#!/usr/bin/env python3 """Reproducible collection benchmark; every restore is hash-checked.""" import argparse import csv import hashlib import json import os from pathlib import Path import platform import random import shutil import subprocess import tarfile import time def digest(path): with path.open('rb') as stream: return hashlib.file_digest(stream, 'sha256').hexdigest() def manifest(root): return {p.relative_to(root).as_posix(): digest(p) for p in sorted(root.rglob('*')) if p.is_file()} def fixtures(root, size): rng = random.Random(20261105) base = rng.randbytes(size) names = ['binary_versions', 'game_assets', 'source_versions', 'simulation', 'independent'] for kind in names: folder = root / kind folder.mkdir(parents=True, exist_ok=False) for version in range(8): if kind == 'binary_versions': data = bytearray(base) for _ in range(8): pos = rng.randrange(size - 32) data[pos:pos+32] = rng.randbytes(32) elif kind == 'game_assets': # Asset packs reorder shared high-entropy objects and add one new object. assets = [base[i:i+32768] for i in range(0, size, 32768)] rng.shuffle(assets) assets[0] = rng.randbytes(len(assets[0])) data = b''.join(assets) elif kind == 'source_versions': lines = [f'int function_{i}(int value) {{ return value + {i % 97}; }}\n' for i in range(max(1, size // 60))] lines[version] = f'/* revision {version} */\n' data = ''.join(lines).encode() elif kind == 'simulation': # Synthetic serialized checkpoints; repeated blocks with local updates. blocks = [base[i:i+4096] for i in range(0, size, 4096)] blocks[version % len(blocks)] = rng.randbytes(len(blocks[version % len(blocks)])) data = b''.join(blocks) else: data = rng.randbytes(size) item = folder / f'revision-{version:02}.dat' item.write_bytes(data) item.chmod(0o644) os.utime(item, (946684800, 946684800)) yield kind, folder def run(command, cwd, logs, tag, stdin=None, stdout=None): timing = logs / (tag + '.time') started = time.perf_counter() with (logs / (tag + '.stderr')).open('wb') as err: result = subprocess.run(['/usr/bin/time', '-f', '%e %M', '-o', str(timing), 'timeout', '--kill-after=5s', '600s', *map(str, command)], cwd=cwd, stdin=stdin, stdout=stdout or subprocess.DEVNULL, stderr=err) if result.returncode: raise RuntimeError(f'{tag} failed: see {logs / (tag + ".stderr")}') elapsed = time.perf_counter() - started _, rss = timing.read_text().split() return round(elapsed, 6), int(rss) def main(): ap = argparse.ArgumentParser(description=__doc__) ap.add_argument('--uc2', type=Path, required=True) ap.add_argument('--output', type=Path, required=True) ap.add_argument('--size-kib', type=int, default=512) ap.add_argument('--repeat', type=int, default=3) ap.add_argument('--corpus', type=Path, help='benchmark this existing flat directory instead of synthetic collections') args = ap.parse_args() uc2 = args.uc2.resolve() root = args.output.resolve() root.mkdir(parents=True, exist_ok=False) seven = next((shutil.which(x) for x in ('7zz', '7z', '7za') if shutil.which(x)), None) if not seven or not shutil.which('zstd'): raise SystemExit('7zz/7z/7za and zstd are required; missing comparisons are not silently skipped') if args.size_kib < 32 or args.repeat < 1: raise SystemExit('size-kib must be >=32 and repeat must be >=1') metadata = {'date_utc': time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime()), 'platform': platform.platform(), 'cpu': platform.processor(), 'uc2_sha256': digest(uc2), 'seed': 20261105, 'size_kib': args.size_kib, 'repeat': args.repeat, 'corpus': (str(args.corpus.resolve()) if args.corpus else 'synthetic, eight revisions per collection; not real user data'), 'rss_scope': 'per process peak RSS in KiB; tar materialization excluded equally', 'versions': {}} for tool, cmd in [('7zip', [seven]), ('zstd', ['zstd', '--version']), ('uc2', [str(uc2), '-h'])]: metadata['versions'][tool] = subprocess.run(cmd, capture_output=True, text=True).stdout.splitlines()[:5] rows = [] if args.corpus: folder = args.corpus.resolve() if not folder.is_dir() or not any(folder.iterdir()) or any(not p.is_file() for p in folder.iterdir()): raise SystemExit('--corpus must be a nonempty flat directory of regular files') collections = [(folder.name, folder)] else: collections = fixtures(root / 'inputs', args.size_kib * 1024) for kind, folder in collections: expected = manifest(folder) total = sum(p.stat().st_size for p in folder.iterdir()) tar_path = root / f'{kind}.tar' with tarfile.open(tar_path, 'w', format=tarfile.USTAR_FORMAT) as tar: for p in sorted(folder.iterdir()): info = tar.gettarinfo(str(p), arcname=p.name) info.mtime = info.uid = info.gid = 0 info.uname = info.gname = '' with p.open('rb') as f: tar.addfile(info, f) (root / f'{kind}.manifest.json').write_text(json.dumps(expected, indent=2)+'\n') for trial in range(args.repeat): for method in ['uc2-files', 'uc2-chunks', 'tar-uc2', '7z', 'tar-zstd']: work = root / f'{kind}-{method}-{trial}' work.mkdir() archive = work / ('archive.7z' if method == '7z' else 'archive') restored = work / 'restored' restored.mkdir() with tar_path.open('rb') as stream: if method in ('uc2-files', 'uc2-chunks'): cmd = [uc2, *(['--chunked'] if method == 'uc2-chunks' else []), '-q', '-w', '-L', '4', archive, *sorted(expected)] inp = None elif method == 'tar-uc2': cmd, inp = [uc2, '--ingest', archive], stream elif method == '7z': cmd, inp = [seven, 'a', '-t7z', '-mx=5', '-mmt=1', archive, *sorted(expected)], None else: cmd, inp = ['zstd', '-q', '-3', '-T1', '-o', archive], stream enc, enc_rss = run(cmd, folder, work, 'create', stdin=inp) if method in ('uc2-files', 'uc2-chunks'): dec, dec_rss = run([uc2, '-q', '-d', restored, archive], root, work, 'restore') elif method == '7z': dec, dec_rss = run([seven, 'x', '-y', '-o'+str(restored), archive], root, work, 'restore') else: restored_tar = work / 'restored.tar' with restored_tar.open('wb') as out: cmd = ([uc2, '--ingest-restore', archive] if method == 'tar-uc2' else ['zstd', '-q', '-d', '-c', archive]) dec, dec_rss = run(cmd, root, work, 'restore', stdout=out) if digest(restored_tar) != digest(tar_path): raise RuntimeError(f'{kind}/{method}: tar bytes changed') with tarfile.open(restored_tar) as tar: tar.extractall(restored, filter='data') if manifest(restored) != expected: raise RuntimeError(f'{kind}/{method}: file manifest mismatch') row = dict(collection=kind, method=method, trial=trial, input_bytes=total, archive_bytes=archive.stat().st_size, compress_seconds=enc, restore_seconds=dec, compress_peak_kib=enc_rss, restore_peak_kib=dec_rss, verified=True) rows.append(row) print(json.dumps(row), flush=True) (root/'results.json').write_text(json.dumps({'metadata':metadata, 'results':rows}, indent=2)+'\n') with (root/'results.csv').open('w') as f: writer=csv.DictWriter(f, fieldnames=list(rows[0])) writer.writeheader(); writer.writerows(rows) if __name__ == '__main__': main()