| 1 | // Command bakeoff measures one app and records the result in |
| 2 | // results/<slug>.json. Run it from the repo root. |
| 3 | // |
| 4 | // bakeoff <ttfb|wake|load|weight|code|build|tokens|behaviour|source|readability> --slug <s> [flags] |
| 5 | package main |
| 6 | |
| 7 | import ( |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "flag" |
| 11 | "fmt" |
| 12 | "io" |
| 13 | "os" |
| 14 | "os/exec" |
| 15 | "path/filepath" |
| 16 | "strconv" |
| 17 | "strings" |
| 18 | "time" |
| 19 | |
| 20 | b "bakeoff/harness" |
| 21 | "bakeoff/harness/report" |
| 22 | ) |
| 23 | |
| 24 | func main() { |
| 25 | if len(os.Args) < 2 { |
| 26 | usage() |
| 27 | } |
| 28 | cmd := os.Args[1] |
| 29 | fs := flag.NewFlagSet(cmd, flag.ExitOnError) |
| 30 | slug := fs.String("slug", "", "app slug, e.g. rust") |
| 31 | url := fs.String("url", "", "app base URL (default https://todo-<slug>.bfj.oncarlos.com)") |
| 32 | results := fs.String("results", "results", "results directory") |
| 33 | dir := fs.String("dir", "", "app directory (default apps/<slug>)") |
| 34 | runs := fs.Int("runs", 3, "build: warm and one-line-change runs") |
| 35 | coldRuns := fs.Int("cold-runs", 1, "build: cold runs") |
| 36 | dur := fs.Duration("duration", 10*time.Second, "load: length of each step") |
| 37 | rest := fs.Duration("rest", 10*time.Second, "load: pause between steps") |
| 38 | levels := fs.String("levels", "1,2,4,8,16,32,64,128,256,512,1024", "load: concurrency levels, doubling") |
| 39 | phase := fs.String("phase", "", "tokens: build|fix|security|review") |
| 40 | logs := fs.String("logs", "builders/logs", "tokens: log directory") |
| 41 | since := fs.String("since", "", "tokens: log whose cumulative totals this resumed phase starts from (default: the previous phase's)") |
| 42 | file := fs.String("file", "", "behaviour: Playwright JSON report") |
| 43 | var rec b.Record |
| 44 | fs.StringVar(&rec.Language, "language", "", "record: display name") |
| 45 | fs.StringVar(&rec.Status, "status", "", "record: LIVE or BLOCKED <reason>") |
| 46 | fs.StringVar(&rec.Version, "version", "", "record: deployed version") |
| 47 | fs.StringVar(&rec.Stack, "stack", "", "record: one-line stack summary") |
| 48 | fs.StringVar(&rec.Source, "source", "", "record: source URL") |
| 49 | fs.StringVar(&rec.Note, "note", "", "record: a note shown in the app's section (added once)") |
| 50 | recURL := fs.String("app-url", "", "record: live URL") |
| 51 | clearNotes := fs.Bool("clear-notes", false, "record: drop existing notes first") |
| 52 | clearWake := fs.Bool("clear", false, "wake: drop recorded samples") |
| 53 | preview := fs.Bool("preview", false, "report: tag strings for copy review") |
| 54 | replies := fs.String("replies", "", "readability: comma-separated files, one scorer reply each") |
| 55 | fs.Parse(os.Args[2:]) |
| 56 | if *slug == "" && cmd != "report" && cmd != "writeup" { |
| 57 | usage() |
| 58 | } |
| 59 | b.ResultsDir = *results |
| 60 | if *url == "" { |
| 61 | *url = b.URL(*slug) |
| 62 | } |
| 63 | if *dir == "" { |
| 64 | *dir = filepath.Join("apps", *slug) |
| 65 | } |
| 66 | var err error |
| 67 | switch cmd { |
| 68 | case "ttfb": |
| 69 | err = ttfb(*slug, *url) |
| 70 | case "wake": |
| 71 | if *clearWake { |
| 72 | err = b.Update(*slug, func(r *b.AppResult) { r.Wake = nil }) |
| 73 | break |
| 74 | } |
| 75 | err = wake(*slug, *url) |
| 76 | case "load": |
| 77 | err = load(*slug, *url, *levels, *dur, *rest) |
| 78 | case "weight": |
| 79 | err = weight(*slug, *url) |
| 80 | case "code": |
| 81 | err = code(*slug, *dir) |
| 82 | case "build": |
| 83 | err = build(*slug, *dir, *coldRuns, *runs) |
| 84 | case "tokens": |
| 85 | err = tokens(*slug, *logs, *phase, *since) |
| 86 | case "behaviour": |
| 87 | err = behaviour(*slug, *file) |
| 88 | case "writeup": |
| 89 | err = renderWriteup(*file) |
| 90 | case "report": |
| 91 | err = renderReport(*file, *preview) |
| 92 | case "findings": |
| 93 | err = findings(*slug, *file) |
| 94 | case "probe": |
| 95 | err = probe(*url, *levels, *dur) |
| 96 | case "memory": |
| 97 | err = memory(*slug, *dir, *dur) |
| 98 | case "record": |
| 99 | if rec.URL = *recURL; rec.URL == "" && strings.HasPrefix(rec.Status, "LIVE") { |
| 100 | rec.URL = *url |
| 101 | } |
| 102 | err = b.Update(*slug, func(r *b.AppResult) { |
| 103 | if *clearNotes { |
| 104 | r.Notes = nil |
| 105 | } |
| 106 | b.ApplyRecord(r, rec) |
| 107 | }) |
| 108 | case "source": |
| 109 | err = source(*dir) |
| 110 | case "readability": |
| 111 | err = readability(*slug, *replies) |
| 112 | default: |
| 113 | usage() |
| 114 | } |
| 115 | if err != nil { |
| 116 | fmt.Fprintf(os.Stderr, "bakeoff %s %s: %v\n", cmd, *slug, err) |
| 117 | os.Exit(1) |
| 118 | } |
| 119 | } |
| 120 | |
| 121 | func usage() { |
| 122 | fmt.Fprintln(os.Stderr, "usage: bakeoff <ttfb|wake|load|weight|code|build|tokens|behaviour|source|readability> --slug <s> [flags]") |
| 123 | os.Exit(2) |
| 124 | } |
| 125 | |
| 126 | func seeded(url string, n int) (*b.Session, error) { |
| 127 | s, err := b.NewSession(url) |
| 128 | if err != nil { |
| 129 | return nil, err |
| 130 | } |
| 131 | if n > 0 { |
| 132 | _, err = b.Seed(s, n) |
| 133 | } |
| 134 | return s, err |
| 135 | } |
| 136 | |
| 137 | func ttfb(slug, url string) error { |
| 138 | empty, err := seeded(url, 0) |
| 139 | if err != nil { |
| 140 | return err |
| 141 | } |
| 142 | e, err := b.MeasureTTFB(empty, "/", 3, 30) |
| 143 | if err != nil { |
| 144 | return err |
| 145 | } |
| 146 | full, err := seeded(url, 50) |
| 147 | if err != nil { |
| 148 | return err |
| 149 | } |
| 150 | f, err := b.MeasureTTFB(full, "/", 3, 30) |
| 151 | if err != nil { |
| 152 | return err |
| 153 | } |
| 154 | fmt.Printf("%s ttfb empty p50 %.1fms, 50 items p50 %.1fms\n", slug, e.P50, f.P50) |
| 155 | return b.Update(slug, func(r *b.AppResult) { r.TTFBEmpty, r.TTFB50 = e, f }) |
| 156 | } |
| 157 | |
| 158 | // wake takes one sample if the instance is asleep, and exits 3 if it is |
| 159 | // awake, so a polling loop can tell "not asleep yet" from failure. |
| 160 | func wake(slug, url string) error { |
| 161 | asleep, err := b.IsAsleep(b.AppName(slug), b.Host(slug)) |
| 162 | if err != nil { |
| 163 | return err |
| 164 | } |
| 165 | if !asleep { |
| 166 | fmt.Printf("%s is awake; no sample\n", slug) |
| 167 | os.Exit(3) |
| 168 | } |
| 169 | // A fresh visitor, without first fetching / (that would wake it). |
| 170 | s, err := b.SessionNoFetch(url) |
| 171 | if err != nil { |
| 172 | return err |
| 173 | } |
| 174 | ms, err := b.MeasureWake(s, "/") |
| 175 | if err != nil { |
| 176 | return err |
| 177 | } |
| 178 | fmt.Printf("%s wake %.1fms\n", slug, ms) |
| 179 | return b.Update(slug, func(r *b.AppResult) { r.Wake = append(r.Wake, ms) }) |
| 180 | } |
| 181 | |
| 182 | // load ramps each scenario through the levels, stopping early when the |
| 183 | // app has stopped scaling or is failing (b.SweepShouldStop). |
| 184 | func load(slug, url, levels string, d, rest time.Duration) error { |
| 185 | s, err := seeded(url, 50) |
| 186 | if err != nil { |
| 187 | return err |
| 188 | } |
| 189 | var out []b.LoadResult |
| 190 | for _, sc := range []string{"read", "toggle", "read-after"} { |
| 191 | var sweep []b.LoadResult |
| 192 | for _, l := range strings.Split(levels, ",") { |
| 193 | c, err := strconv.Atoi(strings.TrimSpace(l)) |
| 194 | if err != nil { |
| 195 | return err |
| 196 | } |
| 197 | if len(out)+len(sweep) > 0 { |
| 198 | time.Sleep(rest) |
| 199 | } |
| 200 | r, err := b.RunLoad(s, sc, c, d) |
| 201 | if err != nil { |
| 202 | // A step that can't run at all (oha failed, or the app is |
| 203 | // down) is a step where every request failed: record it and |
| 204 | // end this scenario, rather than losing the whole run. |
| 205 | fmt.Printf("%s %s c=%d: step failed: %v\n", slug, sc, c, err) |
| 206 | r = b.LoadResult{Scenario: sc, Concurrency: c, ErrorRate: 1, Errors: map[string]int{err.Error(): 1}} |
| 207 | sweep = append(sweep, r) |
| 208 | break |
| 209 | } |
| 210 | fmt.Printf("%s %s c=%d: %.0f rps, p50 %.1fms, p99 %.1fms, errors %.2f%%", slug, sc, c, r.RPS, r.P50Ms, r.P99Ms, r.ErrorRate*100) |
| 211 | if r.ErrorRate > 0 { |
| 212 | fmt.Printf(" status %v errors %v", r.Status, r.Errors) |
| 213 | } |
| 214 | fmt.Println() |
| 215 | sweep = append(sweep, r) |
| 216 | if b.SweepShouldStop(sweep) { |
| 217 | break |
| 218 | } |
| 219 | } |
| 220 | k := b.Knee(sweep) |
| 221 | fmt.Printf("%s %s: peak %.0f rps; slows down after %d (found %v, tested to %d)\n", slug, sc, k.PeakRPS, k.Concurrency, k.Found, k.MaxTested) |
| 222 | out = append(out, sweep...) |
| 223 | } |
| 224 | return b.Update(slug, func(r *b.AppResult) { r.Load = out }) |
| 225 | } |
| 226 | |
| 227 | func weight(slug, url string) error { |
| 228 | s, err := seeded(url, 50) |
| 229 | if err != nil { |
| 230 | return err |
| 231 | } |
| 232 | w, err := b.PageWeight(s, "/") |
| 233 | if err != nil { |
| 234 | return err |
| 235 | } |
| 236 | fmt.Printf("%s page weight (50 items): html %d bytes, js %d bytes\n", slug, w.HTMLBytes, w.JSBytes) |
| 237 | return b.Update(slug, func(r *b.AppResult) { r.Weight = w }) |
| 238 | } |
| 239 | |
| 240 | func code(slug, dir string) error { |
| 241 | c, err := b.CountCode(dir) |
| 242 | if err != nil { |
| 243 | return err |
| 244 | } |
| 245 | fmt.Printf("%s code: %+v\n", slug, c) |
| 246 | return b.Update(slug, func(r *b.AppResult) { r.Code = c }) |
| 247 | } |
| 248 | |
| 249 | func build(slug, dir string, coldRuns, runs int) error { |
| 250 | st, err := b.BuildStats(dir, strings.HasPrefix(slug, "go"), coldRuns, runs) |
| 251 | if err != nil { |
| 252 | return err |
| 253 | } |
| 254 | fmt.Printf("%s build: cold %.1fs, warm %.1fs, edit %.1fs, %d bytes on %s\n", slug, st.ColdSeconds, st.WarmSeconds, st.EditSeconds, st.Bytes, st.Host) |
| 255 | return b.Update(slug, func(r *b.AppResult) { r.Build = st }) |
| 256 | } |
| 257 | |
| 258 | func tokens(slug, logs, phase, since string) error { |
| 259 | parse := b.ParseClaudeLog |
| 260 | if phase == "review" { |
| 261 | parse = b.ParseCodexLog |
| 262 | } else if phase != "build" && phase != "fix" && phase != "security" { |
| 263 | return errors.New("--phase must be build, fix, security or review") |
| 264 | } |
| 265 | files, _ := filepath.Glob(filepath.Join(logs, slug+"."+phase+"*.jsonl")) |
| 266 | if len(files) == 0 { |
| 267 | return fmt.Errorf("no %s logs for %s in %s", phase, slug, logs) |
| 268 | } |
| 269 | var total b.Tokens |
| 270 | for _, f := range files { |
| 271 | fh, err := os.Open(f) |
| 272 | if err != nil { |
| 273 | return err |
| 274 | } |
| 275 | t, err := parse(fh) |
| 276 | fh.Close() |
| 277 | if err != nil { |
| 278 | return fmt.Errorf("%s: %w", f, err) |
| 279 | } |
| 280 | total = total.Add(t) |
| 281 | } |
| 282 | // Fix and security rounds resume the build session, whose result |
| 283 | // events report usage cumulatively: subtract the phase before. |
| 284 | if phase == "fix" || phase == "security" { |
| 285 | prev := []string{"build"} |
| 286 | if phase == "security" { |
| 287 | prev = []string{"fix", "build"} |
| 288 | } |
| 289 | if since != "" { |
| 290 | prev = []string{since} |
| 291 | } |
| 292 | for _, p := range prev { |
| 293 | f := filepath.Join(logs, slug+"."+p+".jsonl") |
| 294 | if since != "" { |
| 295 | f = since |
| 296 | } |
| 297 | fh, err := os.Open(f) |
| 298 | if err != nil { |
| 299 | continue |
| 300 | } |
| 301 | before, err := b.ParseClaudeLog(fh) |
| 302 | fh.Close() |
| 303 | if err != nil { |
| 304 | return fmt.Errorf("%s: %w", f, err) |
| 305 | } |
| 306 | total = total.Since(before) |
| 307 | break |
| 308 | } |
| 309 | } |
| 310 | fmt.Printf("%s %s tokens: %+v\n", slug, phase, total) |
| 311 | return b.Update(slug, func(r *b.AppResult) { |
| 312 | switch phase { |
| 313 | case "build": |
| 314 | r.TokensBuild = total |
| 315 | case "fix": |
| 316 | r.TokensFix = total |
| 317 | case "security": |
| 318 | r.TokensSecurity = total |
| 319 | case "review": |
| 320 | r.TokensReview = total |
| 321 | } |
| 322 | }) |
| 323 | } |
| 324 | |
| 325 | func behaviour(slug, file string) error { |
| 326 | var in io.Reader = os.Stdin |
| 327 | if file != "" { |
| 328 | f, err := os.Open(file) |
| 329 | if err != nil { |
| 330 | return err |
| 331 | } |
| 332 | defer f.Close() |
| 333 | in = f |
| 334 | } |
| 335 | got, err := b.ParsePlaywright(in) |
| 336 | if err != nil { |
| 337 | return err |
| 338 | } |
| 339 | fmt.Printf("%s behaviour: js-on %+v, js-off %+v\n", slug, got["js-on"], got["js-off"]) |
| 340 | return b.Update(slug, func(r *b.AppResult) { r.BehaviourJSOn, r.BehaviourJSOff = got["js-on"], got["js-off"] }) |
| 341 | } |
| 342 | |
| 343 | // source prints the app's own code, file by file, for the readability scorer. |
| 344 | func source(dir string) error { |
| 345 | files, err := b.SourceFiles(dir) |
| 346 | if err != nil { |
| 347 | return err |
| 348 | } |
| 349 | for _, f := range files { |
| 350 | body, err := os.ReadFile(filepath.Join(dir, f)) |
| 351 | if err != nil { |
| 352 | return err |
| 353 | } |
| 354 | fmt.Printf("===== %s =====\n%s\n", f, body) |
| 355 | } |
| 356 | return nil |
| 357 | } |
| 358 | |
| 359 | func readability(slug, replies string) error { |
| 360 | var texts []string |
| 361 | for _, f := range strings.Split(replies, ",") { |
| 362 | body, err := os.ReadFile(f) |
| 363 | if err != nil { |
| 364 | return err |
| 365 | } |
| 366 | texts = append(texts, string(body)) |
| 367 | } |
| 368 | r, err := b.MedianReadability(texts) |
| 369 | if err != nil { |
| 370 | return err |
| 371 | } |
| 372 | fmt.Printf("%s readability: overall %.0f/10 (naming %.0f, structure %.0f, traceability %.0f, errors %.0f, comments %.0f)\n", |
| 373 | slug, r.Overall, r.Naming, r.Structure, r.Traceability, r.Errors, r.Comments) |
| 374 | return b.Update(slug, func(a *b.AppResult) { a.Readability = r }) |
| 375 | } |
| 376 | |
| 377 | // renderReport writes the report page to stdout from every results file, |
| 378 | // in the contestants' fixed order, with prose from the copy file. |
| 379 | func renderReport(copyFile string, preview bool) error { |
| 380 | if copyFile == "" { |
| 381 | copyFile = "copy-review/strings.json" |
| 382 | } |
| 383 | text, err := os.ReadFile(copyFile) |
| 384 | if err != nil { |
| 385 | return err |
| 386 | } |
| 387 | words, err := report.LoadStrings(text) |
| 388 | if err != nil { |
| 389 | return err |
| 390 | } |
| 391 | var rs []b.AppResult |
| 392 | for _, a := range b.Apps { |
| 393 | r, err := b.LoadIn(b.ResultsDir, a.Slug) |
| 394 | if err != nil { |
| 395 | return err |
| 396 | } |
| 397 | if r.Language == "" { |
| 398 | r.Language = a.Language |
| 399 | } |
| 400 | if r.Status == "" { |
| 401 | r.Status = "not built yet" |
| 402 | } |
| 403 | rs = append(rs, r) |
| 404 | } |
| 405 | if preview { |
| 406 | return report.RenderPreview(rs, words, os.Stdout) |
| 407 | } |
| 408 | return report.Render(rs, words, os.Stdout) |
| 409 | } |
| 410 | |
| 411 | // findings records what a review left unfixed and exits 4 if any of it |
| 412 | // is a security finding (which earns the security-only round). |
| 413 | func findings(slug, file string) error { |
| 414 | text, err := os.ReadFile(file) |
| 415 | if err != nil { |
| 416 | return err |
| 417 | } |
| 418 | fs := b.ParseFindings(string(text)) |
| 419 | fmt.Printf("%s: %d findings remain\n", slug, len(fs)) |
| 420 | if err := b.Update(slug, func(r *b.AppResult) { r.ReviewFindings = fs }); err != nil { |
| 421 | return err |
| 422 | } |
| 423 | if b.HasSecurity(fs) { |
| 424 | fmt.Printf("%s: security findings remain\n", slug) |
| 425 | os.Exit(4) |
| 426 | } |
| 427 | return nil |
| 428 | } |
| 429 | |
| 430 | // probe runs one toggle step per level and prints oha's raw status and |
| 431 | // error breakdown, to diagnose failures a ramp reports. |
| 432 | func probe(url, levels string, d time.Duration) error { |
| 433 | s, err := seeded(url, 50) |
| 434 | if err != nil { |
| 435 | return err |
| 436 | } |
| 437 | ids, err := s.IDs("/") |
| 438 | if err != nil { |
| 439 | return err |
| 440 | } |
| 441 | for _, l := range strings.Split(levels, ",") { |
| 442 | out, err := exec.Command("oha", "--no-tui", "--output-format", "json", "-z", d.String(), "-c", strings.TrimSpace(l), |
| 443 | "--redirect", "0", "-H", "Cookie: "+s.CookieHeader(), "-H", "Origin: "+url, |
| 444 | "-m", "POST", "-T", "application/x-www-form-urlencoded", "-d", "filter=all", url+"/todos/"+ids[0]+"/toggle").Output() |
| 445 | if err != nil { |
| 446 | return err |
| 447 | } |
| 448 | var o struct { |
| 449 | Status map[string]int `json:"statusCodeDistribution"` |
| 450 | Errors map[string]int `json:"errorDistribution"` |
| 451 | } |
| 452 | json.Unmarshal(out, &o) |
| 453 | fmt.Printf("c=%s status %v errors %v\n", l, o.Status, o.Errors) |
| 454 | } |
| 455 | return nil |
| 456 | } |
| 457 | |
| 458 | // memory measures the app's own host build (dist/app, from make build) |
| 459 | // locally: resident memory at rest, peak under load, and after. |
| 460 | func memory(slug, dir string, d time.Duration) error { |
| 461 | bin, err := filepath.Abs(filepath.Join(dir, "dist", "app")) |
| 462 | if err != nil { |
| 463 | return err |
| 464 | } |
| 465 | m, err := b.MeasureMemory(bin, nil, d) |
| 466 | if err != nil { |
| 467 | return err |
| 468 | } |
| 469 | fmt.Printf("%s memory: idle %d KB, peak %d KB, after %d KB\n", slug, m.IdleKB, m.PeakKB, m.AfterKB) |
| 470 | return b.Update(slug, func(r *b.AppResult) { r.Memory = m }) |
| 471 | } |
| 472 | |
| 473 | // renderWriteup writes docs/writeup.md (or --file) as the write-up page. |
| 474 | func renderWriteup(file string) error { |
| 475 | if file == "" { |
| 476 | file = "docs/writeup.md" |
| 477 | } |
| 478 | src, err := os.ReadFile(file) |
| 479 | if err != nil { |
| 480 | return err |
| 481 | } |
| 482 | text, err := os.ReadFile("copy-review/strings.json") |
| 483 | if err != nil { |
| 484 | return err |
| 485 | } |
| 486 | words, err := report.LoadStrings(text) |
| 487 | if err != nil { |
| 488 | return err |
| 489 | } |
| 490 | return report.RenderWriteup(string(src), words, os.Stdout) |
| 491 | } |
| 492 | |
amadan