run_pipeline() never completes when the server side closes the connection during the pipeline; call_timeout and asyncio.wait_for don't end it

未关闭
  • 哪里不对？
#630 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
postgresql, python
领域
backend-api-design, database

调研方向

Start by tracing the async run_pipeline() implementation and how it handles socket closure, call timeouts, and task cancellation; the report gives a Python 3.12 Thin mode reproducer using connect_async. Reproduce with a TCP proxy that closes the connection after login, then compare the result with a direct database connection. Done when the pipeline raises on connection closure and timeout or cancellation ends the wait.

由索引模型根据 Issue 内容生成。

描述

Versions

  • python-oracledb 4.0.1, Thin mode (connect_async)
  • Python 3.12.3, Linux x86_64
  • Oracle Database 23ai Free (container image), reached through a TCP proxy

What happens

The connection goes through a TCP proxy that inspects SQL. When the proxy refuses the request carrying the pipeline, it closes both TCP connections without sending a reply. await conn.run_pipeline(pipeline) then never returns or raises: there is no error and no result.

Ten seconds in, the client process holds no TCP socket to the proxy, only what looks like asyncio's internal socket pair. So python-oracledb has seen the close and dropped its socket, but the pending pipeline is never completed.

What we tried (each in its own process, killed by an outside timeout at 150 s):

Setting Result through the proxy Direct to the database
none no answer, killed at 150 s 6 results, 0.2 s
conn.call_timeout = 5000 no answer, 150 s ok
expire_time=1 no answer, 150 s ok
asyncio.wait_for(conn.run_pipeline(p), 10) no answer, 150 s (the cancellation hangs too) ok
call_timeout + expire_time no answer, 150 s ok

Expected

When the connection is closed while a pipeline is waiting for replies, run_pipeline() should raise (for example DPY-4011, "the database or network closed the connection"), as other calls on a closed connection do. call_timeout and task cancellation should also end the wait.

Repro sketch

  1. Put any TCP proxy between the client and a 23ai database that forwards the login, then closes both sides when it receives the first request after login.
  2. Run: 
    import asyncio, oracledb

async def main():
    conn = await oracledb.connect_async(user=..., password=..., dsn="proxyhost:port/FREEPDB1")
    p = oracledb.create_pipeline()
    for _ in range(6):
        p.add_fetchone("select 1 from dual")
    print(await conn.run_pipeline(p))   # never returns

asyncio.run(main())

We can run a debug build or extra logging if that helps.

主要语言
Python
星标
457
派生
120
30 天内没有已合并 PR

环境准备

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

oracle/python-oracledb 的其他 Issue

查看 oracle/python-oracledb 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。