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

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

未关闭
  • 哪里不对？
#2,137 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 3 天内回复

还没有人认领这个 Issue。

评估

难度
3/5
预计耗时
半天
新手友好度
58/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
c, postgresql, sql
领域
databases, distributed-systems

调研方向

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.

由索引模型根据 Issue 内容生成。

描述

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
主要语言
C
星标
1.4k
派生
262
平均合并
4 天 20 小时
30 天内合并 PR
57

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。