Double-write buffer
A MySQL/InnoDB mechanism that protects against "torn pages" — when a page is only partially written during a power failure. InnoDB first writes the page to a double-write buffer, then to its real location, so a correct copy always exists.
Problem: a database page (16 KB) is larger than what the disk guarantees to write atomically (often 4 KB sectors); a power failure mid-write can leave a half-written, corrupt page that the WAL/redo log can't repair (since redo assumes a consistent starting page). Solution: InnoDB first writes the page sequentially to the double-write buffer (on disk), and then to its actual position. On crash, a torn page can be restored from the intact copy in the buffer. Cost: every page is written twice (extra I/O) — can be turned off on filesystems/disks with atomic writes (some SSDs, ZFS). PostgreSQL solves the same problem with "full page writes" in the WAL instead. Related to redo log and checkpoint.