Upgrade to Pro — share decks privately, control downloads, hide ads and more …

JSのウェブフレームワークで高速なルーターを実装する方法

 JSのウェブフレームワークで高速なルーターを実装する方法

https://nseg.connpass.com/event/251366/ での発表資料です。

Taku Amano

June 25, 2022
Tweet

More Decks by Taku Amano

Other Decks in Programming

Transcript

  1. https://github.com/honojs/hono/blob/master/README.md Hono - [炎] means flame in Japanese - is

    a small, simple, and ultrafast web framework for Cloudflare Workers or Service Worker based serverless such as Fastly Compute@Edge. 3 3
  2. Cloudflare Workers https://developers.cloudflare.com/workers/runtime- apis/web-standards/ Cloudflare Workers uses the V8 JavaScript

    engine from Google Chrome. The Workers runtime is updated at least once a week, to at least the version that is currently used by Chrome’s stable release. This means you can safely use the latest JavaScript features, with no need for transpilers. All of the standard built-in objects supported by the current Google Chrome stable release are supported, with a few notable exceptions: 6 6
  3. const app = new Hono(); app.get("/", function a() {}); app.get("/users/:id",

    function b() {}); app.get("/users/:id/posts", function c() {}); GET / => a GET /users/1 => b GET /users/1/posts => c 13 13
  4. const routes = [ [/^\/$/, a], [/^\/users\/[^/]+$/, b], [/^\/users\/[^/]+\/posts$/, c],

    ]; // ここまでを事前に用意しておく const path = request.url.replace(/https?:\/\/[^/]+/, ""); // 実行時にリクエストオブジェクトからパスを取り出す for (const [regexp, handler] of routes) { if (regexp.test(path)) { return handler; } } 15 15
  5. Perlでできること use re 'eval'; my @routes = ($a, $b, $c);

    my @c = (); my $m = undef; $path =~ m{ \A/(?:\z(?{$m=0}) |users/([^/]+)(?:\z(?{$m=1;@c=($1)}) |/posts\z(?{$m=2;@c=($1)}))) }x; 23 23
  6. JavaScriptでできること(全体) const routeData: [number, Array<[string, number]>][] = []; routeData[1] =

    [a, []]; routeData[3] = [b, [["id", 1]]]; routeData[4] = [c, [["id", 1]]]; const re = /^\/(?:$()|users\/([^\/]+)(?:$()|\/posts$()))/; // ここまでを事前に用意しておく const matchResult = path.match(re); if (matchResult) { const [handler, paramMap] = routeData[matchResult.indexOf("", 1)]; const pathParams = {} for (const [key, index] of paramMap) { pathParams[key] = matchResult[index]; } return [handler, pathParams]; } 25 25
  7. const prefix = Array.from({ length: 1000 }, () => "prefix").join("");

    const re = new RegExp(Array.from({ length: 26 }, (_, k) => `${prefix}${String.fromCharCode("a".charCodeAt(0) + k)}` ).join("|")); // `${prefix}a|${prefix}b|${prefix}c|...` suite .add("a", () => { re.test(`${prefix}a`); }) .add("z", () => { re.test(`${prefix}z`); }) .run(); a x 326,194 ops/sec ±0.61% (91 runs sampled) z x 334,725 ops/sec ±0.72% (92 runs sampled) 29 29
  8. const prefix = Array.from({ length: 1000 }, () => "prefix").join("");

    const re = new RegExp(Array.from({ length: 26 }, (_, k) => `with-(capture)${prefix}${String.fromCharCode("a".charCodeAt(0) + k)}` ).join("|")); // `with-(capture)${prefix}a|with-(capture)${prefix}b|...` suite .add("a", () => { re.test(`with-capture${prefix}a`); }) .add("z", () => { re.test(`with-capture${prefix}z`); }) .run(); a x 301,888 ops/sec ±0.54% (93 runs sampled) z x 272,048 ops/sec ±0.42% (96 runs sampled) 30 30
  9. const prefix = Array.from({ length: 1000 }, () => "prefix").join("");

    const re = new RegExp( `with-(capture)(?:${...})` ).join("|")); // `with-(capture)(?:${prefix}a|${prefix}b|${prefix}c|...` suite .add("a", () => { re.test(`with-capture${prefix}a`); }) .add("z", () => { re.test(`with-capture${prefix}z`); }) .run(); a x 310,702 ops/sec ±0.58% (93 runs sampled) z x 314,374 ops/sec ±0.92% (88 runs sampled) 31 31
  10. 最初に出したルーティング定義の例 const app = new Hono(); app.get("/", function a() {});

    app.get("/users/:id", function b() {}); app.get("/users/:id/posts", function c() {}); 35 35
  11. ミドルウェアを適用する例 const app = new Hono(); app.all("*", logger); app.all("/api/*", cors);

    app.get("/api/users/:id", function handler() {}); app.all("/api/*", fallback); 36 36
  12. const routes = [ [/^\//, logger], [/^\/api(?:$|\/)/, cors], [/^\/api\/users\/[^/]+$/, handler],

    [/^\/api(?:$|\/)/, fallback], ]; // ここまでを事前に用意しておく const path = request.url.replace(/https?:\/\/[^/]+/, ""); const handlers = []; for (const [regexp, handler] of routes) { if (regexp.test(path)) { handlers.push(handler); } } return handlers.reduceRight((prev, current) => { current(() => prev()) }, () => undefined) 37 37
  13. Honoでは、ここまでがルーターの仕事 const routes = [ [/^\//, logger], [/^\/api(?:$|\/)/, cors], [/^\/api\/users\/[^/]+$/,

    handler], [/^\/api(?:$|\/)/, fallback], ]; // ここまでを事前に用意しておく const path = request.url.replace(/https?:\/\/[^/]+/, ""); const handlers = []; for (const [regexp, handler] of routes) { if (regexp.test(path)) { handlers.push(handler); } } return handlers 42 42
  14. const app = new Hono(); app.all("*", logger); app.all("/api/*", cors); app.get("/api/users/:id",

    function handler() {}); app.all("/api/*", fallback); GET /api/users/1 => [logger, cors, handler, fallback] 44 44
  15. const app = new Hono(); app.all("*", logger); app.all("/api/*", cors); app.get("/api/users/:id",

    function handler() {}); app.all("/api/*", fallback); つまりここで探し出したいのは handler であり、 かつ handler が見つかったときには、常に [logger, cors, handler, fallback] が返されるはず。 48 48
  16. const routeData: [number, Array<[string, number]>][] = []; routeData[1] = [a,

    []]; routeData[3] = [b, [["id", 1]]]; routeData[4] = [c, [["id", 1]]]; const re = /^\/(?:$()|users\/([^\/]+)(?:$()|\/posts$()))/; // ここまでを事前に用意しておく const path = request.url.replace(/https?:\/\/[^/]+/, ""); const matchResult = path.match(re); if (matchResult) { const [handler, paramMap] = routeData[matchResult.indexOf("", 1)]; const pathParams = {} for (const [key, index] of paramMap) { pathParams[key] = matchResult[index]; } return [handler, pathParams]; } 49 49
  17. const a = [ logger, cors, a ]; const b

    = [ logger, basicAuth, b ]; const c = [ logger, basicAuth, c ]; 50 50
  18. const routeData: [number, Array<[string, number]>][] = []; routeData[1] = [

    a, []]; routeData[3] = [ b, [["id", 1]]]; routeData[4] = [ c, [["id", 1]]]; const re = /^\/(?:$()|users\/([^\/]+)(?:$()|\/posts$()))/; // ここまでを事前に用意しておく const matchResult = url.match(re); if (matchResult) { const [handler, paramMap] = routeData[matchResult.indexOf("", 1)]; const pathParams = {} for (const [key, index] of paramMap) { pathParams[key] = matchResult[index]; } return [handler, pathParams]; } 51 51
  19. の作り方 const app = new Hono(); app.all("*", logger); app.all("/api/*", cors);

    app.get("/api/users/:id", function handler() {}); app.all("/api/*", fallback); 52 52
  20. 登録された状態 [ // [ パス, 登録順, ハンドラ] ["*", 0, logger],

    ["/api/*", 1, cors], ["/api/users/:id", 2, handler], ["/api/*", 3, fallback], ] 53 53
  21. 以下のケースではハンドラ b に適用されるミドルウェアは 決まっているいない const app = new Hono(); app.get("/",

    function a() {}); app.get("/:type/subtype/:action", function a() {}); app.get("/users/:id/posts", function b() {}); GET /users/1/posts => [b] GET /users/subtype/create => [a, b] 58 58
  22. どれが速い? join / += indexOf / lastIndexOf split("/") / split(/\//)

    for / forEach / for-of (V8とSpiderMonkeyで違いがあります) 63 63
  23. join / += const strings = Array.from({ length: 1000 },

    () => "hello"); suite .add("join", () => { return strings.join(""); }) .add("+=", () => { let res = ""; strings.forEach((str) => (res += str)); return res; }) .add("reduce", () => { return strings.reduce((res, str) => res + str, ""); }) .run(); 64 64
  24. join x 48,489 ops/sec ±2.26% (97 runs sampled) += x

    105,353 ops/sec ±1.63% (97 runs sampled) reduce x 130,151 ops/sec ±3.11% (93 runs sampled) Fastest is join by reduce 65 65
  25. indexOf / lastIndexOf const re = new RegExp([...Array(1000)].map((_, i) =>

    `/${i}$()` ).join("|")); const m = "/500".match(re); suite .add("indexOf()", () => { return m.indexOf(""); }) .add("lastIndexOf()", () => { return m.lastIndexOf(""); }) .run(); 66 66
  26. indexOf() x 340,204,824 ops/sec ±0.69% (88 runs sampled) lastIndexOf() x

    545,046 ops/sec ±0.76% (94 runs sampled) Fastest is indexOf() indexOf()はC++ lastIndexOf()はTorque 67 67
  27. split("/") / split(/\//) const string = process.argv[process.argv.length - 1] ||

    "/"; suite .add(`"/"`, () => { return string.split("/"); }) .add("/\\//", () => { return string.split(/\//); }) .run(); 68 68
  28. "/" x 4,386,136 ops/sec ±0.27% (98 runs sampled) /\// x

    9,975,101 ops/sec ±0.24% (95 runs sampled) Fastest is /\// 69 69
  29. const splitter = {}; splitter[Symbol.split] = function (){ return ["42"];

    } console.log("test".split(splitter)); // => ["42"] 74 74
  30. V8のコメントの抜粋 // Redirect to splitter method if {separator[@@split]} is not

    undefined. RegExpの場合には(おそらく)ここで処理されて返る // String and integer conversions. // Shortcut for {limit} == 0. // If the separator string is empty then return the elements in the subject. ... 75 75
  31. function splitByIndexOf(str: string, sep: string): string[] { const res: string[]

    = []; for (let i = 0, j = str.indexOf(sep); ; j = str.indexOf(sep, i)) { if (j === -1) { res.push(str.substring(i)); break; } else { res.push(str.substring(i, j)); i = j + 1; } } return res; } ※ limitには未対応 77 77
  32. const string = process.argv[process.argv.length - 1] || "/"; suite .add(`"/"`,

    () => { return string.split("/"); }) .add("/\\//", () => { return string.split(/\//); }) .add("splitByIndexOf", () => { return splitByIndexOf(string, "/"); }) .run(); 78 78
  33. suite .add( `"/"`, () => string.split("/"), { onStart: () =>

    delete String.prototype[Symbol.split], } ) .add( "prototype", () => string.split("/"), { onStart: () => String.prototype[Symbol.split] = splitByIndexOf, } ) 81 81
  34. while x 13,837,421 ops/sec ±0.31% (97 runs sampled) for x

    13,904,145 ops/sec ±0.68% (92 runs sampled) for with len x 13,942,264 ops/sec ±1.15% (99 runs sampled) for with decrement x 7,646,750 ops/sec ±0.21% (96 runs sampled) forEach x 4,837,059 ops/sec ±0.08% (100 runs sampled) reduce x 7,477,822 ops/sec ±1.64% (98 runs sampled) for-of x 5,279,595 ops/sec ±0.20% (97 runs sampled) Fastest is for 86 86
  35. while x 5,340,699 ops/sec ±0.19% (98 runs sampled) for x

    5,312,270 ops/sec ±0.58% (98 runs sampled) for with len x 6,251,252 ops/sec ±0.55% (94 runs sampled) for with decrement x 5,374,107 ops/sec ±0.21% (100 runs sampled) forEach x 5,178,295 ops/sec ±1.84% (98 runs sampled) for-of x 4,522,059 ops/sec ±0.10% (97 runs sampled) Fastest is for with len 88 88