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] Mirrors crash during WAL replay when wal_compression is pglz/lz4

オープン
  • 何か問題がありますか？
#2,137 コメント 0 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
3/5
見積もり時間
半日
初心者へのやさしさ
58/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
c, postgresql, sql
領域
databases, distributed-systems

調査の方向性

Start in RestoreBlockImage() in src/backend/access/transam/xlogreader.c (around line 2096), where zstd_decompress_backupblock() runs before the bimg_info method check. Compare it with XLogCompressBackupBlock() on the write side to see which methods write images. Done when pglz and lz4 images replay without error; reproduce with the SQL in the issue on a cluster with mirrors.

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 (on a cluster with mirrors):

  CREATE TABLE w (a int, b text) DISTRIBUTED BY (a);
  INSERT INTO w SELECT i, repeat('x', 100) FROM generate_series(1, 1000) i;
  CHECKPOINT;
  SET wal_compression = pglz;
  UPDATE w SET b = 'y' WHERE a < 100;

Actual: the UPDATE succeeds, but all segment mirrors stop WAL replay and shut down, and FTS marks them down:
FATAL: invalid compressed image at 0/C7CFF070, block 0 (error computing decompression size)
LOG: startup process (PID ...) exited with exit code 1
LOG: shutting down due to startup process failure
gprecoverseg (incremental) reports success, but the mirrors replay the same WAL record and crash again. Only gprecoverseg -F (full recovery) brings them back.

Expected: the mirrors replay the WAL normally, as they do with SET wal_compression = zstd. With zstd, the same steps produce no errors.

Cause: in RestoreBlockImage() (src/backend/access/transam/xlogreader.c, around line 2096), every compressed page image is passed to zstd_decompress_backupblock() before the code checks bimg_info for the method
actually used. That zstd-only call dates from commit 60915ec6b2e ("Use zstd for wal_compression", 2020), when Greenplum always compressed WAL with zstd. Since the merges from PostgreSQL,
XLogCompressBackupBlock() writes images with the method chosen by wal_compression (pglz, lz4 or zstd), so pglz and lz4 images reach the zstd decoder and fail. A likely fix is to remove the unconditional zstd
call and rely on the per-method code that follows it.

Impact:

  • A session-level SET wal_compression plus an ordinary write is enough. It leaves the cluster without mirror protection until an administrator runs a full recovery.
  • Not verified: lz4 (expected to fail the same way, since the same code runs first), and whether a primary's own crash recovery also fails after replaying such a record.
What you think should happen instead

No response

How to reproduce

Reproduce (on a cluster with mirrors):

  CREATE TABLE w (a int, b text) DISTRIBUTED BY (a);
  INSERT INTO w SELECT i, repeat('x', 100) FROM generate_series(1, 1000) i;
  CHECKPOINT;
  SET wal_compression = pglz;
  UPDATE w SET b = 'y' WHERE a < 100;
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。