incomplete parent process receipt')
|
||||||
|
require(SHA.fullmatch(value.get('job_sha256', '')) and (job_sha is None or value['job_sha256'] == job_sha), 'Parent job binding differs')
|
||||||
|
require(expected_sha is None or hashlib.sha256(raw).hexdigest() == expected_sha, 'Latest committed process receipt SHA differs')
|
||||||
|
code = value['exit_code']
|
||||||
|
require(code is None or value['exit_observed_by_parent'] is True, 'Exit code lacks parent observation')
|
||||||
|
require(not value['complete'] or (type(code) is int and code == 0 and not value['timed_out']), 'Completion contradicts recorded exit')
|
||||||
|
return dict(path=str(path), present=True, exit_code=code, timed_out=value['timed_out'],
|
||||||
|
exit_observed_by_parent=value['exit_observed_by_parent'], complete=value['complete'],
|
||||||
|
actual_exit='RECORDED-BY-PARENT' if code is not None else 'UNKNOWN',
|
||||||
|
OS_liveness_checked=False, sha_checked_against_latest_entry=expected_sha is not None)
|
||||||
|
|
||||||
|
|
||||||
|
def _progress(exp, control, expected_registration_sha):
|
||||||
|
regpath = exp / 'protocol/preregistered.json'
|
||||||
|
if not regpath.exists():
|
||||||
|
return dict(status='PREPARING-OR-NOT-REGISTERED', registered=False,
|
||||||
|
reason='No complete registration file is available; preparation activity is not inferred')
|
||||||
|
reg, raw = read_json(regpath)
|
||||||
|
regsha = hashlib.sha256(raw).hexdigest()
|
||||||
|
require(expected_registration_sha is None or regsha == expected_registration_sha, 'External registration SHA differs')
|
||||||
|
require(reg.get('schema') == 'normal-conditioning-registration-v1' and
|
||||||
|
reg.get('scientific_scope') == 'normal-acquisition-only' and reg.get('subsequent_assays_authorized') is False,
|
||||||
|
'Unexpected registration schema/scope')
|
||||||
|
inventory = reg['normal_inventory']
|
||||||
|
for field in ('train_calls', 'probe_calls', 'brain_calls', 'original_E_calls', 'training_cpu_validation_calls'):
|
||||||
|
require(type(inventory[field]) is int and inventory[field] >= 0, 'Missing registered inventory: ' + field)
|
||||||
|
require(inventory['brain_calls'] == inventory['train_calls'] + inventory['probe_calls'] and
|
||||||
|
inventory['original_E_calls'] == inventory['training_cpu_validation_calls'] == inventory['train_calls'],
|
||||||
|
'Registered normal inventory is inconsistent')
|
||||||
|
expected_processes = 1 + inventory['brain_calls'] + inventory['original_E_calls'] + inventory['training_cpu_validation_calls']
|
||||||
|
ledger = exp / '.append-ledger'
|
||||||
|
files = directory_files(ledger / 'entries')
|
||||||
|
starts = directory_files(ledger / 'started')
|
||||||
|
results = directory_files(exp / 'results')
|
||||||
|
require(all(RESULT_NAME.fullmatch(p.name) for p in results), 'Unrecognized/partial result file')
|
||||||
|
if ledger.exists():
|
||||||
|
header, _ = read_json(ledger / 'header.json')
|
||||||
|
require(header.get('schema') == 'normal-append-ledger-v1' and header.get('root') == str(exp) and
|
||||||
|
header.get('registration_sha256') == regsha and
|
||||||
|
header.get('initial_inventory', {}).get('protocol/preregistered.json', {}).get('sha256') == regsha,
|
||||||
|
'Ledger header registration binding differs')
|
||||||
|
seen, committed_results, counts = set(), set(), Counter()
|
||||||
|
latest = None
|
||||||
|
for number, path in enumerate(files, 1):
|
||||||
|
require(SEQUENCE_NAME.fullmatch(path.name) and path.name == f'{number:016d}.json', 'Noncontiguous/partial ledger sequence')
|
||||||
|
row, _ = read_json(path)
|
||||||
|
entry_id = row.get('entry_id')
|
||||||
|
require(row.get('schema') == 'normal-append-ledger-v1-entry' and type(row.get('sequence')) is int and
|
||||||
|
row['sequence'] == number and row.get('registration_sha256') == regsha and
|
||||||
|
type(entry_id) is str and ENTRY_ID.fullmatch(entry_id) and entry_id not in seen and
|
||||||
|
row.get('scientific_gate_claimed') is False, 'Ledger entry identity/schema differs')
|
||||||
|
metadata = row['metadata']
|
||||||
|
require(type(metadata.get('observed_returncode')) is int and metadata['observed_returncode'] == 0 and
|
||||||
|
metadata.get('job_file') == str(exp / 'jobs' / (entry_id + '.json')) and
|
||||||
|
metadata.get('process_receipt_file') == str(exp / 'receipts' / entry_id / 'process.json'), 'Ledger process metadata differs')
|
||||||
|
outgoing = row['new_outputs']
|
||||||
|
require(type(outgoing) is list and len({x['path'] for x in outgoing}) == len(outgoing), 'Invalid ledger output rows')
|
||||||
|
process_rows = [x for x in outgoing if x['path'] == metadata['process_receipt_file']]
|
||||||
|
require(len(process_rows) == 1 and SHA.fullmatch(process_rows[0]['sha256']), 'Committed process receipt is not ledger owned')
|
||||||
|
result_rows = [x for x in outgoing if Path(x['path']).parent == exp / 'results']
|
||||||
|
require(len(result_rows) <= 1, 'Multiple task results in one process entry')
|
||||||
|
if entry_id == 'input-generation':
|
||||||
|
require(number == 1 and not result_rows, 'Sampler must be the first committed process')
|
||||||
|
counts['sampler'] += 1
|
||||||
|
else:
|
||||||
|
require('input-generation' in seen, 'Task commit precedes sampler commit')
|
||||||
|
key, phase = entry_id.rsplit('--', 1)
|
||||||
|
counts[phase] += 1
|
||||||
|
if phase in ('update', 'verify'):
|
||||||
|
require(key + '--brain' in seen and (phase != 'verify' or key + '--update' in seen), 'Training commit lacks earlier brain/update')
|
||||||
|
if result_rows:
|
||||||
|
expected = str(exp / 'results' / (key + '.json'))
|
||||||
|
require(result_rows[0]['path'] == expected and expected not in committed_results and phase in ('brain', 'verify'),
|
||||||
|
'Unexpected/repeated committed task result')
|
||||||
|
committed_results.add(expected)
|
||||||
|
counts['probe_complete' if phase == 'brain' else 'train_complete'] += 1
|
||||||
|
else:
|
||||||
|
require(phase != 'verify', 'Committed training verification lacks its task result')
|
||||||
|
seen.add(entry_id)
|
||||||
|
latest = dict(path=path, row=row, process=process_rows[0], result=result_rows[0] if result_rows else None)
|
||||||
|
require(len(files) <= expected_processes and counts['train_complete'] <= inventory['train_calls'] and
|
||||||
|
counts['probe_complete'] <= inventory['probe_calls'], 'Recorded progress exceeds registered inventory')
|
||||||
|
start_ids = {p.stem for p in starts}
|
||||||
|
require(len(start_ids) == len(starts) and all(p.suffix == '.json' and ENTRY_ID.fullmatch(p.stem) for p in starts) and
|
||||||
|
seen <= start_ids, 'Missing/invalid ledger start inventory')
|
||||||
|
pending = sorted(start_ids - seen)
|
||||||
|
require(len(pending) <= 1, 'More than one uncommitted child in serial ledger')
|
||||||
|
result_paths = {str(p) for p in results}
|
||||||
|
require(committed_results <= result_paths, 'Ledger-owned result file is missing')
|
||||||
|
latest_report = None
|
||||||
|
if latest is not None:
|
||||||
|
row = latest['row']; entry_id = row['entry_id']
|
||||||
|
start, _ = read_json(ledger / 'started' / (entry_id + '.json'))
|
||||||
|
require(start['entry_id'] == entry_id and start['registration_sha256'] == regsha and start['next_sequence'] == len(files), 'Latest start binding differs')
|
||||||
|
receipt = process_record(exp, entry_id, regsha, expected_sha=latest['process']['sha256'], job_sha=start['job']['sha256'])
|
||||||
|
require(receipt['present'] and receipt['complete'] and receipt['exit_code'] == 0 and not receipt['timed_out'], 'Latest committed process is not recorded successful')
|
||||||
|
latest_report = dict(sequence=len(files), entry_id=entry_id, process=receipt)
|
||||||
|
if latest['result'] is not None:
|
||||||
|
record, raw_result = read_json(latest['result']['path'])
|
||||||
|
task = record['task']; kind = 'train' if entry_id.endswith('--verify') else 'probe'
|
||||||
|
key = 'task-' + hashlib.sha256(task['task_id'].encode()).hexdigest()
|
||||||
|
require(hashlib.sha256(raw_result).hexdigest() == latest['result']['sha256'] and
|
||||||
|
record.get('schema') == 'normal-task-result-v1' and record.get('registration_sha256') == regsha and
|
||||||
|
record.get('complete') is True and task['kind'] == kind and entry_id.startswith(key + '--'), 'Latest task result binding differs')
|
||||||
|
latest_report['task'] = {k: task[k] for k in ('task_id', 'kind', 'block_id', 'arm')}
|
||||||
|
pending_report = None
|
||||||
|
if pending:
|
||||||
|
entry_id = pending[0]
|
||||||
|
start, _ = read_json(ledger / 'started' / (entry_id + '.json'))
|
||||||
|
require(start.get('schema') == 'normal-append-ledger-v1-start' and start.get('entry_id') == entry_id and
|
||||||
|
start.get('registration_sha256') == regsha and start.get('next_sequence') == len(files) + 1 and
|
||||||
|
start.get('terminal_state') == 'unknown' and start.get('retry_authorized') is False, 'Pending start identity differs')
|
||||||
|
pending_report = dict(entry_id=entry_id, process=process_record(exp, entry_id, regsha, job_sha=start['job']['sha256']))
|
||||||
|
control_records = {name: named_record(control / filename, regsha) for name, filename in
|
||||||
|
(('launch_attempt', 'launch-attempt.json'), ('launch', 'launch.json'), ('supervisor_start', 'supervisor-start.json'), ('exit', 'exit.json'))}
|
||||||
|
launch = control_records['launch']
|
||||||
|
if launch is not None:
|
||||||
|
require(launch.get('launch_observed') is True and launch.get('final_exit_observed') is False and
|
||||||
|
launch.get('not_evidence_of_completion') is True and type(launch.get('supervisor_pid')) is int, 'Incomplete launch observation')
|
||||||
|
attempt = control_records['launch_attempt']
|
||||||
|
if attempt is not None:
|
||||||
|
require(attempt.get('complete') is False and attempt.get('terminal_state') == 'UNKNOWN' and
|
||||||
|
attempt.get('automatic_retry') is False, 'Incomplete launch attempt record')
|
||||||
|
supervisor = control_records['supervisor_start']
|
||||||
|
if supervisor is not None:
|
||||||
|
require(supervisor.get('complete') is False and type(supervisor.get('pid')) is int, 'Incomplete supervisor start record')
|
||||||
|
stopped = named_record(exp / 'audit/stopped.json', regsha)
|
||||||
|
completion = named_record(exp / 'audit/execution-complete.json', regsha)
|
||||||
|
exit_record = control_records['exit']
|
||||||
|
status = 'REGISTERED-NOT-STARTED'
|
||||||
|
if attempt is not None:
|
||||||
|
status = 'LAUNCH-ATTEMPT-RECORDED-EXIT-UNKNOWN'
|
||||||
|
if control_records['launch'] is not None or control_records['supervisor_start'] is not None:
|
||||||
|
status = 'LAUNCH-RECORDED-EXIT-UNKNOWN'
|
||||||
|
if files:
|
||||||
|
status = 'BETWEEN-COMMITS-OR-PARENT-STATE-UNKNOWN'
|
||||||
|
if pending_report:
|
||||||
|
receipt = pending_report['process']
|
||||||
|
status = 'SAMPLER-STARTED-EXIT-UNKNOWN' if pending[0] == 'input-generation' else 'CHILD-STARTED-UNCOMMITTED-EXIT-UNKNOWN'
|
||||||
|
if receipt['present']:
|
||||||
|
if receipt['timed_out'] or (receipt['exit_code'] is not None and receipt['exit_code'] != 0): status = 'CHILD-FAILED-UNCOMMITTED'
|
||||||
|
elif receipt['complete']: status = 'CHILD-EXITED-ZERO-NOT-COMMITTED'
|
||||||
|
elif receipt['exit_code'] == 0: status = 'CHILD-EXITED-ZERO-NOT-VALIDATED'
|
||||||
|
if completion is not None:
|
||||||
|
require(completion.get('complete') is True and completion.get('inventory') == inventory and
|
||||||
|
len(files) == expected_processes and not pending, 'Execution completion contradicts ledger inventory')
|
||||||
|
status = 'EXECUTION-COMPLETION-RECORDED-PARENT-EXIT-UNKNOWN'
|
||||||
|
if exit_record is not None:
|
||||||
|
require(type(exit_record.get('final_exit_observed')) is bool and type(exit_record.get('complete')) is bool and
|
||||||
|
(exit_record.get('exit_code') is None or type(exit_record.get('exit_code')) is int), 'Incomplete supervisor exit record')
|
||||||
|
code = exit_record['exit_code']
|
||||||
|
require((code is None) == (exit_record['final_exit_observed'] is False) and exit_record['complete'] is (code == 0), 'Contradictory supervisor exit record')
|
||||||
|
status = 'SUPERVISOR-ERROR-EXIT-UNKNOWN' if code is None else ('PARENT-EXITED-ZERO' if code == 0 else 'PARENT-EXITED-NONZERO')
|
||||||
|
if stopped is not None:
|
||||||
|
require(stopped.get('complete') is False, 'Malformed stopped record')
|
||||||
|
status = 'STOPPED-RECORD-PRESENT'
|
||||||
|
return dict(status=status, registered=True, registration_sha256=regsha, registered_inventory=inventory,
|
||||||
|
expected_process_entries=expected_processes, committed_process_entries=len(files),
|
||||||
|
sampler_committed=counts['sampler'] == 1, brain_entries=counts['brain'], update_entries=counts['update'],
|
||||||
|
training_epochs_completed=counts['train_complete'], probes_completed=counts['probe_complete'],
|
||||||
|
result_files_present=len(results), task_results_ledger_owned=len(committed_results),
|
||||||
|
result_files_without_commit=sorted(result_paths - committed_results), latest_entry=latest_report,
|
||||||
|
uncommitted_child=pending_report, control_records=control_records, stopped_record=stopped,
|
||||||
|
execution_completion_record_present=completion is not None,
|
||||||
|
registered_workflow_inventory_recorded_complete=(len(files) == expected_processes and
|
||||||
|
counts['train_complete'] == inventory['train_calls'] and counts['probe_complete'] == inventory['probe_calls'] and not pending),
|
||||||
|
ledger_writer_lock_present=(ledger / 'writer.lock').exists(),
|
||||||
|
entries_appeared_during_read=len(directory_files(ledger / 'entries')) != len(files))
|
||||||
|
|
||||||
|
|
||||||
|
def read_progress(exp, control, registration_sha256=None):
|
||||||
|
"""Unknown/inconsistent metadata never become completed work or learning."""
|
||||||
|
base = dict(schema='normal-fast-progress-v1', scope='READ-ONLY-METADATA-NOT-INTEGRITY',
|
||||||
|
integrity_verified=False, historical_receipts_reverified=False, OS_liveness_checked=False,
|
||||||
|
learning_pass_claimed=False, scientific_metrics_evaluated=False, candidate_generated=False,
|
||||||
|
writes_performed=False, model_executed=False, automatic_retry_authorized=False,
|
||||||
|
snapshot_scope='Non-atomic observation; concurrent publication can require reading again')
|
||||||
|
try:
|
||||||
|
exp, control = safe(exp), safe(control)
|
||||||
|
require(registration_sha256 is None or SHA.fullmatch(registration_sha256), 'Invalid external registration SHA')
|
||||||
|
return dict(base, **_progress(exp, control, registration_sha256))
|
||||||
|
except (OSError, ValueError, KeyError, TypeError, AttributeError, IndexError) as error:
|
||||||
|
return dict(base, status='UNKNOWN-METADATA-INCONSISTENT', error=type(error).__name__ + ': ' + str(error),
|
||||||
|
completion_counts_available=False)
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == '__main__':
|
||||||
|
parser = argparse.ArgumentParser(description=__doc__)
|
||||||
|
parser.add_argument('--exp', required=True)
|
||||||
|
parser.add_argument('--control', required=True)
|
||||||
|
parser.add_argument('--registration-sha256')
|
||||||
|
args = parser.parse_args()
|
||||||
|
print(json.dumps(read_progress(args.exp, args.control, args.registration_sha256), ensure_ascii=False, sort_keys=True, indent=2))
|
||||||
@@ -0,0 +1,168 @@
|
|||||||
|
"""Temporary metadata fixtures only; never starts a model or reads its arrays."""
|
||||||
|
import hashlib
|
||||||
|
import importlib.util
|
||||||
|
import json
|
||||||
|
from pathlib import Path
|
||||||
|
import tempfile
|
||||||
|
import unittest
|
||||||
|
|
||||||
|
spec = importlib.util.spec_from_file_location('_progress_subject', Path(__file__).with_name('progress_normal.py'))
|
||||||
|
P = importlib.util.module_from_spec(spec)
|
||||||
|
spec.loader.exec_module(P)
|
||||||
|
|
||||||
|
|
||||||
|
def write(path, value):
|
||||||
|
path.parent.mkdir(parents=True, exist_ok=True)
|
||||||
|
raw = (json.dumps(value, sort_keys=True) + '\n').encode()
|
||||||
|
path.write_bytes(raw)
|
||||||
|
return hashlib.sha256(raw).hexdigest()
|
||||||
|
|
||||||
|
|
||||||
|
class Fixture:
|
||||||
|
def __init__(self, root, register=True):
|
||||||
|
self.exp, self.control = root / 'EXP', root / 'CONTROL'
|
||||||
|
self.exp.mkdir(); self.control.mkdir()
|
||||||
|
self.inventory = dict(train_calls=1, probe_calls=1, brain_calls=2, original_E_calls=1, training_cpu_validation_calls=1)
|
||||||
|
self.count = 0
|
||||||
|
if register:
|
||||||
|
self.sha = write(self.exp / 'protocol/preregistered.json', dict(schema='normal-conditioning-registration-v1',
|
||||||
|
scientific_scope='normal-acquisition-only', subsequent_assays_authorized=False, normal_inventory=self.inventory))
|
||||||
|
write(self.exp / '.append-ledger/header.json', dict(schema='normal-append-ledger-v1', root=str(self.exp),
|
||||||
|
registration_sha256=self.sha, initial_inventory={'protocol/preregistered.json': {'sha256': self.sha}}))
|
||||||
|
|
||||||
|
def start(self, entry_id):
|
||||||
|
job = self.exp / 'jobs' / (entry_id + '.json')
|
||||||
|
jobsha = write(job, dict(kind='sample' if entry_id == 'input-generation' else 'fixture'))
|
||||||
|
value = dict(schema='normal-append-ledger-v1-start', entry_id=entry_id, registration_sha256=self.sha,
|
||||||
|
next_sequence=self.count + 1, prev_entry_sha256=None, terminal_state='unknown', retry_authorized=False,
|
||||||
|
job=dict(path=str(job), sha256=jobsha))
|
||||||
|
write(self.exp / '.append-ledger/started' / (entry_id + '.json'), value)
|
||||||
|
return jobsha
|
||||||
|
|
||||||
|
def process(self, entry_id, jobsha, code=0, complete=True, timed_out=False):
|
||||||
|
path = self.exp / 'receipts' / entry_id / 'process.json'
|
||||||
|
digest = write(path, dict(entry_id=entry_id, registration_sha256=self.sha, job_sha256=jobsha,
|
||||||
|
complete=complete, exit_code=code, timed_out=timed_out, exit_observed_by_parent=code is not None or timed_out))
|
||||||
|
return dict(path=str(path), relative_path=path.relative_to(self.exp).as_posix(), sha256=digest, bytes=path.stat().st_size)
|
||||||
|
|
||||||
|
def result(self, task):
|
||||||
|
key = 'task-' + hashlib.sha256(task['task_id'].encode()).hexdigest()
|
||||||
|
path = self.exp / 'results' / (key + '.json')
|
||||||
|
digest = write(path, dict(schema='normal-task-result-v1', registration_sha256=self.sha, complete=True, task=task))
|
||||||
|
return dict(path=str(path), relative_path=path.relative_to(self.exp).as_posix(), sha256=digest, bytes=path.stat().st_size)
|
||||||
|
|
||||||
|
def commit(self, entry_id, task=None):
|
||||||
|
jobsha = self.start(entry_id)
|
||||||
|
outputs = [self.process(entry_id, jobsha)]
|
||||||
|
if task is not None:
|
||||||
|
outputs.append(self.result(task))
|
||||||
|
self.count += 1
|
||||||
|
return write(self.exp / f'.append-ledger/entries/{self.count:016d}.json', dict(
|
||||||
|
schema='normal-append-ledger-v1-entry', entry_id=entry_id, sequence=self.count, registration_sha256=self.sha,
|
||||||
|
scientific_gate_claimed=False, new_outputs=outputs, metadata=dict(observed_returncode=0,
|
||||||
|
job_file=str(self.exp / 'jobs' / (entry_id + '.json')),
|
||||||
|
process_receipt_file=str(self.exp / 'receipts' / entry_id / 'process.json'))))
|
||||||
|
|
||||||
|
def task(self, kind):
|
||||||
|
task = dict(task_id='block/normal/paired/' + ('000' if kind == 'train' else 'checkpoint-00/probe-A'),
|
||||||
|
kind=kind, block_id='block', arm='paired')
|
||||||
|
key = 'task-' + hashlib.sha256(task['task_id'].encode()).hexdigest()
|
||||||
|
return task, key
|
||||||
|
|
||||||
|
def read(self):
|
||||||
|
return P.read_progress(self.exp, self.control)
|
||||||
|
|
||||||
|
|
||||||
|
class ProgressTests(unittest.TestCase):
|
||||||
|
def test_no_registration_not_learning_or_inferred_preparation_liveness(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp), register=False)
|
||||||
|
value = f.read()
|
||||||
|
self.assertEqual(value['status'], 'PREPARING-OR-NOT-REGISTERED')
|
||||||
|
self.assertFalse(value['registered'] or value['OS_liveness_checked'] or value['learning_pass_claimed'])
|
||||||
|
|
||||||
|
def test_registered_and_launch_pid_are_not_completion(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp))
|
||||||
|
self.assertEqual(f.read()['status'], 'REGISTERED-NOT-STARTED')
|
||||||
|
write(f.control / 'launch.json', dict(registration_sha256=f.sha, supervisor_pid=123,
|
||||||
|
launch_observed=True, final_exit_observed=False, not_evidence_of_completion=True))
|
||||||
|
value = f.read()
|
||||||
|
self.assertEqual(value['status'], 'LAUNCH-RECORDED-EXIT-UNKNOWN')
|
||||||
|
self.assertEqual(value['expected_process_entries'], 5)
|
||||||
|
self.assertEqual(value['committed_process_entries'], 0)
|
||||||
|
|
||||||
|
def test_sampler_started_vs_committed_and_latest_actual_exit(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); f.start('input-generation')
|
||||||
|
self.assertEqual(f.read()['status'], 'SAMPLER-STARTED-EXIT-UNKNOWN')
|
||||||
|
f.commit('input-generation')
|
||||||
|
value = f.read()
|
||||||
|
self.assertTrue(value['sampler_committed'])
|
||||||
|
self.assertEqual(value['latest_entry']['entry_id'], 'input-generation')
|
||||||
|
self.assertTrue(value['latest_entry']['process']['sha_checked_against_latest_entry'])
|
||||||
|
self.assertFalse(value['integrity_verified'])
|
||||||
|
|
||||||
|
def test_written_result_never_substitutes_for_verify_commit(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); f.commit('input-generation')
|
||||||
|
task, key = f.task('train')
|
||||||
|
f.commit(key + '--brain'); f.commit(key + '--update')
|
||||||
|
jobsha = f.start(key + '--verify'); f.result(task)
|
||||||
|
before = f.read()
|
||||||
|
self.assertEqual(before['training_epochs_completed'], 0)
|
||||||
|
self.assertEqual(before['result_files_present'], 1)
|
||||||
|
self.assertEqual(len(before['result_files_without_commit']), 1)
|
||||||
|
f.process(key + '--verify', jobsha)
|
||||||
|
self.assertEqual(f.read()['status'], 'CHILD-EXITED-ZERO-NOT-COMMITTED')
|
||||||
|
f.commit(key + '--verify', task)
|
||||||
|
after = f.read()
|
||||||
|
self.assertEqual(after['training_epochs_completed'], 1)
|
||||||
|
self.assertEqual(after['result_files_without_commit'], [])
|
||||||
|
|
||||||
|
def test_probe_and_train_commits_separate_and_no_neural_files_read(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); f.commit('input-generation')
|
||||||
|
train, key = f.task('train')
|
||||||
|
f.commit(key + '--brain'); f.commit(key + '--update'); f.commit(key + '--verify', train)
|
||||||
|
probe, key = f.task('probe'); f.commit(key + '--brain', probe)
|
||||||
|
state = f.exp / 'runs' / key / 'state.npz'; state.parent.mkdir(parents=True); state.write_bytes(b'not-an-array')
|
||||||
|
value = f.read()
|
||||||
|
self.assertEqual((value['training_epochs_completed'], value['probes_completed']), (1, 1))
|
||||||
|
self.assertEqual(value['task_results_ledger_owned'], 2)
|
||||||
|
self.assertEqual(value['latest_entry']['task']['kind'], 'probe')
|
||||||
|
self.assertFalse(value['scientific_metrics_evaluated'] or value['model_executed'])
|
||||||
|
|
||||||
|
def test_failed_child_and_real_supervisor_exit_and_stop_record(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); jobsha = f.start('input-generation')
|
||||||
|
f.process('input-generation', jobsha, code=7, complete=False)
|
||||||
|
self.assertEqual(f.read()['status'], 'CHILD-FAILED-UNCOMMITTED')
|
||||||
|
write(f.control / 'exit.json', dict(registration_sha256=f.sha, final_exit_observed=True, exit_code=1, complete=False))
|
||||||
|
self.assertEqual(f.read()['status'], 'PARENT-EXITED-NONZERO')
|
||||||
|
write(f.exp / 'audit/stopped.json', dict(registration_sha256=f.sha, complete=False, error='fixture'))
|
||||||
|
self.assertEqual(f.read()['status'], 'STOPPED-RECORD-PRESENT')
|
||||||
|
|
||||||
|
def test_missing_fields_bad_sequence_or_registration_are_unknown(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); f.commit('input-generation')
|
||||||
|
value = P.read_progress(f.exp, f.control, '0' * 64)
|
||||||
|
self.assertEqual(value['status'], 'UNKNOWN-METADATA-INCONSISTENT')
|
||||||
|
path = f.exp / '.append-ledger/entries/0000000000000001.json'
|
||||||
|
row, _ = P.read_json(path); del row['metadata']['observed_returncode']; write(path, row)
|
||||||
|
self.assertEqual(f.read()['status'], 'UNKNOWN-METADATA-INCONSISTENT')
|
||||||
|
path.rename(path.with_name('0000000000000002.json'))
|
||||||
|
self.assertEqual(f.read()['status'], 'UNKNOWN-METADATA-INCONSISTENT')
|
||||||
|
|
||||||
|
def test_latest_receipt_mutation_is_unknown_not_metadata_success(self):
|
||||||
|
with tempfile.TemporaryDirectory() as temp:
|
||||||
|
f = Fixture(Path(temp)); f.commit('input-generation')
|
||||||
|
path = f.exp / 'receipts/input-generation/process.json'
|
||||||
|
row, _ = P.read_json(path); row['exit_code'] = 9; row['complete'] = False; write(path, row)
|
||||||
|
value = f.read()
|
||||||
|
self.assertEqual(value['status'], 'UNKNOWN-METADATA-INCONSISTENT')
|
||||||
|
self.assertFalse(value['completion_counts_available'])
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == '__main__':
|
||||||
|
unittest.main()
|
||||||
Reference in New Issue
Block a user