{
  "topic": "mysql-acid",
  "type": "fill_blank",
  "schema_version": "1.0.0",
  "generated": "2026-09-07T00:00:00+08:00",
  "questions": [
    {
      "id": "fb-001",
      "type": "fill_blank",
      "difficulty": 1,
      "tags": [
        "原子性",
        "undo log",
        "InnoDB"
      ],
      "question": "InnoDB 保证事务原子性依赖的日志是 ____，它记录的是修改前的旧值，回滚时据此恢复数据。",
      "answer": [
        "undo log"
      ],
      "answer_rule": "any",
      "explanation": "undo log 记录的是事务修改前的旧值（逻辑逆操作）。当事务回滚或崩溃恢复时，InnoDB 依据 undo log 将数据恢复到事务开始前的状态，从而保证原子性（要么全部提交要么全部回滚）。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-002",
      "type": "fill_blank",
      "difficulty": 1,
      "tags": [
        "持久性",
        "WAL",
        "redo log"
      ],
      "question": "InnoDB 保证持久性采用 Write-Ahead Logging 策略，即在数据页落盘前必须先将对应的 ____ 日志写入磁盘。",
      "answer": [
        "redo log"
      ],
      "answer_rule": "any",
      "explanation": "WAL 的核心是“写日志先行”：任何数据页修改在刷入磁盘之前，必须先把对应的 redo log 记录写入磁盘。这样一旦崩溃，通过重放 redo log 即可恢复未落盘的已提交修改，从而保证持久性。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-003",
      "type": "fill_blank",
      "difficulty": 2,
      "tags": [
        "redo log",
        "物理日志",
        "环形缓冲",
        "InnoDB"
      ],
      "question": "InnoDB 的 redo log 在逻辑上是固定大小的环形缓冲，写入是循环的，当写满时会覆盖最 ____ 的日志。",
      "answer": [
        "旧"
      ],
      "answer_rule": "any",
      "explanation": "redo log 由若干固定大小的文件组成环形结构，物理日志循环写入，当文件写满会回绕覆盖最旧（最早写入）的日志记录。因此 redo log 并不记录全部历史，而只保留最近一段时间的重做记录。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-004",
      "type": "fill_blank",
      "difficulty": 2,
      "tags": [
        "redo log",
        "物理日志",
        "undo log",
        "逻辑日志"
      ],
      "question": "redo log 记录的是对数据页字节修改的____日志（物理层），而 undo log 记录的是如何“改回旧值”的____日志（逻辑层）。",
      "answer": [
        "物理",
        "逻辑"
      ],
      "answer_rule": "ordered",
      "explanation": "redo log 属于物理日志，描述对哪个页的哪个位置的字节进行了什么修改，用于崩溃后重放；undo log 属于逻辑日志，记录的是把数据改回旧状态所需的逆操作逻辑。两个空位必须按顺序填“物理 / 逻辑”，所以使用 ordered 规则。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-005",
      "type": "fill_blank",
      "difficulty": 2,
      "tags": [
        "MVCC",
        "undo log",
        "隔离性"
      ],
      "question": "InnoDB 的 MVCC 通过 undo log 维护行的多____链，配合事务的 ____ 快照决定当前读可见的版本，从而在不加锁的情况下实现一致性的快照读。",
      "answer": [
        "版本",
        "ReadView"
      ],
      "answer_rule": "ordered",
      "explanation": "InnoDB 通过 undo log 维护行的多版本链，每次更新原记录时旧版本被放入版本链。一致性读（快照读）通过 ReadView（一个事务范围内的可见性判断结构）沿版本链找到当前事务可见的版本，实现高并发的非锁定读。两空位按顺序填“版本 / ReadView”。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-006",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "崩溃恢复",
        "redo log",
        "undo log"
      ],
      "question": "InnoDB 崩溃恢复的三阶段为：一、扫描 ____ 确定需要恢复的范围；二、对其中的已提交事务进行 ____；三、对未提交事务用 ____ 进行回滚。",
      "answer": [
        "redo log",
        "重做",
        "undo log"
      ],
      "answer_rule": "ordered",
      "explanation": "崩溃恢复流程：先扫描 redo log 确定恢复起点；然后重做（应用）崩溃前已提交但未刷盘的事务，使其修改落到数据页；最后根据两阶段提交状态，对未提交事务用 undo log 回滚，恢复到崩溃前的一致状态。三个空位需按顺序作答。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-007",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "binlog",
        "redo log",
        "两阶段提交"
      ],
      "question": "为协调 InnoDB 与 binlog 的一致性，MySQL 对两者采用____阶段提交：先将 redo log 写入____（prepare）阶段，再写 binlog，最后把 redo 标记为 commit。",
      "answer": [
        "两",
        "prepare"
      ],
      "answer_rule": "ordered",
      "explanation": "MySQL 对 redo log 与 binlog 用两阶段提交（2PC / InnoDB&binlog 内部协调）：第一步把 redo log 写入 prepare 阶段（此刻事务处于预提交），第二步写 binlog，第三步将 redo 标记为 commit。崩溃重启时依据 binlog 是否完整来决定事务是提交还是回滚，从而保证两者一致。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-008",
      "type": "fill_blank",
      "difficulty": 2,
      "tags": [
        "innodb_flush_log_at_trx_commit",
        "持久性"
      ],
      "question": "innodb_flush_log_at_trx_commit=1 时，每次事务提交都会调用 ____ 强制把 redo log 刷到磁盘，保证最强的持久性（也最浪费 IO）。",
      "answer": [
        "fsync"
      ],
      "answer_rule": "any",
      "explanation": "取值 1 意味着每次提交时 InnoDB 都调用 fsync 把 redo log 从 OS 缓存强制刷入磁盘，确保提交的事务真正持久化，因此符合严格 ACID 的持久性要求，但会带来较多磁盘同步 IO。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-009",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "redo log",
        "binlog",
        "主从复制",
        "审计"
      ],
      "question": "binlog 是 MySQL ____ 层的逻辑日志，主要用于主从复制、基于时间点恢复（PITR）和 ____；而 redo log 是 InnoDB 引擎层用于崩溃恢复的日志。",
      "answer": [
        "Server",
        "审计"
      ],
      "answer_rule": "any",
      "explanation": "binlog 由 MySQL 的 Server 层记录逻辑变更，其主要用途包括主从复制（从库重放 binlog）、基于时间点恢复（PITR）以及审计。而 redo log 属于 InnoDB 存储引擎层，只用于崩溃恢复时重做已提交事务。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-010",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "Buffer Pool",
        "脏页",
        "刷盘",
        "WAL"
      ],
      "question": "Buffer Pool 中被修改而未刷盘的数据页称为 ____ 页，InnoDB 会延迟到合适时机（如 CHECKPOINT）才批量 ____ 并将其落盘；由于 redo log 已先行持久化，即使这期间崩溃也能通过重放恢复。",
      "answer": [
        "脏",
        "刷盘"
      ],
      "answer_rule": "ordered",
      "explanation": "Buffer Pool 中已修改但尚未写回磁盘的页称为脏页。InnoDB 通常在检查点（checkpoint）或后台刷脏线程等时机批量刷盘，而不是每次修改都立即落盘。因为 redo log 已先持久化，脏页即便在刷盘前崩溃，也能通过重放 redo 恢复，这正是 WAL + 延迟刷盘配合的高效设计。",
      "source": null,
      "related": []
    }
  ]
}