m/market labResearch, with a record.
Source on GitHub ↗

benchmark_prepared_import.py

1"""Compare the former Python export loop with native batches, then import a full day.23Outputs small JSON evidence; all large generated files live in an owned temporary4workspace and are cleaned on exit. No network, model calls or source modification.5"""67import argparse8import csv9import json10from pathlib import Path11import random12import resource13import sys14import tempfile15import time1617ROOT = Path(__file__).resolve().parents[1]18sys.path.insert(0, str(ROOT))19from sim_platform.market.session import MarketSession20from sim_platform.market.prepared.builder import build21from sim_platform.market.prepared.codec import encode_event, encode_state22from sim_platform.market.prepared.reader import PreparedMarket23from benchmark_prepared_runtime import distribution as milliseconds242526def distribution(samples):27    values = milliseconds(samples)28    return dict(29        count=values['count'],30        **{31            key.replace('_ms', '_us'): value * 1000 if value is not None else None32            for key, value in values.items()33            if key != 'count'34        },35    )363738def prefix(source, destination, count):39    """Copy complete BinaryFILE frames and add an explicit fixture terminator."""40    with source.open('rb') as incoming, destination.open('wb') as outgoing:41        for _ in range(count):42            header = incoming.read(2)43            if len(header) != 2:44                raise ValueError('source ends before the requested prefix')45            length = int.from_bytes(header, 'big')46            if not length:47                break48            payload = incoming.read(length)49            if len(payload) != length:50                raise ValueError('source has a truncated frame')51            outgoing.write(header + payload)52        outgoing.write(b'\0\0')535455def reference_export(source, target):56    """Former per-frame JSON/Python path, retained only as a benchmark reference."""57    market = MarketSession('AAPL')58    market.load(source)59    market.play()60    started = time.perf_counter()61    pump_ns = encode_write_ns = frames = versions = 062    previous = encode_state(market.snapshot, 5)63    try:64        with (65            (target / 'events.bin').open('wb') as events,66            (target / 'states.bin').open('wb') as states,67        ):68            states.write(previous)69            while True:70                tick = time.perf_counter_ns()71                event = market.pump(1)[0]72                pump_ns += time.perf_counter_ns() - tick73                if market.status == 'completed':74                    break75                if market.status == 'failed':76                    raise RuntimeError(event.message)77                tick = time.perf_counter_ns()78                raw = encode_state(event.snapshot, 5)79                if raw != previous:80                    versions += 181                    states.write(raw)82                    previous = raw83                events.write(encode_event(event, versions))84                frames += 185                encode_write_ns += time.perf_counter_ns() - tick86        return dict(87            seconds=time.perf_counter() - started,88            source_frames=frames,89            native_json_python_pump_seconds=pump_ns / 1e9,90            python_encode_write_seconds=encode_write_ns / 1e9,91            artifact_data_bytes=sum(p.stat().st_size for p in target.iterdir()),92            instrumented=True,93        )94    finally:95        market.close()969798def queries(reader, count=5000):99    m = reader.manifest100    rng = random.Random(731)101    points = [rng.randint(m['first_ns'], m['last_ns']) for _ in range(count)]102    runs = []103    for _ in range(2):104        samples = []105        for point in points:106            started = time.perf_counter_ns()107            reader.snapshot_at(point)108            samples.append(time.perf_counter_ns() - started)109        runs.append(distribution(samples))110    return dict(111        first=runs[0],112        repeated=runs[1],113        os_cold=False,114        native_calls_per_query=0,115        replayed_frames_per_query=0,116    )117118119def native_import(source, target, *, implicit=False, profile=False):120    stages = {}121    last_progress = [time.monotonic()]122123    def progress(frames):124        if time.monotonic() - last_progress[0] >= 10:125            print('import progress: ' + str(frames) + ' source frames', flush=True)126            last_progress[0] = time.monotonic()127128    started = time.perf_counter()129    build(130        source,131        target,132        allow_implicit_eof=implicit,133        timings=stages,134        profile_native=profile,135        progress=progress,136    )137    seconds = time.perf_counter() - started138    opens = []139    for _ in range(5):140        started = time.perf_counter_ns()141        with PreparedMarket(target) as reader:142            opens.append(time.perf_counter_ns() - started)143    with PreparedMarket(target) as reader:144        m = reader.manifest145        return dict(146            seconds=seconds,147            source_bytes=source.stat().st_size,148            source_frames=m['source_frames'],149            retained_events=m['frames'],150            state_versions=m['states'],151            artifact_bytes=sum(p.stat().st_size for p in target.iterdir()),152            open=distribution(opens),153            queries=queries(reader),154            instrumented=profile,155            native_stage_timings=stages,156            lifetime_peak_rss_bytes=resource.getrusage(resource.RUSAGE_SELF).ru_maxrss,157        )158159160def compare_csv(target, csv_path):161    """All book events against existing pre-change CSV, including depth and source ID."""162163    def price(value):164        return f'{value // 10000}.{value % 10000:04d}'165166    def levels(values):167        return '|'.join(f'{price(p)}:{q}' for p, q in values)168169    count = 0170    with PreparedMarket(target) as reader, csv_path.open(newline='') as stream:171        rows = csv.DictReader(stream)172        for index in range(reader.manifest['frames']):173            event = reader.event(index)174            if event.source_type not in 'AFECXDU':175                continue176            row = next(rows)177            s = event.snapshot178            actual = [179                str(reader.source_sequence(index + 1)),180                str(event.time_ns),181                event.source_type,182                price(s.bids[0][0]) if s.bids else '',183                str(s.bids[0][1]) if s.bids else '',184                price(s.asks[0][0]) if s.asks else '',185                str(s.asks[0][1]) if s.asks else '',186                str(s.bid_levels),187                str(s.ask_levels),188                levels(s.bids),189                levels(s.asks),190            ]191            if actual != list(row.values()):192                raise AssertionError('prepared differs from existing CSV at row ' + str(count + 1))193            count += 1194        if next(rows, None) is not None:195            raise AssertionError('prepared omitted CSV events')196    return dict(197        rows=count,198        all_fields_equal=True,199        note='pre-existing CSV from original native core; not a second independent book',200    )201202203def main():204    parser = argparse.ArgumentParser(description=__doc__)205    parser.add_argument('--source', type=Path, required=True)206    parser.add_argument('--output', type=Path, required=True)207    parser.add_argument('--csv', type=Path)208    parser.add_argument('--prefix-frames', type=int, default=1_000_000)209    parser.add_argument('--full-day', action='store_true')210    args = parser.parse_args()211    report = dict(212        symbol='AAPL',213        depth=5,214        source=str(args.source.resolve()),215        cold_cache=False,216        scope='local import feasibility, not live or financial throughput',217    )218219    def save():220        args.output.write_text(json.dumps(report, indent=2) + '\n')221222    with tempfile.TemporaryDirectory(prefix='prepared-import-') as workspace:223        root = Path(workspace)224        sample = root / 'prefix.bin'225        prefix(args.source, sample, args.prefix_frames)226        old = root / 'reference'227        old.mkdir()228        report['prefix_reference'] = reference_export(sample, old)229        save()230        print('reference prefix complete', flush=True)231        report['prefix_native'] = native_import(sample, root / 'native')232        save()233        print('native prefix complete', flush=True)234        report['prefix_profiled'] = native_import(sample, root / 'profiled', profile=True)235        save()236        if args.full_day:237            report['full_day'] = native_import(args.source, root / 'full', implicit=True)238            save()239            print('full-day import complete', flush=True)240            if args.csv:241                report['full_day']['existing_csv_comparison'] = compare_csv(root / 'full', args.csv)242                save()243                print('full-day CSV comparison complete', flush=True)244    print(json.dumps(report, indent=2), flush=True)245246247if __name__ == '__main__':248    main()