Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

[Bug] GPORCA crashes the coordinator with dependencies extended statistics

オープン
  • 何か問題がありますか？
#2,135 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 3 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
30/100
issue の種類
バグ
明瞭さ
おおむね明確
活発さ
活発
技術スタック
c, cpp, postgresql, sql
領域
databases

調査の方向性

Start by running the reproduction SQL with optimizer = on to confirm the SIGSEGV and capture a backtrace. The payload points to CExtendedStatsProcessor.cpp, where colid_to_attno_mapping->Find(&colid) is dereferenced without a null check around lines 279 and 355/365, while the check near line 431 does test for null. Done means the query returns one row (NULL | NULL | f) with optimizer on and the crash no longer occurs, with a regression test added.

索引モデルが issue の本文から書いたものです。

説明

type: Bug
Apache Cloudberry version

No response

What happened

Reproduce:

CREATE TABLE t1 (c0 name, c1 numeric, c2 boolean) DISTRIBUTED BY (c2);
INSERT INTO t1 VALUES ('a', 0.9, false), (NULL, NULL, false), ('', 0.1, false), ('1', 0.7, true), ('b', 0.5, NULL), ('', NULL, NULL), ('c', NULL, NULL);
CREATE STATISTICS s1 ON c0, c2 FROM t1;
CREATE STATISTICS s2 ON c1, c2 FROM t1;
ANALYZE t1;

SET optimizer = on;

SELECT * FROM (SELECT c0 AS a, c1 AS b, c2 AS c FROM t1 WHERE c2 IS FALSE) AS r WHERE r.a IS NULL AND r.b IS NULL AND r.c = FALSE;
  -- server closed the connection unexpectedly

Expected: one row, NULL | NULL | f. That's what you get with optimizer = off, or with optimizer = on after dropping s1 and s2.

  Actual: the coordinator backend crashes, and all other sessions are reset while the cluster recovers:
  PANIC: Unexpected internal error: Master process received signal SIGSEGV
  server process (PID ...) was terminated by signal 11: Segmentation fault

  Stack (top frames):
  gpnaucrates::CFilterStatsProcessor::MakeHistHashMapConjFilter
  gpnaucrates::CFilterStatsProcessor::MakeStatsFilter
  gpnaucrates::CFilterStatsProcessor::MakeStatsFilterForScalarExpr
  gpopt::CLogicalSelect::PstatsDerive
  gpopt::CExpressionHandle::DeriveRootStats
  ...
  gpopt::CEngine::FinalizeExploration

Possible cause: MakeHistHashMapConjFilter calls CExtendedStatsProcessor::ApplyCorrelatedStatsToScaleFactorFilterCalculation. In that function, the result of colid_to_attno_mapping->Find(&colid) is dereferenced
without a null check (CExtendedStatsProcessor.cpp, around lines 279 and 355/365), while another call site (around line 431) does check for null. The filter here sits above a subquery and spans columns from two
different dependency statistics objects, which may produce a column ID with no mapping. I haven't confirmed this in a debugger.

What you think should happen instead

No response

How to reproduce
CREATE TABLE t1 (c0 name, c1 numeric, c2 boolean) DISTRIBUTED BY (c2);
INSERT INTO t1 VALUES ('a', 0.9, false), (NULL, NULL, false), ('', 0.1, false), ('1', 0.7, true), ('b', 0.5, NULL), ('', NULL, NULL), ('c', NULL, NULL);
CREATE STATISTICS s1 ON c0, c2 FROM t1;
CREATE STATISTICS s2 ON c1, c2 FROM t1;
ANALYZE t1;

SET optimizer = on;

SELECT * FROM (SELECT c0 AS a, c1 AS b, c2 AS c FROM t1 WHERE c2 IS FALSE) AS r WHERE r.a IS NULL AND r.b IS NULL AND r.c = FALSE;
  -- server closed the connection unexpectedly
Operating System

any

Anything else

No response

Are you willing to submit PR?
  • Yes, I am willing to submit a PR!
Code of Conduct
主要言語
C
スター
1.4k
フォーク
262
平均マージ
4日 20時間
マージ済み PR（30日）
57

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

apache/cloudberry のほかの issue

apache/cloudberry の issue をすべて見る

似ている issue

C の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。