{
  "topic": "cache-read-write-strategy",
  "type": "fill_blank",
  "schema_version": "1.0.0",
  "generated": "09-2026-09-05T00:00:00+08:00",
  "questions": [
    {
      "id": "fb-001",
      "type": "fill_blank",
      "difficulty": 1,
      "tags": [
        "Cache-Aside",
        "写策略"
      ],
      "question": "Cache-Aside 策略的写流程是：先写______，再______缓存。",
      "answer": [
        "数据库",
        "删除（失效）"
      ],
      "answer_rule": "ordered",
      "explanation": "Cache-Aside 的标准写流程：先把数据库当作真相源写入，再把缓存中的该 key 删除使其失效，等下次读请求 miss 后再回填，从而避免缓存与数据库长时间不一致。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-002",
      "type": "fill_blank",
      "difficulty": 1,
      "tags": [
        "Cache-Aside",
        "Read-Through",
        "CacheLoader"
      ],
      "question": "Cache-Aside 的读流程是：先查缓存，若 miss 则查询______，并把结果______到缓存后返回。",
      "answer": [
        "数据库",
        "回填（写入）"
      ],
      "answer_rule": "ordered",
      "explanation": "读请求先查缓存，miss 时由应用查数据库拿到最新值并回填缓存，再返回给调用方。这样后续相同 key 的读可以直接命中缓存，减少数据库压力。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-003",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "Write-Through",
        "Write-Behind"
      ],
      "question": "Write-Through 策略写操作是______地把数据写入缓存和数据库；而 Write-Behind 是先写入缓存、数据库由后台______批量刷写。",
      "answer": [
        "同步",
        "异步"
      ],
      "answer_rule": "ordered",
      "explanation": "Write-Through 写缓存即同步写数据库，强一致但慢；Write-Behind 写入缓存后立即返回成功，数据库由后台攒批/定时异步刷写，吞吐高但有丢失与乱序风险。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-004",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "Write-Behind",
        "Write-Back",
        "数据安全"
      ],
      "question": "Write-Behind 策略在返回成功到数据落库之间若缓存进程宕机，会带来______风险；因此它只适合最终一致且可容忍______的数据，不适合交易、余额、库存。",
      "answer": [
        "丢失",
        "丢失"
      ],
      "answer_rule": "any",
      "explanation": "Write-Behind 先缓存后异步落库，若缓存宕机且尚未刷库则数据永久丢失。因此只适用于「最终一致 + 可容忍丢失」的数据，例如计数、会话、中间态；交易、余额等绝不可用。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-005",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "多级缓存",
        "L1",
        "L2"
      ],
      "question": "经典多级缓存结构是 L1 ______ 缓存 → L2 ______ 缓存 → DB。",
      "answer": [
        "本地（进程内）",
        "Redis（共享）"
      ],
      "answer_rule": "ordered",
      "explanation": "经典多级缓存为「L1 本地进程内缓存（如 Caffeine）→ L2 Redis（共享）→ DB」。本地命中最快零网络，Redis 次之毫秒级，DB 兜底保证最终一致。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-006",
      "type": "fill_blank",
      "difficulty": 4,
      "tags": [
        "多级缓存",
        "一致性"
      ],
      "question": "多级缓存架构最大的代价是其______：同一数据在多个缓存层、多个节点上有副本，更新后若任一副本残留旧值就会不一致。写路径需要在先写数据库后，逐层________。",
      "answer": [
        "一致性",
        "失效（删除/清理）"
      ],
      "answer_rule": "ordered",
      "explanation": "多级缓存中同一数据在 L1、L2 及多节点均有副本，跨层跨节点难以保证实时一致，所以一致性是最大代价。写路径以数据库为真相源，写后须逐层失效：删除 Redis 并广播清理各节点本地缓存。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-007",
      "type": "fill_blank",
      "difficulty": 4,
      "tags": [
        "双删",
        "延时",
        "竞态"
      ],
      "question": "双删方案是：更新数据库后先删一次缓存，延时约______毫秒后再删一次，用来清掉第一次删除之后刚被回填的旧值。",
      "answer": [
        "50"
      ],
      "answer_rule": "any",
      "explanation": "标准双删的延时通常在几十毫秒（教材常取约 50ms），目的是让可能已读取旧值并发起回填的并发读请求完成回填，然后用第二次删除把它清掉。延时过短则回填可能还没发生，过长则会放大读 miss 窗口。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-008",
      "type": "fill_blank",
      "difficulty": 4,
      "tags": [
        "版本号回填",
        "竞态"
      ],
      "question": "在解决「旧值被回填」竞态问题的方案里，最彻底的是______回填：缓存数据带版本号，回填前比较，旧版本直接丢弃。",
      "answer": [
        "版本号"
      ],
      "answer_rule": "any",
      "explanation": "版本号回填在缓存里同时存业务值和版本号，回填前比较新旧版本，旧版本直接丢弃，杜绝读线程把库里的旧值重新写回缓存。相比双删的「概率性清理」更彻底，是根治竞态窗口来源2的方案。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-009",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "CacheLoader",
        "refreshAfterWrite",
        "Read-Through"
      ],
      "question": "Caffeine 用 ________ 把 Cache-Aside 封装成 Read-Through，缓存 miss 时自动加载；用 ________ 在写后到点异步刷新，防止热点 key 被击穿。",
      "answer": [
        "CacheLoader",
        "refreshAfterWrite"
      ],
      "answer_rule": "ordered",
      "explanation": "CacheLoader 让 Caffeine 在 miss 时自动回调加载数据（Read-Through 的本地版）；refreshAfterWrite 在写入指定时长后异步刷新，避免过期瞬间大量请求同时穿透数据库造成击穿。",
      "source": null,
      "related": []
    },
    {
      "id": "fb-010",
      "type": "fill_blank",
      "difficulty": 3,
      "tags": [
        "Caffeine",
        "TinyLFU"
      ],
      "question": "Caffeine 的淘汰算法是 ________，它通过记录 key 的访问________来判定淘汰，比 LRU 更能保留高频 key。",
      "answer": [
        "TinyLFU",
        "频率"
      ],
      "answer_rule": "ordered",
      "explanation": "Caffeine 使用 TinyLFU 淘汰算法，维护计数器记录 key 的访问频率并带时间衰减与最近访问，可避免 LRU 被批量顺序扫描冲垮，更适合热点集中的真实负载。",
      "source": null,
      "related": []
    }
  ]
}