[Bug]

未关闭
  • 哪里不对？
#2,255 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 2 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
40/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
c, postgresql
领域
backend, databases

调研方向

Start with src/backend/gpopt/translate/CTranslatorQueryToDXL.cpp and the optimizer_enable_multiple_distinct_aggs definition in guc_gp.c; the issue identifies the guard that causes fallback. Check how the optimizer handles multiple distinct qualified aggregates and review existing tests for that behavior. Done means establishing whether enabling the option is supported and safe, with evidence about plan behavior and the reason it defaults off.

由索引模型根据 Issue 内容生成。

描述

type: Bug
Apache Cloudberry version
Environment
  • Cloudberry version: <Ausgabe von SELECT version(); einfügen>
  • Segments: 4
  • optimizer = on
  • join_collapse_limit = 20, from_collapse_limit = 20 (defaults)
What happened
Environment
  • Cloudberry version: <Ausgabe von SELECT version(); einfügen>
  • Segments: 4
  • optimizer = on
  • join_collapse_limit = 20, from_collapse_limit = 20 (defaults)
Description

A BI tool (Yellowfin) generates queries with several COUNT(DISTINCT ...) on different columns.
GPORCA refuses these queries and falls back to the Postgres-based planner:

INFO:  GPORCA failed to produce a plan, falling back to planner
DETAIL:  Feature not supported: Multiple Distinct Qualified Aggregates are disabled in the optimizer

In our case the query joins ~20 tables (LEFT JOINs) with 7 COUNT(DISTINCT ...) and a wide GROUP BY.
After the fallback, the Postgres planner (with join_collapse_limit = 20) needs a very long time to plan:

Planning Time: 192018.556 ms
Optimizer: Postgres query optimizer
Execution Time: 50.880 ms

So the fallback itself turns a trivial query into a ~3 minute planning-only operation.

Questions
  1. Is this limitation still in place in the latest Cloudberry release, or is there a version where ORCA handles multiple DQAs by default?
  2. Is it safe/supported to set optimizer_enable_multiple_distinct_aggs = on in production (e.g. per role), given it is classified as a developer option? Are there known issues or plan-quality regressions (CXformGbAggWithMDQA2Join)?
  3. Is there a reason why the default is off (performance, correctness)?
  4. Are there recommendations to limit the planning time after an ORCA fallback for queries with many joins (e.g. lowering join_collapse_limit / from_collapse_limit for such roles)?

Thanks!

What you think should happen instead

No response

How to reproduce
Minimal reproduction
CREATE TABLE t_mdqa (a int, b int, c int) DISTRIBUTED BY (a);
SET optimizer = on;
SET optimizer_trace_fallback = on;
EXPLAIN SELECT count(DISTINCT a), count(DISTINCT b) FROM t_mdqa GROUP BY c;
-- -> falls back to Postgres-based planner

As far as I can see, the check is still present in current main
(src/backend/gpopt/translate/CTranslatorQueryToDXL.cpp:
if (1 < num_dqa && !optimizer_enable_multiple_distinct_aggs)),
and optimizer_enable_multiple_distinct_aggs is defined as a DEVELOPER_OPTIONS GUC in guc_gp.c.

Operating System

NAME="Red Hat Enterprise Linux" VERSION="8.10 (Ootpa)" ID="rhel" ID_LIKE="fedora" VERSION_ID="8.10" PLATFORM_ID="platform:el8" PRETTY_NAME="Red Hat Enterprise Linux 8.10 (Ootpa)" ANSI_COLOR="0;31" CPE_NAME="cpe:/o:redhat:enterprise_linux:8::baseos" HOME_URL="https://www.redhat.com/" DOCUMENTATION_URL="https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/8" BUG_REPORT_URL="https://issues.redhat.com/" REDHAT_BUGZILLA_PRODUCT="Red Hat Enterprise Linux 8" REDHAT_BUGZILLA_PRODUCT_VERSION=8.10 REDHAT_SUPPORT_PRODUCT="Red Hat Enterprise Linux" REDHAT_SUPPORT_PRODUCT_VERSION="8.10" Red Hat Enterprise Linux release 8.10 (Ootpa) Red Hat Enterprise Linux release 8.10 (Ootpa)

Anything else

No response

Are you willing to submit PR?
  • Yes, I am willing to submit a PR!
Code of Conduct
主要语言
C
星标
1.4k
派生
268
平均合并
4 天 3 小时
30 天内合并 PR
78

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。

apache/cloudberry 的其他 Issue

查看 apache/cloudberry 的全部 Issue

相似的 Issue

更多 C Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。