# coding=utf-8 import os import json import commands import hashlib from multiprocessing import Pool from settings import max_running_number, lrun_gid, lrun_uid, judger_workspace from language import languages from result import result from judge_exceptions import JudgeClientError from utils import parse_lrun_output from logger import logger # 下面这个函数作为代理访问实例变量，否则Python2会报错，是Python2的已知问题 # http://stackoverflow.com/questions/1816958/cant-pickle-type-instancemethod-when-using-pythons-multiprocessing-pool-ma/7309686 def _run(instance, test_case_id): return instance._judge_one(test_case_id) class JudgeClient(object): def __init__(self, language_code, exe_path, max_cpu_time, max_real_time, max_memory, test_case_dir): """ :param language_code: 语言编号 :param exe_path: 可执行文件路径 :param max_cpu_time: 最大cpu时间，单位ms :param max_real_time: 最大执行时间，单位ms :param max_memory: 最大内存，单位MB :param test_case_dir: 测试用例文件夹路径 :return:返回结果list """ self._language = languages[language_code] self._exe_path = exe_path self._max_cpu_time = max_cpu_time self._max_real_time = max_real_time self._max_memory = max_memory self._test_case_dir = test_case_dir # 进程池 self._pool = Pool(processes=max_running_number) # 测试用例配置项 self._test_case_info = self._load_test_case_info() def _load_test_case_info(self): # 读取测试用例信息 转换为dict try: f = open(self._test_case_dir + "info") return json.loads(f.read()) except IOError: raise JudgeClientError("Test case config file not found") except ValueError: raise JudgeClientError("Test case config file format error") def _generate_command(self, test_case_id): """ 设置相关运行限制 进制访问网络 如果启用tmpfs 就把代码输出写入tmpfs，否则写入硬盘 """ # todo 系统调用白名单 chroot等参数 command = "lrun" + \ " --max-cpu-time " + str(self._max_cpu_time / 1000.0) + \ " --max-real-time " + str(self._max_real_time / 1000.0 * 2) + \ " --max-memory " + str(self._max_memory * 1000 * 1000) + \ " --network false" + \ " --remount-dev true " + \ " --reset-env true " + \ " --syscalls '" + self._language["syscalls"] + "'" + \ " --max-nprocess 20" + \ " --uid " + str(lrun_uid) + \ " --gid " + str(lrun_gid) execute_command = self._language["execute_command"].format(exe_path=self._exe_path) command += (" " + execute_command + # 0就是stdin " 0<" + self._test_case_dir + str(test_case_id) + ".in" + # 1就是stdout " 1>" + judger_workspace + str(test_case_id) + ".out" + # 3是stderr，包含lrun的输出和程序的异常输出 " 3>&2") return command def _parse_lrun_output(self, output): # 要注意的是 lrun把结果输出到了stderr，所以有些情况下lrun的输出可能与程序的一些错误输出的混合的，要先分离一下 error = None # 倒序找到MEMORY的位置,lrun的 MEMORY 输出后面有3个空格，而 EXCEEDED 也有可能是MEMORY，所以需要判断空格 output_start = output.rfind("MEMORY ") if output_start == -1: logger.error("Lrun result parse error") logger.error(output) raise JudgeClientError("Lrun result parse error") # 如果不是0，说明lrun输出前面有输出，也就是程序的stderr有内容 if output_start != 0: error = output[0:output_start] # 分离出lrun的输出 output = output[output_start:] return error, parse_lrun_output(output) def _compare_output(self, test_case_id): test_case_config = self._test_case_info["test_cases"][str(test_case_id)] output_path = judger_workspace + str(test_case_id) + ".out" try: f = open(output_path, "rb") except IOError: # 文件不存在等引发的异常 返回结果错误 return "", False if "striped_output_md5" not in test_case_config: # 计算输出文件的md5 和之前测试用例文件的md5进行比较 # 兼容之前没有striped_output_md5的测试用例 # 现在比较的是完整的文件 md5 = hashlib.md5() while True: data = f.read(2 ** 8) if not data: break md5.update(data) output_md5 = md5.hexdigest() return output_md5, output_md5 == test_case_config["output_md5"] else: # 这时候需要去除用户输出最后的空格和换行 再去比较md5 md5 = hashlib.md5() # 比较和返回去除空格后的md5比较结果 md5.update(f.read().rstrip()) output_md5 = md5.hexdigest() return output_md5, output_md5 == test_case_config["striped_output_md5"] def _judge_one(self, test_case_id): # 运行lrun程序 接收返回值 command = self._generate_command(test_case_id) status_code, output = commands.getstatusoutput(command) if status_code: raise JudgeClientError(output) error, run_result = self._parse_lrun_output(output) run_result["test_case_id"] = test_case_id # 代表内存或者时间超过限制了 程序被终止掉 要在runtime error 之前判断 if run_result["exceed"]: if run_result["exceed"] == "memory": run_result["result"] = result["memory_limit_exceeded"] elif run_result["exceed"] in ["cpu_time", "real_time"]: run_result["result"] = result["time_limit_exceeded"] else: logger.error("Error exceeded type: " + run_result["exceed"]) logger.error(output) raise JudgeClientError("Error exceeded type: " + run_result["exceed"]) return run_result # 如果返回值非0 或者信号量不是0 或者程序的stderr有输出 代表非正常结束 if run_result["exit_code"] or run_result["term_sig"] or run_result["siginaled"] or error: run_result["result"] = result["runtime_error"] return run_result # 下面就是代码正常运行了 需要判断代码的输出是否正确 output_md5, r = self._compare_output(test_case_id) if r: run_result["result"] = result["accepted"] else: run_result["result"] = result["wrong_answer"] run_result["output_md5"] = output_md5 return run_result def run(self): # 添加到任务队列 _results = [] results = [] for i in range(self._test_case_info["test_case_number"]): _results.append(self._pool.apply_async(_run, (self, i + 1))) self._pool.close() self._pool.join() for item in _results: # 注意多进程中的异常只有在get()的时候才会被引发 # http://stackoverflow.com/questions/22094852/how-to-catch-exceptions-in-workers-in-multiprocessing try: results.append(item.get()) except Exception as e: logger.error("system error") logger.error(e) results.append({"result": result["system_error"]}) return results def __getstate__(self): # 不同的pool之间进行pickle的时候要排除自己，否则报错 # http://stackoverflow.com/questions/25382455/python-notimplementederror-pool-objects-cannot-be-passed-between-processes self_dict = self.__dict__.copy() del self_dict['_pool'] return self_dict