All Your GUCs in a Row: full_page_writes

PostgreSQL's full_page_writes GUC controls whether the database logs entire 8kB pages to WAL after each checkpoint, protecting against torn page writes caused by mismatched sector sizes. Because WAL records are physiological (not byte-level diffs), recovery cannot safely apply changes to a partially-written page, so PostgreSQL logs a full page image on the first modification after each checkpoint. This increases WAL volume in a sawtooth pattern tied to checkpoint intervals. Counterintuitively, shortening checkpoint intervals worsens this overhead; lengthening them and enabling wal_compression (lz4 or zstd) are the correct levers. Turning full_page_writes off is only defensible on storage that guarantees atomic 8kB writes (e.g., ZFS with matching recordsize), but the failure mode is silent data corruption discovered months later. Additionally, pg_basebackup and pg_rewind re-enable or require full page writes regardless of the setting, because they face torn-read hazards during concurrent writes. The recommendation is to leave it on.

Nguồn: https://postgr.es/p/9qq. 8sync News chỉ tóm tắt và dẫn link; bản quyền nội dung thuộc tác giả và nguồn gốc.