import csv import re import subprocess import time from pathlib import Path CONFIGS = [(1, 1), (2, 2), (4, 4)] COUNT_PER_SIZE = 20 RESULTS_DIR = Path("benchmark_results") RESULTS_DIR.mkdir(exist_ok=True) RSS_CODE = """ import os total_kb = 0 current_pid = os.getpid() for name in os.listdir("/proc"): if not name.isdigit(): continue if int(name) == current_pid: continue try: with open(f"/proc/{name}/status") as file: for line in file: if line.startswith("VmRSS:"): total_kb += int(line.split()[1]) break except (OSError, ValueError): pass print(total_kb / 1024) """ def command(args, check=True): return subprocess.run( args, text=True, capture_output=True, check=check, ) def remove_test_worker(): command( ["docker", "rm", "-f", "benchmark-worker"], check=False, ) def start_worker(concurrency, prefetch): remove_test_worker() result = command([ "docker", "compose", "run", "-d", "--name", "benchmark-worker", "--no-deps", "-e", f"WORKER_CONCURRENCY={concurrency}", "-e", f"WORKER_PREFETCH={prefetch}", "worker", ]) expected = ( f"[Worker] CONCURRENCY={concurrency}, " f"PREFETCH={prefetch}" ) for _ in range(30): logs = command( ["docker", "logs", "benchmark-worker"], check=False, ) if expected in logs.stdout + logs.stderr: print(expected) return time.sleep(1) raise RuntimeError( f"Worker {concurrency}/{prefetch} не запустился.\n" + command( ["docker", "logs", "benchmark-worker"], check=False, ).stderr ) def measure_rss(): result = command( [ "docker", "exec", "benchmark-worker", "python", "-c", RSS_CODE, ], check=False, ) if result.returncode != 0: return None try: return float(result.stdout.strip()) except ValueError: return None def extract_number(text, label): match = re.search( rf"^{re.escape(label)}:\s*([0-9.]+)", text, re.MULTILINE, ) return float(match.group(1)) if match else None def run_benchmark(concurrency, prefetch): config = f"{concurrency}_{prefetch}" print(f"\n{'=' * 55}") print(f"ЭКСПЕРИМЕНТ: {concurrency}/{prefetch}") print(f"{'=' * 55}") start_worker(concurrency, prefetch) log_path = RESULTS_DIR / f"load_{config}.log" rss_samples = [] load_command = [ "docker", "compose", "exec", "-T", "-w", "/tmp", "-e", "PYTHONPATH=/app", "app1", "python", "-m", "app.load_test", "--count", str(COUNT_PER_SIZE), "--timeout", "300", ] print("Запускаем 60 фотографий...") start_time = time.monotonic() with log_path.open("w") as log_file: process = subprocess.Popen( load_command, stdout=log_file, stderr=subprocess.STDOUT, ) while process.poll() is None: elapsed = time.monotonic() - start_time rss = measure_rss() if rss is not None: rss_samples.append((elapsed, rss)) time.sleep(0.5) return_code = process.wait() output = log_path.read_text() if return_code != 0: print(output) raise RuntimeError( f"Нагрузочный тест {concurrency}/{prefetch} завершился с ошибкой" ) print(output) if not rss_samples: raise RuntimeError("Не удалось измерить RSS") rss_values = [value for _, value in rss_samples] peak_rss = max(rss_values) mean_rss = sum(rss_values) / len(rss_values) samples_path = RESULTS_DIR / f"rss_{config}.csv" with samples_path.open("w", newline="") as file: writer = csv.writer(file) writer.writerow(["elapsed_s", "rss_mb"]) writer.writerows(rss_samples) csv_match = re.search( r"CSV:\s*(benchmark_[a-zA-Z0-9]+\.csv)", output, ) if csv_match: filename = csv_match.group(1) copy_result = command( [ "docker", "compose", "cp", f"app1:/tmp/{filename}", str(RESULTS_DIR / filename), ], check=False, ) if copy_result.returncode != 0: print("Не удалось скопировать CSV:", copy_result.stderr) result = { "config": f"{concurrency}/{prefetch}", "total": int(extract_number(output, "Total jobs") or 0), "ready": int(extract_number(output, "Ready") or 0), "failed": int(extract_number(output, "Failed") or 0), "duration_s": extract_number(output, "Duration"), "throughput": extract_number(output, "Throughput"), "peak_processing": extract_number( output, "Peak observed processing" ), "peak_rss_mb": round(peak_rss, 2), "mean_rss_mb": round(mean_rss, 2), } print(f"\n[RSS] Пиковое значение: {peak_rss:.2f} MB") print(f"[RSS] Среднее значение: {mean_rss:.2f} MB") print(f"[RSS] Количество замеров: {len(rss_samples)}") return result def save_summary(results): output = RESULTS_DIR / "summary.csv" with output.open("w", newline="") as file: writer = csv.DictWriter( file, fieldnames=[ "config", "total", "ready", "failed", "duration_s", "throughput", "peak_processing", "peak_rss_mb", "mean_rss_mb", ], ) writer.writeheader() writer.writerows(results) print("\n" + "=" * 70) print("ИТОГОВАЯ ТАБЛИЦА") print("=" * 70) for item in results: print( f"{item['config']:5} | " f"Ready: {item['ready']:3} | " f"Time: {item['duration_s']} s | " f"Throughput: {item['throughput']} img/s | " f"Peak RSS: {item['peak_rss_mb']} MB" ) print(f"\nРезультаты сохранены в {RESULTS_DIR}/") def main(): results = [] try: print("Останавливаем обычный Worker...") command(["docker", "compose", "stop", "worker"]) for concurrency, prefetch in CONFIGS: result = run_benchmark(concurrency, prefetch) results.append(result) finally: print("\nВосстанавливаем обычный Worker...") remove_test_worker() restore = command( ["docker", "compose", "up", "-d", "worker"], check=False, ) if restore.returncode != 0: print("Ошибка восстановления Worker:") print(restore.stderr) if results: save_summary(results) if __name__ == "__main__": main()