Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

Cypher on Oracle: a list literal of 1,001 items read three times through a WITH spends 815 MB of PGA and keeps 217 MB

未关闭
  • 哪里不对？
#241 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

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
45/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
databases, performance, testing

调研方向

Start by reproducing the Oracle memory measurements described here, especially the 1,001-item list carried through WITH and read three times. Compare the generated SQL and PGA use with the single-read cases, then trace the Oracle compiler and connection execution path; the issue says the Oracle json_array and json_object compilation is unchanged. Done means the statement no longer causes excessive PGA use or ORA-04036 under the reported test-suite conditions, while the list literal test remains meaningful.

由索引模型根据 Issue 内容生成。

描述

bug

Found while porting liberation-data/drivine4j#362 (#192).

What happens. On Oracle, WITH <list of 1,001 items> AS xs RETURN [size(xs), xs[-1], xs[999]] makes a session spend 815 MB of PGA. The list mixes seven kinds of item, a null among them: [0, 's1', null, 3.5, true, [5, 'n'], {k: 6}, 7, ...]. The compiled statement is only 34 KB. After the statement ends, the session still holds 217 MB. Two such sessions, or one beside the rest of the test suite's sessions, pass Oracle Free's 2 GB PGA_AGGREGATE_LIMIT, and the statement fails with ORA-04036.

In two full runs of ORACLE_TESTS=true uv run pytest, the sync copy of #192's test failed on exactly this step with ORA-04036. Its async copy, run earlier in the same suite, passed both times, and so does the test run on its own.

Measured on a fresh gvenzl/oracle-free:23-faststart session, the statement prepared with cache_statement=False as OracleConnection runs it. Values are session pga memory max and, after the statement, session pga memory:

statement items SQL PGA peak PGA after
RETURN <list> 250 8 KB 51 MB 17 MB
WITH <list> AS xs RETURN [size(xs), xs[-1], xs[248]] 250 10 KB 214 MB 60 MB
RETURN <list> 1,001 33 KB 181 MB 50 MB
UNWIND <list> AS x RETURN x 1,001 33 KB 182 MB 50 MB
RETURN [x IN <list> WHERE x IS NOT NULL | x] 1,001 33 KB 182 MB 50 MB
WITH <list> AS xs RETURN [size(xs), xs[-1], xs[999]] 1,001 34 KB 815 MB 217 MB

Reading the carried list three times in the projection costs about 4.5 times what reading it once does. Each read is a CASE JSON_VALUE(w2.c1, '$.type()') ... over the whole array. Even once, a 1,001-item literal costs 181 MB.

Not caused by #192. That change leaves Oracle's SQL for these statements as it was (the Oracle compiler's json_array/json_object are untouched).

Workaround in the test. #192's test_a_list_literal_of_more_than_100_items_... asks Oracle no list of 1,001 items (see the comment below).

Other libraries. drivine (TS) compiles nearly the same Oracle SQL for this statement (42,170 vs 42,196 characters), so it should behave the same. Its port of the test passed in its full suite, where the Oracle sessions happen to leave enough headroom. drivine4j isn't measured.

主要语言
Python
星标
1
派生
0
平均合并
1 小时 21 分钟
30 天内合并 PR
128

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

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。

liberation-data/drivine-python 的其他 Issue

查看 liberation-data/drivine-python 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。