新增 SQL 题型:SQLite 内联判题(不依赖外部沙箱)

- SQL 作为语言接入现有提交流程,judge_task 按 language 分流到 SQLJudgeDispatcher
- judge/sql_runner.py:内存 SQLite 判题核心,查询题比结果集/增删改题比表状态,
  authorizer + progress_handler + max_page_count 三重防护
- dispatcher 提取 _process_judge_result/_push_status 供 SQL 判题复用(行为不变)
- 测试点通道支持 1.sql..N.sql 压缩包,出题只需数据脚本+标准答案
- Problem 增加 sql_config 字段;options 数据迁移注册 SQL 语言

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
2026-07-02 17:23:13 -06:00
parent d1a1c05b9f
commit 9f7f818d51
10 changed files with 551 additions and 93 deletions

View File

@@ -28,6 +28,7 @@ def process_pending_task():
if cache.llen(CacheKey.waiting_queue):
# 防止循环引入
from judge.tasks import judge_task
tmp_data = cache.rpop(CacheKey.waiting_queue)
if tmp_data:
data = json.loads(tmp_data.decode("utf-8"))
@@ -42,8 +43,7 @@ class ChooseJudgeServer:
with transaction.atomic():
cutoff = timezone.now() - timedelta(seconds=6)
server = (
JudgeServer.objects
.select_for_update(skip_locked=True)
JudgeServer.objects.select_for_update(skip_locked=True)
.filter(
is_disabled=False,
last_heartbeat__gte=cutoff,
@@ -78,8 +78,6 @@ class DispatcherBase(object):
logger.exception(e)
class JudgeDispatcher(DispatcherBase):
def __init__(self, submission_id, problem_id):
super().__init__()
@@ -93,6 +91,24 @@ class JudgeDispatcher(DispatcherBase):
else:
self.problem = Problem.objects.get(id=problem_id)
def _push_status(self, result, status, extra=None):
data = {
"type": "submission_update",
"submission_id": str(self.submission.id),
"result": result,
"status": status,
}
if extra:
data.update(extra)
try:
push_submission_update(
submission_id=str(self.submission.id),
user_id=self.submission.user_id,
data=data,
)
except Exception as e:
logger.error(f"Failed to push submission update: {str(e)}")
def _compute_statistic_info(self, resp_data):
# 用时和内存占用保存为多个测试点中最长的那个
self.submission.statistic_info["time_cost"] = max([x["cpu_time"] for x in resp_data])
@@ -131,7 +147,7 @@ class JudgeDispatcher(DispatcherBase):
"max_memory": 1024 * 1024 * self.problem.memory_limit,
"test_case_id": self.problem.test_case_id,
"output": False,
"io_mode": self.problem.io_mode
"io_mode": self.problem.io_mode,
}
with ChooseJudgeServer() as server:
@@ -139,57 +155,30 @@ class JudgeDispatcher(DispatcherBase):
data = {"submission_id": self.submission.id, "problem_id": self.problem.id}
cache.lpush(CacheKey.waiting_queue, json.dumps(data))
# 推送排队状态
try:
push_submission_update(
submission_id=str(self.submission.id),
user_id=self.submission.user_id,
data={
"type": "submission_update",
"submission_id": str(self.submission.id),
"result": JudgeStatus.PENDING,
"status": "pending",
}
)
except Exception as e:
logger.error(f"Failed to push submission update: {str(e)}")
self._push_status(JudgeStatus.PENDING, "pending")
return
Submission.objects.filter(id=self.submission.id).update(result=JudgeStatus.JUDGING)
# 推送判题中状态
try:
push_submission_update(
submission_id=str(self.submission.id),
user_id=self.submission.user_id,
data={
"type": "submission_update",
"submission_id": str(self.submission.id),
"result": JudgeStatus.JUDGING,
"status": "judging",
}
)
except Exception as e:
logger.error(f"Failed to push submission update: {str(e)}")
self._push_status(JudgeStatus.JUDGING, "judging")
resp = self._request(urljoin(server.service_url, "/judge"), data=data)
if not resp:
Submission.objects.filter(id=self.submission.id).update(result=JudgeStatus.SYSTEM_ERROR)
# 推送系统错误状态
try:
push_submission_update(
submission_id=str(self.submission.id),
user_id=self.submission.user_id,
data={
"type": "submission_update",
"submission_id": str(self.submission.id),
"result": JudgeStatus.SYSTEM_ERROR,
"status": "error",
}
)
except Exception as e:
logger.error(f"Failed to push submission update: {str(e)}")
self._push_status(JudgeStatus.SYSTEM_ERROR, "error")
return
self._process_judge_result(resp)
def _process_judge_result(self, resp):
"""判题结果的统一后处理状态聚合、AST 钩子、统计、排名、WebSocket 推送。
resp 结构与外部 judger 返回一致:{"err": "CompileError"|None, "data": ...}
SQLJudgeDispatcher 组装同构 resp 后也走这里。
"""
language = self.submission.language
if resp["err"]:
self.submission.result = JudgeStatus.COMPILE_ERROR
self.submission.statistic_info["err_info"] = resp["data"]
@@ -220,31 +209,22 @@ class JudgeDispatcher(DispatcherBase):
self.submission.save(update_fields=["result", "info", "statistic_info"])
# 推送判题完成状态
try:
push_submission_update(
submission_id=str(self.submission.id),
user_id=self.submission.user_id,
data={
"type": "submission_update",
"submission_id": str(self.submission.id),
"result": self.submission.result,
"status": "finished",
"time_cost": self.submission.statistic_info.get("time_cost"),
"memory_cost": self.submission.statistic_info.get("memory_cost"),
"score": self.submission.statistic_info.get("score", 0),
}
)
except Exception as e:
logger.error(f"Failed to push submission update: {str(e)}")
self._push_status(
self.submission.result,
"finished",
extra={
"time_cost": self.submission.statistic_info.get("time_cost"),
"memory_cost": self.submission.statistic_info.get("memory_cost"),
"score": self.submission.statistic_info.get("score", 0),
},
)
if self.contest_id:
# 以提交时刻(而非判题时刻)是否落在比赛时间窗内为准,
# 避免临界提交因判题排队延迟到比赛结束后才处理而被丢弃
in_contest = self.contest.start_time <= self.submission.create_time <= self.contest.end_time
if not in_contest or \
User.objects.get(id=self.submission.user_id).is_contest_admin(self.contest):
logger.info(
"Contest debug mode, id: " + str(self.contest_id) + ", submission id: " + self.submission.id)
if not in_contest or User.objects.get(id=self.submission.user_id).is_contest_admin(self.contest):
logger.info("Contest debug mode, id: " + str(self.contest_id) + ", submission id: " + self.submission.id)
return
with transaction.atomic():
self.update_contest_problem_status()
@@ -286,8 +266,7 @@ class JudgeDispatcher(DispatcherBase):
score = self.submission.statistic_info["score"]
if not is_accepted(oi_problems_status[problem_id]["status"]):
# minus last time score, add this tim score
profile.add_score(this_time_score=score,
last_time_score=oi_problems_status[problem_id]["score"])
profile.add_score(this_time_score=score, last_time_score=oi_problems_status[problem_id]["score"])
oi_problems_status[problem_id]["score"] = score
oi_problems_status[problem_id]["status"] = JudgeStatus.ACCEPTED if is_accepted(self.submission.result) else self.submission.result
if is_accepted(self.submission.result):
@@ -331,15 +310,12 @@ class JudgeDispatcher(DispatcherBase):
score = self.submission.statistic_info["score"]
if problem_id not in oi_problems_status:
user_profile.add_score(score)
oi_problems_status[problem_id] = {"status": profile_status,
"_id": self.problem._id,
"score": score}
oi_problems_status[problem_id] = {"status": profile_status, "_id": self.problem._id, "score": score}
if is_accepted(self.submission.result):
user_profile.accepted_number += 1
elif not is_accepted(oi_problems_status[problem_id]["status"]):
# minus last time score, add this time score
user_profile.add_score(this_time_score=score,
last_time_score=oi_problems_status[problem_id]["score"])
user_profile.add_score(this_time_score=score, last_time_score=oi_problems_status[problem_id]["score"])
oi_problems_status[problem_id]["score"] = score
oi_problems_status[problem_id]["status"] = profile_status
if is_accepted(self.submission.result):
@@ -425,4 +401,3 @@ class JudgeDispatcher(DispatcherBase):
info["error_number"] = 1
rank.submission_info[str(self.submission.problem_id)] = info
rank.save(update_fields=["submission_info", "total_time", "accepted_number", "submission_number"])

View File

@@ -174,4 +174,6 @@ languages = [
{"config": _py3_lang_config, "name": "Python3", "description": "Python 3.12", "content_type": "text/x-python"},
{"config": _go_lang_config, "name": "Golang", "description": "Golang 1.22", "content_type": "text/x-go"},
{"config": _node_lang_config, "name": "JavaScript", "description": "Node.js 20", "content_type": "text/javascript"},
# SQL 题不走外部 judger 沙箱judge/sql_dispatcher.py 在 worker 内用 sqlite3 判题config 仅占位
{"config": {"template": ""}, "name": "SQL", "description": "SQLite 3", "content_type": "text/x-sql"},
]

101
judge/sql_dispatcher.py Normal file
View File

@@ -0,0 +1,101 @@
"""SQL 题判题调度:不走 JudgeServer 沙箱,在 dramatiq worker 内用 sqlite3 直接判题。
复用 JudgeDispatcher 的 _process_judge_result 完成状态聚合、统计、排名和 WebSocket 推送,
因此比赛排名、rejudge、题目统计的语义与沙箱判题完全一致。
"""
import json
import logging
import os
from django.conf import settings
from judge.dispatcher import JudgeDispatcher
from judge.sql_runner import SQLCaseError, run_case
from submission.models import JudgeStatus, Submission
from utils.shortcuts import natural_sort_key
logger = logging.getLogger(__name__)
class SQLProblemConfigError(Exception):
pass
class SQLJudgeDispatcher(JudgeDispatcher):
def judge(self):
Submission.objects.filter(id=self.submission.id).update(result=JudgeStatus.JUDGING)
self._push_status(JudgeStatus.JUDGING, "judging")
try:
ref_sql, mode, order_sensitive, init_scripts = self._load_problem_config()
except SQLProblemConfigError as e:
self._system_error(str(e))
return
cases = []
for index, init_sql in enumerate(init_scripts, start=1):
try:
case = run_case(
init_sql,
ref_sql,
self.submission.code,
mode=mode,
order_sensitive=order_sensitive,
time_limit_ms=self.problem.time_limit,
memory_limit_mb=self.problem.memory_limit,
)
except SQLCaseError as e:
# 初始化/标准答案执行失败,属出题配置问题
self._system_error(e.message)
return
case["test_case"] = str(index)
# 语法错误与数据无关首个测试点即可确认整题按编译错误处理ACM 不罚时,前端展示 err_info
if index == 1 and case["result"] == JudgeStatus.COMPILE_ERROR:
self._process_judge_result({"err": "CompileError", "data": case["error_message"]})
return
cases.append(case)
self._process_judge_result({"err": None, "data": cases})
def _system_error(self, message):
logger.error(f"SQL judge system error, submission {self.submission.id}, problem {self.problem.id}: {message}")
Submission.objects.filter(id=self.submission.id).update(result=JudgeStatus.SYSTEM_ERROR, statistic_info={"err_info": message})
self._push_status(JudgeStatus.SYSTEM_ERROR, "error")
def _load_problem_config(self):
"""校验并加载 SQL 题配置,返回 (标准答案, mode, order_sensitive, 各测试点初始化脚本)。"""
sql_config = self.problem.sql_config or {}
mode = sql_config.get("mode")
if mode not in ("query", "modify"):
raise SQLProblemConfigError("题目缺少 SQL 配置(题型)")
ref_sql = None
for item in self.problem.answers or []:
if item.get("language") == "SQL" and item.get("code", "").strip():
ref_sql = item["code"]
break
if not ref_sql:
raise SQLProblemConfigError("题目缺少 SQL 标准答案")
test_case_dir = os.path.join(settings.TEST_CASE_DIR, self.problem.test_case_id)
try:
with open(os.path.join(test_case_dir, "info"), encoding="utf-8") as f:
info = json.load(f)
except (OSError, json.JSONDecodeError) as e:
raise SQLProblemConfigError(f"测试点信息读取失败: {e}")
if not info.get("sql"):
raise SQLProblemConfigError("测试点不是 SQL 类型,请重新上传 SQL 测试点压缩包")
init_scripts = []
# 按 "1","2",… 自然序遍历,保证与 test_case_score 的下标对应OI 计分依赖顺序)
for key in sorted(info["test_cases"].keys(), key=natural_sort_key):
input_name = info["test_cases"][key]["input_name"]
try:
with open(os.path.join(test_case_dir, input_name), encoding="utf-8") as f:
init_scripts.append(f.read())
except OSError as e:
raise SQLProblemConfigError(f"测试点脚本 {input_name} 读取失败: {e}")
if not init_scripts:
raise SQLProblemConfigError("题目没有任何测试点")
return ref_sql, mode, sql_config.get("order_sensitive", False), init_scripts

246
judge/sql_runner.py Normal file
View File

@@ -0,0 +1,246 @@
"""SQL 题判题核心:在内存 SQLite 中分别执行标准答案和学生 SQL 并比对结果。
查询题mode="query")比对最后一条 SELECT 的结果集;
增删改题mode="modify")比对执行后所有用户表的最终状态。
学生 SQL 通过 authorizer禁 ATTACH/PRAGMA查询题只读
progress_handler墙钟超时和 max_page_count内存上限三重防护。
"""
import sqlite3
import time
from submission.models import JudgeStatus
# 单结果集/单表最大行数,防 CROSS JOIN 撑爆 worker 内存
ROW_LIMIT = 10_000
# progress_handler 检查粒度SQLite VM 指令数)
PROGRESS_STEP = 1_000
ERROR_MESSAGE_MAX_LEN = 200
# prepare 阶段的语法类错误,映射为 COMPILE_ERROR
_SYNTAX_ERROR_MARKERS = ("syntax error", "unrecognized token", "incomplete input")
# 两种模式都禁止的授权码:挂载外部库 / 数据库参数
_DENIED_ALWAYS = {sqlite3.SQLITE_ATTACH, sqlite3.SQLITE_DETACH, sqlite3.SQLITE_PRAGMA}
# 查询题允许的授权码(白名单外一律拒绝,防先 INSERT 伪造数据再 SELECT
_QUERY_MODE_ALLOWED = {getattr(sqlite3, name) for name in ("SQLITE_SELECT", "SQLITE_READ", "SQLITE_FUNCTION", "SQLITE_RECURSIVE", "SQLITE_TRANSACTION") if hasattr(sqlite3, name)}
class SQLCaseError(Exception):
"""携带 JudgeStatus 的判题异常。SYSTEM_ERROR 级别(出题配置问题)会传播到 dispatcher。"""
def __init__(self, result, message):
super().__init__(message)
self.result = result
self.message = message
def _truncate(message):
message = str(message)
if len(message) > ERROR_MESSAGE_MAX_LEN:
return message[:ERROR_MESSAGE_MAX_LEN] + "..."
return message
def split_statements(script):
"""用 sqlite3.complete_statement 累积切分多条语句,正确处理字符串/注释里的分号;末尾缺分号自动补。"""
statements = []
buf = ""
for part in script.split(";"):
buf += part + ";"
if sqlite3.complete_statement(buf):
stmt = buf.strip()
buf = ""
if stmt and stmt != ";":
statements.append(stmt)
# 残句(未闭合的引号/注释,或末尾缺分号但上面已补),交给 execute 报错或执行
remainder = buf.strip()
if remainder and remainder != ";":
statements.append(remainder)
return statements
def _canonical_value(v):
"""值归一化并打类型标签,防止 NULL/"NULL"、1/"1" 碰撞数值统一比对1 == 1.0,浮点 6 位有效数字)。"""
if v is None:
return ("null",)
if isinstance(v, int):
return ("num", str(v))
if isinstance(v, float):
if v.is_integer() and abs(v) < 2**53:
return ("num", str(int(v)))
return ("num", format(v, ".6g"))
if isinstance(v, bytes):
return ("blob", v.hex())
return ("str", str(v))
def _canonical_row(row):
return tuple(_canonical_value(v) for v in row)
def _new_db(memory_limit_mb):
conn = sqlite3.connect(":memory:", isolation_level=None) # autocommit脚本行为可预期
conn.execute("PRAGMA page_size=4096")
# 4096B/页 × 256 页/MB超限报 "database or disk is full"
conn.execute(f"PRAGMA max_page_count={max(int(memory_limit_mb), 1) * 256}")
return conn
def _execute_statements(conn, script, deadline=None):
"""逐条执行,返回最后一条产生结果集的语句的 (列数, 行列表);无结果集返回 None。"""
last_result = None
for stmt in split_statements(script):
if deadline is not None and time.monotonic() > deadline:
raise sqlite3.OperationalError("interrupted")
cursor = conn.execute(stmt)
if cursor.description is not None:
rows = cursor.fetchmany(ROW_LIMIT + 1)
if len(rows) > ROW_LIMIT:
raise SQLCaseError(JudgeStatus.MEMORY_LIMIT_EXCEEDED, f"查询结果超过 {ROW_LIMIT}")
last_result = (len(cursor.description), [_canonical_row(r) for r in rows])
cursor.close()
return last_result
def _dump_tables(conn):
"""dump 所有用户表:{表名: (列数, 行多重集)},行内排序,表状态天然无序。"""
cursor = conn.execute("SELECT name FROM sqlite_master WHERE type='table' AND name NOT LIKE 'sqlite_%' ORDER BY name")
tables = [r[0] for r in cursor.fetchall()]
state = {}
for table in tables:
quoted = table.replace('"', '""')
cur = conn.execute(f'SELECT * FROM "{quoted}"')
rows = cur.fetchmany(ROW_LIMIT + 1)
if len(rows) > ROW_LIMIT:
raise SQLCaseError(JudgeStatus.MEMORY_LIMIT_EXCEEDED, f"{table} 超过 {ROW_LIMIT}")
state[table] = (len(cur.description), sorted(_canonical_row(r) for r in rows))
return state
def _execute_trusted(conn, script, deadline, error_prefix):
"""执行受信脚本,任何失败都是出题问题 → SYSTEM_ERROR。"""
conn.set_progress_handler(lambda: 1 if time.monotonic() > deadline else 0, PROGRESS_STEP)
try:
return _execute_statements(conn, script, deadline)
except SQLCaseError as e:
raise SQLCaseError(JudgeStatus.SYSTEM_ERROR, f"{error_prefix}: {e.message}")
except sqlite3.Error as e:
raise SQLCaseError(JudgeStatus.SYSTEM_ERROR, f"{error_prefix}: {_truncate(e)}")
finally:
conn.set_progress_handler(None, PROGRESS_STEP)
def _run_student(conn, script, mode, deadline):
"""带三重防护执行学生 SQL异常映射为学生级 JudgeStatus。"""
denied_hints = []
def authorizer(action, arg1, arg2, db_name, trigger):
if action in _DENIED_ALWAYS:
denied_hints.append("禁止使用 ATTACH/DETACH/PRAGMA 等数据库管理语句")
return sqlite3.SQLITE_DENY
if mode == "query" and action not in _QUERY_MODE_ALLOWED:
denied_hints.append("本题为查询题禁止修改数据或表结构INSERT/UPDATE/DELETE/CREATE 等)")
return sqlite3.SQLITE_DENY
return sqlite3.SQLITE_OK
conn.set_authorizer(authorizer)
conn.set_progress_handler(lambda: 1 if time.monotonic() > deadline else 0, PROGRESS_STEP)
try:
last_result = _execute_statements(conn, script, deadline)
except sqlite3.Error as e:
# 注意authorizer 拒绝抛的是 DatabaseError 基类而非 OperationalError统一按消息映射
msg = str(e)
if "interrupted" in msg:
raise SQLCaseError(JudgeStatus.CPU_TIME_LIMIT_EXCEEDED, "SQL 执行超时")
if "database or disk is full" in msg:
raise SQLCaseError(JudgeStatus.MEMORY_LIMIT_EXCEEDED, "数据量超出内存限制")
if "not authorized" in msg or "prohibited" in msg:
hint = denied_hints[-1] if denied_hints else "本题禁止使用该语句"
raise SQLCaseError(JudgeStatus.RUNTIME_ERROR, hint)
if any(marker in msg for marker in _SYNTAX_ERROR_MARKERS):
raise SQLCaseError(JudgeStatus.COMPILE_ERROR, _truncate(msg))
raise SQLCaseError(JudgeStatus.RUNTIME_ERROR, _truncate(msg))
finally:
conn.set_progress_handler(None, PROGRESS_STEP)
conn.set_authorizer(None)
if mode == "query":
return last_result
# dump 是我们自己的读取,不应吃学生的超时/授权限制(上面已清除)
return _dump_tables(conn)
def _compare(expected, actual, mode, order_sensitive):
if mode == "query":
exp_cols, exp_rows = expected
act_cols, act_rows = actual
if exp_cols != act_cols:
return False
if order_sensitive:
return exp_rows == act_rows
return sorted(exp_rows) == sorted(act_rows)
# modify: dump dict 里行已排序
return expected == actual
def run_case(init_sql, ref_sql, student_sql, *, mode, order_sensitive, time_limit_ms, memory_limit_mb):
"""判一个测试点,返回与外部 judger 单测试点同构的 dict。
学生错误CE/WA/TLE/MLE/RE体现在返回值里
出题配置错误(初始化/标准答案失败)抛 SQLCaseError(SYSTEM_ERROR),由 dispatcher 处理。
"""
time_limit_s = time_limit_ms / 1000
# 受信脚本(初始化/标准答案)的运行上限放宽,避免出题数据较大时误报;仍防 worker 永久阻塞
trusted_limit_s = max(time_limit_s * 5, 10)
ref_conn = _new_db(memory_limit_mb)
try:
_execute_trusted(ref_conn, init_sql, time.monotonic() + trusted_limit_s, "初始化脚本执行失败")
last_result = _execute_trusted(ref_conn, ref_sql, time.monotonic() + trusted_limit_s, "标准答案执行失败")
if mode == "query":
expected = last_result
else:
try:
expected = _dump_tables(ref_conn)
except SQLCaseError as e:
raise SQLCaseError(JudgeStatus.SYSTEM_ERROR, f"标准答案结果超出限制: {e.message}")
finally:
ref_conn.close()
if mode == "query" and expected is None:
raise SQLCaseError(JudgeStatus.SYSTEM_ERROR, "标准答案未产生查询结果集")
case = {
"test_case": "",
"result": JudgeStatus.ACCEPTED,
"cpu_time": 0,
"real_time": 0,
"memory": 0,
"signal": 0,
"exit_code": 0,
"error": 0,
"output_md5": "",
"error_message": None,
}
stu_conn = _new_db(memory_limit_mb)
try:
_execute_trusted(stu_conn, init_sql, time.monotonic() + trusted_limit_s, "初始化脚本执行失败")
start = time.monotonic()
try:
actual = _run_student(stu_conn, student_sql, mode, start + time_limit_s)
except SQLCaseError as e:
elapsed = int((time.monotonic() - start) * 1000)
case.update(result=e.result, error_message=e.message, cpu_time=elapsed, real_time=elapsed)
return case
elapsed = int((time.monotonic() - start) * 1000)
finally:
stu_conn.close()
case["cpu_time"] = case["real_time"] = elapsed
if mode == "query" and actual is None:
case.update(result=JudgeStatus.WRONG_ANSWER, error_message="提交的 SQL 未产生查询结果集")
elif not _compare(expected, actual, mode, order_sensitive):
case["result"] = JudgeStatus.WRONG_ANSWER
return case

View File

@@ -2,13 +2,18 @@ import dramatiq
from account.models import User
from judge.dispatcher import JudgeDispatcher
from judge.sql_dispatcher import SQLJudgeDispatcher
from submission.models import Submission
from utils.shortcuts import DRAMATIQ_WORKER_ARGS
@dramatiq.actor(**DRAMATIQ_WORKER_ARGS())
def judge_task(submission_id, problem_id):
uid = Submission.objects.get(id=submission_id).user_id
if User.objects.get(id=uid).is_disabled:
submission = Submission.objects.get(id=submission_id)
if User.objects.get(id=submission.user_id).is_disabled:
return
JudgeDispatcher(submission_id, problem_id).judge()
# SQL 题不依赖 JudgeServer 沙箱,在 worker 内用 sqlite3 判题
if submission.language == "SQL":
SQLJudgeDispatcher(submission_id, problem_id).judge()
else:
JudgeDispatcher(submission_id, problem_id).judge()