From 6a26c8b499f878770a6d9bfe8e906eac113753be Mon Sep 17 00:00:00 2001 From: wonder Date: Mon, 24 Aug 2026 03:14:45 +0000 Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=E6=89=80=E6=9C=89=20M?= =?UTF-8?q?ermaid=20=E5=9B=BE=E8=A1=A8=E8=AF=AD=E6=B3=95=E5=85=BC=E5=AE=B9?= =?UTF-8?q?=E6=80=A7=E9=97=AE=E9=A2=98?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- docs/algorithm/bloom-filter.md | 69 ++++++++--------- docs/algorithm/heavykeeper.md | 75 ++++++------------- .../cache-breakdown-avalanche-penetration.md | 66 +++++++--------- 3 files changed, 81 insertions(+), 129 deletions(-) diff --git a/docs/algorithm/bloom-filter.md b/docs/algorithm/bloom-filter.md index 7fd30eb..9bd025f 100644 --- a/docs/algorithm/bloom-filter.md +++ b/docs/algorithm/bloom-filter.md @@ -20,30 +20,28 @@ ```mermaid graph LR - subgraph 插入元素x["插入元素 x"] - X["x"] --> H1["h₁(x) = 3"] - X --> H2["h₂(x) = 7"] - X --> H3["h₃(x) = 11"] - H1 --> B3["bit[3] = 1"] - H2 --> B7["bit[7] = 1"] - H3 --> B11["bit[11] = 1"] - end + X["元素 x"] --> H1["h1(x) = 3"] + X --> H2["h2(x) = 7"] + X --> H3["h3(x) = 11"] + H1 --> B3["bit[3] = 1"] + H2 --> B7["bit[7] = 1"] + H3 --> B11["bit[11] = 1"] ``` ```mermaid graph LR - subgraph 查询y["查询元素 y — 一定不存在"] - Y["y"] --> YH1["h₁(y) = 3 → bit[3]=1 ✓"] - Y --> YH2["h₂(y) = 5 → bit[5]=0 ✗"] - YH1 --> YR["结论：一定不存在"] + subgraph qy["查询元素 y - 一定不存在"] + Y["y"] --> YH1["h1(y)=3 bit[3]=1"] + Y --> YH2["h2(y)=5 bit[5]=0"] + YH1 --> YR["结论: 一定不存在"] YH2 --> YR end - subgraph 查询z["查询元素 z — 假阳性"] - Z["z"] --> ZH1["h₁(z)=3 → bit[3]=1 ✓"] - Z --> ZH2["h₂(z)=7 → bit[7]=1 ✓"] - Z --> ZH3["h₃(z)=11 → bit[11]=1 ✓"] - ZH1 --> ZR["结论：可能存在
⚠️ 实际未插入（假阳性）"] + subgraph qz["查询元素 z - 假阳性"] + Z["z"] --> ZH1["h1(z)=3 bit[3]=1"] + Z --> ZH2["h2(z)=7 bit[7]=1"] + Z --> ZH3["h3(z)=11 bit[11]=1"] + ZH1 --> ZR["结论: 可能存在 / 实际未插入"] ZH2 --> ZR ZH3 --> ZR end @@ -51,15 +49,8 @@ graph LR ```mermaid graph LR - subgraph 位数组["位数组 (m=16)"] - direction LR - b0["0"] --- b1["0"] --- b2["0"] --- b3["1"] --- b4["0"] --- b5["0"] --- b6["0"] --- b7["1"] - b7 --- b8["0"] --- b9["0"] --- b10["0"] --- b11["1"] --- b12["0"] --- b13["0"] --- b14["0"] --- b15["0"] - end - - 插入元素x -.->|h₁ h₂ h₃| b3 - 插入元素x -.-> b7 - 插入元素x -.-> b11 + b0["0"] --- b1["0"] --- b2["0"] --- b3["1"] --- b4["0"] --- b5["0"] --- b6["0"] --- b7["1"] + b7 --- b8["0"] --- b9["0"] --- b10["0"] --- b11["1"] --- b12["0"] --- b13["0"] --- b14["0"] --- b15["0"] ``` ### 假阳性率 @@ -81,32 +72,32 @@ $$p \approx \left(1 - e^{-kn/m}\right)^k$$ ```mermaid graph TB - subgraph BF["布隆过滤器 — 极致省空间"] + subgraph BF["布隆过滤器 - 极致省空间"] B1["位数组 m bit"] --> B2["k 个哈希函数"] B2 --> B3["插入: O(k)"] B2 --> B4["查询: O(k)"] - B2 --> B5["删除: ❌"] + B2 --> B5["删除: 不支持"] B2 --> B6["假阳性: 有"] end - subgraph HS["HashSet — 精确但占空间"] + subgraph HS["HashSet - 精确但占空间"] HS1["存完整元素"] --> HS2["哈希表"] HS2 --> HS3["插入: O(1)"] HS2 --> HS4["查询: O(1)"] - HS2 --> HS5["删除: ✅"] + HS2 --> HS5["删除: 支持"] HS2 --> HS6["假阳性: 无"] end - subgraph CBF["Counting Bloom — 可删除"] + subgraph CBF["Counting Bloom - 可删除"] C1["计数器数组 4m bit"] --> C2["k 个哈希函数"] C2 --> C3["插入: O(k)"] C2 --> C4["查询: O(k)"] - C2 --> C5["删除: ✅"] + C2 --> C5["删除: 支持"] C2 --> C6["假阳性: 有"] end BF -.->|"约 4x 空间差"| CBF - HS -.->|"约 4~10x 空间差"| BF + HS -.->|"约 4-10x 空间差"| BF ``` | | 布隆过滤器 | HashSet | Counting Bloom Filter | @@ -207,16 +198,16 @@ sequenceDiagram participant DB as DB C->>BF: 查询 user:50 - BF-->>C: 一定不存在 ✗ - C-->>C: 直接返回，不查缓存/DB + BF-->>C: 一定不存在 + C-->>C: 直接返回 不查缓存和DB C->>BF: 查询 user:100 - BF-->>C: 可能存在 ✓ + BF-->>C: 可能存在 C->>R: GET user:100 alt 缓存命中 - R-->>C: 返回数据 ✅ - else 缓存 miss - C->>DB: SELECT * FROM users WHERE id=100 + R-->>C: 返回数据 + else 缓存miss + C->>DB: SELECT DB-->>C: 返回数据 C->>R: SET user:100 + TTL end diff --git a/docs/algorithm/heavykeeper.md b/docs/algorithm/heavykeeper.md index f7a430b..64fa859 100644 --- a/docs/algorithm/heavykeeper.md +++ b/docs/algorithm/heavykeeper.md @@ -19,24 +19,14 @@ ```mermaid graph LR - subgraph 数据流["高速数据流"] - F1["流 A: 10000次 🐘"] - F2["流 B: 8000次 🐘"] - F3["流 C: 1次 🐭"] - F4["流 D: 2次 🐭"] - F5["流 E: 1次 🐭"] - F6["流 ...: 1次 🐭"] - end + F1["流A: 10000次"] --> HK["HeavyKeeper"] + F2["流B: 8000次"] --> HK + F3["流C: 1次"] --> HK + F4["流D: 2次"] --> HK + F5["流E: 1次"] --> HK - F1 --> HK["HeavyKeeper"] - F2 --> HK - F3 --> HK - F4 --> HK - F5 --> HK - F6 --> HK - - HK --> TOP["Top-K 结果:
流 A: ~10000
流 B: ~8000"] - HK -.->|老鼠流被衰减淘汰| GONE["流 C~E: 归零 🗑️"] + HK --> TOP["Top-K: 流A~10000 流B~8000"] + HK -.->|"老鼠流衰减淘汰"| GONE["流C~E: 归零"] ``` ### 背景：为什么需要 HeavyKeeper @@ -53,26 +43,21 @@ graph LR ```mermaid graph TB Input["插入流 x"] --> Loop["遍历 d 行"] - - Loop --> Hash["pos = hᵢ(x) mod w"] + Loop --> Hash["pos = hi(x) mod w"] Hash --> Check{"bucket[pos]?"} - Check -->|"flowID == x"| Hit["count++"] Check -->|"flowID == 空"| Empty["写入 (x, 1)"] - Check -->|"flowID ≠ x"| Conflict{"random() < p?"} - + Check -->|"flowID != x"| Conflict{"random() < p?"} Conflict -->|"否"| Skip["跳过"] Conflict -->|"是"| Decay["count--"] - Decay --> Zero{"count == 0?"} Zero -->|"是"| Replace["替换为 (x, 1)"] - Zero -->|"否"| Skip2["保留原流"] - + Zero -->|"否"| Keep["保留原流"] Hit --> NextRow["下一行"] Empty --> NextRow Skip --> NextRow Replace --> NextRow - Skip2 --> NextRow + Keep --> NextRow NextRow --> Loop style Hit fill:#4caf50,color:#fff @@ -83,17 +68,6 @@ graph TB ```mermaid graph TB - subgraph 数据结构["HeavyKeeper 结构 (d=3, w=8)"] - direction TB - Row0["行 0: h₀(x)"] - Row1["行 1: h₁(x)"] - Row2["行 2: h₂(x)"] - - Row0 --- B00["流C:3"] --- B01["流A:10000"] --- B02[" "] --- B03["流D:1"] --- B04["流A:9500"] --- B05[" "] --- B06["流B:8000"] --- B07["流A:9800"] - Row1 --- B10["流B:7500"] --- B11[" "] --- B12["流A:10200"] --- B13["流E:1"] --- B14[" "] --- B15["流B:8200"] --- B16[" "] --- B17["流A:9000"] - Row2 --- B20[" "] --- B21["流A:10100"] --- B22["流C:2"] --- B23["流B:7800"] --- B24[" "] --- B25["流A:9900"] --- B26[" "] --- B27["流D:0🗑️"] - end - Query["查询流 A"] --> Max["取各行最大 count = 10200"] ``` @@ -126,20 +100,20 @@ Top-k 查询： ```mermaid graph LR - subgraph p0["p = 0 — 无衰减"] - P0E["老鼠流永不清除
噪声累积，Top-K 精度低"] + subgraph p0["p=0 无衰减"] + P0E["老鼠流永不清除 / 噪声累积 / Top-K精度低"] end - subgraph pok["p 适中 (0.01~0.1) — 最佳"] - POKE["老鼠流: count=1 → 一次衰减→0 🗑️
大象流: count=10000 → -1 无感 ✅
Top-K 精度高"] + subgraph pok["p适中 0.01-0.1 最佳"] + POKE["老鼠流count=1 一次衰减归零 / 大象流count=10000 -1无感 / Top-K精度高"] end - subgraph pbig["p 过大 (>0.3)"] - PBIGE["大象流也被严重衰减
估计偏低，Top-K 不准"] + subgraph pbig["p过大 大于0.3"] + PBIGE["大象流也被严重衰减 / 估计偏低 / Top-K不准"] end - p0 -->|"增大 p"| pok - pok -->|"继续增大 p"| pbig + p0 -->|"增大p"| pok + pok -->|"继续增大p"| pbig ``` | 衰减概率 p | 效果 | @@ -304,14 +278,9 @@ func sortDesc(items []Item) { ```mermaid graph LR - Log["Nginx Access Log
stdin 流式读取"] --> Parse["提取 Client IP"] - Parse --> HK["HeavyKeeper
4行 × 65536桶 × 5%衰减
≈ 2MB"] - HK --> TopK["Top-10 热门 IP"] - - subgraph 流量模型 - Elephant["192.168.1.100 🐘
10.0.0.5 🐘"] -->|"高频"| HK - Mouse["172.16.x.x 🐭
大量低频 IP"] -->|"衰减淘汰"| HK - end + Log["Nginx Access Log"] --> Parse["提取 Client IP"] + Parse --> HK["HeavyKeeper / 4行x65536桶 / 5%衰减 / 约2MB"] + HK --> TopK["Top-10 热门IP"] ``` ```go diff --git a/docs/architecture/cache/cache-breakdown-avalanche-penetration.md b/docs/architecture/cache/cache-breakdown-avalanche-penetration.md index 9ec6354..c49ee7c 100644 --- a/docs/architecture/cache/cache-breakdown-avalanche-penetration.md +++ b/docs/architecture/cache/cache-breakdown-avalanche-penetration.md @@ -17,22 +17,22 @@ ```mermaid graph TB - subgraph 击穿["🔓 缓存击穿 — 单热点 key 过期"] + subgraph bd["缓存击穿 - 单热点key过期"] B1["热 key 过期"] --> B2["大量并发同时 miss"] B2 --> B3["同时回源 DB"] - B3 --> B4["DB 瞬时尖峰 💥"] + B3 --> B4["DB 瞬时尖峰"] end - subgraph 雪崩["❄️ 缓存雪崩 — 大面积 key 集中失效"] - A1["大量 key 同时刻过期
或缓存节点宕机"] --> A2["请求大面积 miss"] + subgraph av["缓存雪崩 - 大面积key集中失效"] + A1["大量 key 同时刻过期 或 缓存节点宕机"] --> A2["请求大面积 miss"] A2 --> A3["全部涌向 DB"] - A3 --> A4["DB 持续高压 💥"] + A3 --> A4["DB 持续高压"] end - subgraph 穿透["🕳️ 缓存穿透 — 查不存在的数据"] + subgraph pn["缓存穿透 - 查不存在的数据"] C1["请求查询不存在的 key"] --> C2["缓存永远 miss"] C2 --> C3["每次直达 DB"] - C3 --> C4["DB 持续高压 💥"] + C3 --> C4["DB 持续高压"] end ``` @@ -51,18 +51,18 @@ graph TB ```mermaid graph LR - subgraph 击穿["击穿"] + subgraph bd["击穿"] A1["A1 永不过期"] - A2["A2 加锁排队
singleflight"] + A2["A2 加锁排队 / singleflight"] end - subgraph 雪崩["雪崩"] + subgraph av["雪崩"] B1["B1 加锁/限流"] B2["B2 随机失效"] - B3["B3 Redis 高可用
多级缓存"] + B3["B3 Redis高可用 / 多级缓存"] end - subgraph 穿透["穿透"] + subgraph pn["穿透"] C1["C1 参数校验"] C2["C2 缓存空对象"] C3["C3 布隆过滤器"] @@ -245,23 +245,23 @@ func (s *Service) GetWithRateLimit(ctx context.Context, key string) (any, error) ```mermaid graph LR - subgraph 无 Jitter["❌ 无随机偏移"] + subgraph no_jitter["无随机偏移"] T1["key_a TTL=30min"] --> E1["同时过期"] T2["key_b TTL=30min"] --> E1 T3["key_c TTL=30min"] --> E1 - E1 --> CRASH["DB 雪崩 💥"] + E1 --> CRASH["DB 雪崩"] end - subgraph 有 Jitter["✅ 有随机偏移"] + subgraph has_jitter["有随机偏移"] J1["key_a TTL=30m+2m"] --> D1["32min 过期"] J2["key_b TTL=30m+7m"] --> D2["37min 过期"] J3["key_c TTL=30m+4m"] --> D3["34min 过期"] - D1 --> SAFE["DB 压力分散 ✅"] + D1 --> SAFE["DB 压力分散"] D2 --> SAFE D3 --> SAFE end - 无 Jitter -.->|加 jitter| 有 Jitter + no_jitter -.->|"加 jitter"| has_jitter ``` ```go @@ -284,20 +284,12 @@ for _, item := range items { ```mermaid graph TB - subgraph 多级缓存架构 - Client["客户端请求"] - L1["L1 本地缓存
bigcache / ristretto"] - L2["L2 Redis
Sentinel / Cluster"] - L3["L3 DB"] - - Client --> L1 - L1 -->|miss| L2 - L2 -->|miss| L3 - L3 --> L2 - L2 --> L1 - - L1 -.->|Redis 不可用时
降级到本地| Client - end + Client["客户端请求"] --> L1["L1 本地缓存 / bigcache / ristretto"] + L1 -->|"miss"| L2["L2 Redis / Sentinel / Cluster"] + L2 -->|"miss"| L3["L3 DB"] + L3 --> L2 + L2 --> L1 + L1 -.->|"Redis不可用时降级"| Client ``` | 方案 | 说明 | @@ -391,12 +383,12 @@ func (s *Service) GetWithNullCache(ctx context.Context, key string) (any, error) ```mermaid graph LR Req["请求"] --> BF{"布隆过滤器"} - BF -->|"一定不存在 ✗"| Reject["直接拒绝
不查缓存/DB"] - BF -->|"可能存在 ✓"| Cache{"Redis 缓存"} - Cache -->|hit| Return["返回数据"] - Cache -->|miss| DB["查询 DB"] - DB -->|有数据| Cache - DB -->|无数据| Null["缓存空对象
短 TTL"] + BF -->|"一定不存在"| Reject["直接拒绝 / 不查缓存和DB"] + BF -->|"可能存在"| Cache{"Redis 缓存"} + Cache -->|"hit"| Return["返回数据"] + Cache -->|"miss"| DB["查询 DB"] + DB -->|"有数据"| Cache2["回写缓存"] + DB -->|"无数据"| Null["缓存空对象 / 短TTL"] ``` ```go