Hacktoberfest 2026: die Issues, die Maintainer für den Oktober markiert haben – offen und einsteigerfreundlich. Hacktoberfest-Issues durchsuchen

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

Offen
  • Stimmt etwas nicht?
#2,137 0 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 2 Tagen

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
3/5
Geschätzter Aufwand
Ein halber Tag
Anfängerfreundlichkeit
58/100
Issue-Typ
Bug
Klarheit
Klar beschrieben
Aktivitätsstatus
Aktiv
Tech-Stack
c, postgresql, sql
Bereich
databases, distributed-systems

Rechercherichtung

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.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

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
Vorherrschende Sprache
C
Sterne
1.4k
Forks
265
Ø Merge
3 T. 12 Std.
Gemergte PRs (30 T.)
60

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus apache/cloudberry

Alle Issues in apache/cloudberry

Ähnliche Issues

Weitere Issues zu C

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.