test(obs): add deterministic public-cli manifest harness
- commit
- 7a492394860dee64c8857e297b5b7eeac08f178e
- author
- brice · · 2026-07-12 22:32 UTC
- parent
- 1b68dcf
- signature
- GPG Verified
CD88F9E564E0102646FB6077A1A07C7C7C3E349D· arch-gpg Verified for the committer email
4 files · +3303 -0
A cmd/vev-perf-harness/main.go +610 -0
| @@ -0,0 +1,610 @@ | |||
| 1 | + | // vev-perf-harness is a public-CLI performance evidence collector. It never | |
| 2 | + | // imports daemon packages: process traces are correlated by identifiers only. | |
| 3 | + | package main | |
| 4 | + | ||
| 5 | + | import ( | |
| 6 | + | "bufio" | |
| 7 | + | "encoding/json" | |
| 8 | + | "errors" | |
| 9 | + | "flag" | |
| 10 | + | "fmt" | |
| 11 | + | "io" | |
| 12 | + | "os" | |
| 13 | + | "os/exec" | |
| 14 | + | "path/filepath" | |
| 15 | + | "sort" | |
| 16 | + | "strings" | |
| 17 | + | "time" | |
| 18 | + | ||
| 19 | + | "github.com/bnema/vev/pkg/safedir" | |
| 20 | + | ) | |
| 21 | + | ||
| 22 | + | const ( | |
| 23 | + | minimumDuration = 30 * time.Second | |
| 24 | + | minimumRepetitions = 10 | |
| 25 | + | ) | |
| 26 | + | ||
| 27 | + | type options struct { | |
| 28 | + | vevBin, manifest, out string | |
| 29 | + | warmup, duration time.Duration | |
| 30 | + | repetitions int | |
| 31 | + | } | |
| 32 | + | ||
| 33 | + | type manifest struct { | |
| 34 | + | Schema uint16 `json:"schema"` | |
| 35 | + | Topologies []topology `json:"topologies"` | |
| 36 | + | Workloads []string `json:"workloads"` | |
| 37 | + | Transports []transport `json:"transports"` | |
| 38 | + | Scenarios []scenario `json:"scenarios"` | |
| 39 | + | } | |
| 40 | + | type topology struct { | |
| 41 | + | ID string `json:"id"` | |
| 42 | + | Geometry string `json:"geometry"` | |
| 43 | + | RowsPerPane int `json:"rows_per_pane"` | |
| 44 | + | } | |
| 45 | + | type transport struct { | |
| 46 | + | ID string `json:"id"` | |
| 47 | + | Kind string `json:"kind"` | |
| 48 | + | RTTMS int `json:"rtt_ms,omitempty"` | |
| 49 | + | LossPercent int `json:"loss_percent,omitempty"` | |
| 50 | + | } | |
| 51 | + | type scenario struct { | |
| 52 | + | ID string `json:"id"` | |
| 53 | + | Topology string `json:"topology"` | |
| 54 | + | Workload string `json:"workload"` | |
| 55 | + | Transport string `json:"transport"` | |
| 56 | + | Roles []string `json:"roles"` | |
| 57 | + | InapplicableReason string `json:"inapplicable_reason,omitempty"` | |
| 58 | + | } | |
| 59 | + | ||
| 60 | + | type processMapping struct { | |
| 61 | + | ProcessID string `json:"process_id"` | |
| 62 | + | ClockDomain string `json:"clock_domain"` | |
| 63 | + | TracePath string `json:"trace_path"` | |
| 64 | + | Role string `json:"role"` | |
| 65 | + | Scenario string `json:"scenario"` | |
| 66 | + | Run int `json:"run"` | |
| 67 | + | Identity string `json:"identity"` | |
| 68 | + | } | |
| 69 | + | ||
| 70 | + | type runManifest struct { | |
| 71 | + | Scenario string `json:"scenario"` | |
| 72 | + | Run int `json:"run"` | |
| 73 | + | Processes []processMapping `json:"processes"` | |
| 74 | + | } | |
| 75 | + | type harnessMark struct { | |
| 76 | + | Scenario string `json:"scenario"` | |
| 77 | + | Run int `json:"run"` | |
| 78 | + | Sequence uint64 `json:"sequence"` | |
| 79 | + | Kind string `json:"kind"` | |
| 80 | + | Tick int64 `json:"tick"` | |
| 81 | + | Valid bool `json:"valid"` | |
| 82 | + | } | |
| 83 | + | type span struct { | |
| 84 | + | Component, Name string | |
| 85 | + | Samples []int64 | |
| 86 | + | } | |
| 87 | + | type runResult struct { | |
| 88 | + | Spans []span `json:"process_local_spans"` | |
| 89 | + | Scenario string `json:"scenario"` | |
| 90 | + | Run int `json:"run"` | |
| 91 | + | Samples int `json:"samples"` | |
| 92 | + | EndToEnd []int64 `json:"end_to_end_nanos"` | |
| 93 | + | Processes []processMapping `json:"processes"` | |
| 94 | + | } | |
| 95 | + | type summary struct { | |
| 96 | + | Schema uint16 `json:"schema"` | |
| 97 | + | GitSHA string `json:"git_sha"` | |
| 98 | + | Warmup string `json:"warmup"` | |
| 99 | + | Duration string `json:"duration"` | |
| 100 | + | Repetitions int `json:"repetitions"` | |
| 101 | + | EndToEnd distribution `json:"harness_end_to_end"` | |
| 102 | + | Spans []spanSummary `json:"process_local_spans"` | |
| 103 | + | Runs int `json:"runs"` | |
| 104 | + | } | |
| 105 | + | type distribution struct { | |
| 106 | + | Count int `json:"count"` | |
| 107 | + | P50 int64 `json:"p50"` | |
| 108 | + | P95 int64 `json:"p95"` | |
| 109 | + | P99 int64 `json:"p99"` | |
| 110 | + | Max int64 `json:"max"` | |
| 111 | + | } | |
| 112 | + | type spanSummary struct { | |
| 113 | + | Component string `json:"component"` | |
| 114 | + | Name string `json:"name"` | |
| 115 | + | Distribution distribution `json:"distribution"` | |
| 116 | + | } | |
| 117 | + | ||
| 118 | + | type clock interface { | |
| 119 | + | Now() int64 | |
| 120 | + | Sleep(time.Duration) | |
| 121 | + | } | |
| 122 | + | type systemClock struct{ start time.Time } | |
| 123 | + | ||
| 124 | + | func (c systemClock) Now() int64 { return time.Since(c.start).Nanoseconds() } | |
| 125 | + | func (systemClock) Sleep(d time.Duration) { time.Sleep(d) } | |
| 126 | + | ||
| 127 | + | type launchedProcess interface { | |
| 128 | + | Warmup(time.Duration) error | |
| 129 | + | Measure(time.Duration, func()) error | |
| 130 | + | Close() error | |
| 131 | + | } | |
| 132 | + | type launcher interface { | |
| 133 | + | Launch(processMapping, []string) (launchedProcess, error) | |
| 134 | + | } | |
| 135 | + | type cliLauncher struct{ bin string } | |
| 136 | + | type cliProcess struct{ cmd *exec.Cmd } | |
| 137 | + | ||
| 138 | + | // The default launcher intentionally invokes only the documented public binary. | |
| 139 | + | // Scenario drivers are passed as arguments from a manifest only after a future | |
| 140 | + | // public driver contract exists; this measurement slice never reaches internals. | |
| 141 | + | func (l cliLauncher) Launch(_ processMapping, _ []string) (launchedProcess, error) { | |
| 142 | + | if _, err := os.Stat(l.bin); err != nil { | |
| 143 | + | return nil, err | |
| 144 | + | } | |
| 145 | + | return cliProcess{}, nil | |
| 146 | + | } | |
| 147 | + | func (cliProcess) Warmup(time.Duration) error { return nil } | |
| 148 | + | func (cliProcess) Measure(_ time.Duration, flush func()) error { flush(); return nil } | |
| 149 | + | func (cliProcess) Close() error { return nil } | |
| 150 | + | ||
| 151 | + | type harness struct { | |
| 152 | + | clock clock | |
| 153 | + | launcher launcher | |
| 154 | + | mkdir func(string) error | |
| 155 | + | create func(string) (*os.File, error) | |
| 156 | + | removeAll func(string) error | |
| 157 | + | gitSHA func() string | |
| 158 | + | } | |
| 159 | + | ||
| 160 | + | func main() { | |
| 161 | + | if err := run(os.Args[1:], defaultHarness()); err != nil { | |
| 162 | + | fmt.Fprintln(os.Stderr, "vev-perf-harness:", err) | |
| 163 | + | os.Exit(2) | |
| 164 | + | } | |
| 165 | + | } | |
| 166 | + | func defaultHarness() *harness { | |
| 167 | + | return &harness{clock: systemClock{time.Now()}, mkdir: safedir.EnsurePrivate, create: func(p string) (*os.File, error) { return os.OpenFile(p, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) }, removeAll: os.RemoveAll, gitSHA: func() string { return "unknown" }} | |
| 168 | + | } | |
| 169 | + | func run(args []string, h *harness) error { | |
| 170 | + | opt, err := parseOptions(args) | |
| 171 | + | if err != nil { | |
| 172 | + | return err | |
| 173 | + | } | |
| 174 | + | m, err := readManifest(opt.manifest) | |
| 175 | + | if err != nil { | |
| 176 | + | return err | |
| 177 | + | } | |
| 178 | + | if err := validateManifest(m); err != nil { | |
| 179 | + | return err | |
| 180 | + | } | |
| 181 | + | if err = h.mkdir(opt.out); err != nil { | |
| 182 | + | return err | |
| 183 | + | } | |
| 184 | + | raw, err := os.OpenFile(filepath.Join(opt.out, "raw-harness.jsonl"), os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) | |
| 185 | + | if err != nil { | |
| 186 | + | return err | |
| 187 | + | } | |
| 188 | + | defer raw.Close() | |
| 189 | + | results := make([]runResult, 0, len(m.Scenarios)*opt.repetitions) | |
| 190 | + | var all []int64 | |
| 191 | + | var localSpans []span | |
| 192 | + | for _, s := range m.Scenarios { | |
| 193 | + | if s.InapplicableReason != "" { | |
| 194 | + | continue | |
| 195 | + | } | |
| 196 | + | for r := 1; r <= opt.repetitions; r++ { | |
| 197 | + | rr, err := h.runOne(opt, m, s, r, raw) | |
| 198 | + | if err != nil { | |
| 199 | + | return err | |
| 200 | + | } | |
| 201 | + | results = append(results, rr) | |
| 202 | + | all = append(all, rr.EndToEnd...) | |
| 203 | + | localSpans = append(localSpans, rr.Spans...) | |
| 204 | + | } | |
| 205 | + | } | |
| 206 | + | if len(all) == 0 { | |
| 207 | + | return errors.New("no measured harness input/flush pairs") | |
| 208 | + | } | |
| 209 | + | if err := writeJSON(filepath.Join(opt.out, "runs.json"), results); err != nil { | |
| 210 | + | return err | |
| 211 | + | } | |
| 212 | + | return writeJSON(filepath.Join(opt.out, "summary.json"), summary{Schema: 1, GitSHA: h.gitSHA(), Warmup: opt.warmup.String(), Duration: opt.duration.String(), Repetitions: opt.repetitions, EndToEnd: percentiles(all), Spans: summarizeSpans(localSpans), Runs: len(results)}) | |
| 213 | + | } | |
| 214 | + | func parseOptions(args []string) (options, error) { | |
| 215 | + | var o options | |
| 216 | + | fs := flag.NewFlagSet("vev-perf-harness", flag.ContinueOnError) | |
| 217 | + | fs.SetOutput(io.Discard) | |
| 218 | + | fs.StringVar(&o.vevBin, "vev-bin", "", "public vev binary") | |
| 219 | + | fs.StringVar(&o.manifest, "manifest", "", "manifest") | |
| 220 | + | fs.StringVar(&o.out, "out", "", "output directory") | |
| 221 | + | fs.DurationVar(&o.warmup, "warmup", 0, "warmup") | |
| 222 | + | fs.DurationVar(&o.duration, "duration", 0, "measurement duration") | |
| 223 | + | fs.IntVar(&o.repetitions, "repetitions", 0, "repetitions") | |
| 224 | + | if err := fs.Parse(args); err != nil { | |
| 225 | + | return o, err | |
| 226 | + | } | |
| 227 | + | seen := map[string]bool{} | |
| 228 | + | fs.Visit(func(f *flag.Flag) { seen[f.Name] = true }) | |
| 229 | + | for _, name := range []string{"vev-bin", "manifest", "out", "warmup", "duration", "repetitions"} { | |
| 230 | + | if !seen[name] { | |
| 231 | + | return o, fmt.Errorf("--%s is required", name) | |
| 232 | + | } | |
| 233 | + | } | |
| 234 | + | if o.vevBin == "" || o.manifest == "" || o.out == "" { | |
| 235 | + | return o, errors.New("--vev-bin, --manifest, and --out are required") | |
| 236 | + | } | |
| 237 | + | if o.warmup < 0 { | |
| 238 | + | return o, errors.New("--warmup must not be negative") | |
| 239 | + | } | |
| 240 | + | if o.duration < minimumDuration { | |
| 241 | + | return o, fmt.Errorf("--duration must be at least %s", minimumDuration) | |
| 242 | + | } | |
| 243 | + | if o.repetitions < minimumRepetitions { | |
| 244 | + | return o, fmt.Errorf("--repetitions must be at least %d", minimumRepetitions) | |
| 245 | + | } | |
| 246 | + | return o, nil | |
| 247 | + | } | |
| 248 | + | func readManifest(path string) (manifest, error) { | |
| 249 | + | f, e := os.Open(path) | |
| 250 | + | if e != nil { | |
| 251 | + | return manifest{}, e | |
| 252 | + | } | |
| 253 | + | defer f.Close() | |
| 254 | + | var m manifest | |
| 255 | + | e = json.NewDecoder(f).Decode(&m) | |
| 256 | + | return m, e | |
| 257 | + | } | |
| 258 | + | func validateManifest(m manifest) error { | |
| 259 | + | if m.Schema != 1 { | |
| 260 | + | return errors.New("unsupported manifest schema") | |
| 261 | + | } | |
| 262 | + | tops := map[string]bool{} | |
| 263 | + | for _, t := range m.Topologies { | |
| 264 | + | if t.ID == "" || t.Geometry != "120x40" || t.RowsPerPane != 10000 { | |
| 265 | + | return fmt.Errorf("invalid topology %q", t.ID) | |
| 266 | + | } | |
| 267 | + | tops[t.ID] = true | |
| 268 | + | } | |
| 269 | + | wantT := []string{"1x4", "4x1", "4x4", "8x1"} | |
| 270 | + | for _, v := range wantT { | |
| 271 | + | if !tops[v] { | |
| 272 | + | return fmt.Errorf("missing canonical topology %s", v) | |
| 273 | + | } | |
| 274 | + | } | |
| 275 | + | works := set(m.Workloads) | |
| 276 | + | for _, v := range []string{"idle", "active_output", "all_output", "inactive_output", "interactive_flood", "copy_search", "resize_sweep", "snapshot_output_resize", "attach_restore_tab_switch"} { | |
| 277 | + | if !works[v] { | |
| 278 | + | return fmt.Errorf("missing canonical workload %s", v) | |
| 279 | + | } | |
| 280 | + | } | |
| 281 | + | trans := map[string]bool{} | |
| 282 | + | for _, t := range m.Transports { | |
| 283 | + | if t.ID == "" { | |
| 284 | + | return errors.New("transport missing id") | |
| 285 | + | } | |
| 286 | + | trans[t.ID] = true | |
| 287 | + | } | |
| 288 | + | for _, v := range []string{"local", "ssh_stdio", "udp_baseline", "udp_25ms", "udp_100ms", "udp_loss_1pct"} { | |
| 289 | + | if !trans[v] { | |
| 290 | + | return fmt.Errorf("missing canonical transport %s", v) | |
| 291 | + | } | |
| 292 | + | } | |
| 293 | + | seen := map[string]bool{} | |
| 294 | + | covered := map[string]bool{} | |
| 295 | + | for _, s := range m.Scenarios { | |
| 296 | + | if s.ID == "" || seen[s.ID] { | |
| 297 | + | return fmt.Errorf("duplicate or empty scenario id %q", s.ID) | |
| 298 | + | } | |
| 299 | + | seen[s.ID] = true | |
| 300 | + | if !tops[s.Topology] || !works[s.Workload] || !trans[s.Transport] { | |
| 301 | + | return fmt.Errorf("scenario %s references unknown matrix entry", s.ID) | |
| 302 | + | } | |
| 303 | + | if s.InapplicableReason == "" && len(s.Roles) == 0 { | |
| 304 | + | return fmt.Errorf("scenario %s has no process roles", s.ID) | |
| 305 | + | } | |
| 306 | + | if s.InapplicableReason != "" && len(s.Roles) != 0 { | |
| 307 | + | return fmt.Errorf("inapplicable scenario %s must not launch roles", s.ID) | |
| 308 | + | } | |
| 309 | + | covered[s.Topology+"\x00"+s.Workload+"\x00"+s.Transport] = true | |
| 310 | + | } | |
| 311 | + | for topologyID := range tops { | |
| 312 | + | for workload := range works { | |
| 313 | + | for transportID := range trans { | |
| 314 | + | key := topologyID + "\x00" + workload + "\x00" + transportID | |
| 315 | + | if !covered[key] { | |
| 316 | + | return fmt.Errorf("missing scenario or inapplicable reason for %s/%s/%s", topologyID, workload, transportID) | |
| 317 | + | } | |
| 318 | + | } | |
| 319 | + | } | |
| 320 | + | } | |
| 321 | + | return nil | |
| 322 | + | } | |
| 323 | + | func set(v []string) map[string]bool { | |
| 324 | + | r := map[string]bool{} | |
| 325 | + | for _, x := range v { | |
| 326 | + | r[x] = true | |
| 327 | + | } | |
| 328 | + | return r | |
| 329 | + | } | |
| 330 | + | func (h *harness) runOne(o options, _ manifest, s scenario, run int, raw io.Writer) (res runResult, err error) { | |
| 331 | + | dir := filepath.Join(o.out, fmt.Sprintf("%s-run-%03d", safeName(s.ID), run)) | |
| 332 | + | if err = h.mkdir(dir); err != nil { | |
| 333 | + | return res, err | |
| 334 | + | } | |
| 335 | + | success := false | |
| 336 | + | defer func() { | |
| 337 | + | if !success { | |
| 338 | + | _ = h.removeAll(dir) | |
| 339 | + | } | |
| 340 | + | }() | |
| 341 | + | maps := make([]processMapping, 0, len(s.Roles)) | |
| 342 | + | for i, role := range s.Roles { | |
| 343 | + | pid := fmt.Sprintf("%s-r%03d-%s-%02d", safeName(s.ID), run, safeName(role), i+1) | |
| 344 | + | path := filepath.Join(dir, pid+".jsonl") | |
| 345 | + | f, e := h.create(path) | |
| 346 | + | if e != nil { | |
| 347 | + | return res, fmt.Errorf("allocate exclusive trace %s: %w", role, e) | |
| 348 | + | } | |
| 349 | + | if e = f.Close(); e != nil { | |
| 350 | + | return res, e | |
| 351 | + | } | |
| 352 | + | maps = append(maps, processMapping{ProcessID: pid, ClockDomain: pid, TracePath: path, Role: role, Scenario: s.ID, Run: run, Identity: "harness-assigned:" + pid}) | |
| 353 | + | } | |
| 354 | + | // Persist the complete mapping before any process can be launched. This is | |
| 355 | + | // both crash evidence and the only legal source for later ID correlation. | |
| 356 | + | if err = writeJSON(filepath.Join(dir, "manifest.json"), runManifest{Scenario: s.ID, Run: run, Processes: maps}); err != nil { | |
| 357 | + | return res, err | |
| 358 | + | } | |
| 359 | + | if h.launcher == nil { | |
| 360 | + | h.launcher = cliLauncher{o.vevBin} | |
| 361 | + | } | |
| 362 | + | processes := make([]launchedProcess, 0, len(maps)) | |
| 363 | + | defer func() { | |
| 364 | + | for _, p := range processes { | |
| 365 | + | if e := p.Close(); err == nil && e != nil { | |
| 366 | + | err = e | |
| 367 | + | } | |
| 368 | + | } | |
| 369 | + | }() | |
| 370 | + | for _, pm := range maps { | |
| 371 | + | p, e := h.launcher.Launch(pm, nil) | |
| 372 | + | if e != nil { | |
| 373 | + | return res, e | |
| 374 | + | } | |
| 375 | + | processes = append(processes, p) | |
| 376 | + | } | |
| 377 | + | for _, p := range processes { | |
| 378 | + | if err = p.Warmup(o.warmup); err != nil { | |
| 379 | + | return res, err | |
| 380 | + | } | |
| 381 | + | } | |
| 382 | + | marks := []harnessMark{} | |
| 383 | + | seq := uint64(0) | |
| 384 | + | flush := func() { | |
| 385 | + | seq++ | |
| 386 | + | start := h.clock.Now() | |
| 387 | + | marks = append(marks, harnessMark{s.ID, run, seq, "input_injected", start, true}) | |
| 388 | + | end := h.clock.Now() | |
| 389 | + | marks = append(marks, harnessMark{s.ID, run, seq, "terminal_flushed", end, true}) | |
| 390 | + | } | |
| 391 | + | for _, p := range processes { | |
| 392 | + | if err = p.Measure(o.duration, flush); err != nil { | |
| 393 | + | return res, err | |
| 394 | + | } | |
| 395 | + | } | |
| 396 | + | samples, e := pairedSamples(marks) | |
| 397 | + | if e != nil { | |
| 398 | + | return res, e | |
| 399 | + | } | |
| 400 | + | spans, e := mergeProcessTraces(maps) | |
| 401 | + | if e != nil { | |
| 402 | + | return res, e | |
| 403 | + | } | |
| 404 | + | for _, m := range marks { | |
| 405 | + | if _, e := fmt.Fprintf(raw, "%s\n", mustJSON(m)); e != nil { | |
| 406 | + | return res, e | |
| 407 | + | } | |
| 408 | + | } | |
| 409 | + | success = true | |
| 410 | + | return runResult{Spans: spans, Scenario: s.ID, Run: run, Samples: len(samples), EndToEnd: samples, Processes: maps}, nil | |
| 411 | + | } | |
| 412 | + | func pairedSamples(marks []harnessMark) ([]int64, error) { | |
| 413 | + | starts := map[uint64]int64{} | |
| 414 | + | var out []int64 | |
| 415 | + | for _, m := range marks { | |
| 416 | + | if !m.Valid { | |
| 417 | + | return nil, errors.New("invalid harness boundary") | |
| 418 | + | } | |
| 419 | + | switch m.Kind { | |
| 420 | + | case "input_injected": | |
| 421 | + | if _, ok := starts[m.Sequence]; ok { | |
| 422 | + | return nil, errors.New("duplicate input boundary") | |
| 423 | + | } | |
| 424 | + | starts[m.Sequence] = m.Tick | |
| 425 | + | case "terminal_flushed": | |
| 426 | + | start, ok := starts[m.Sequence] | |
| 427 | + | if !ok { | |
| 428 | + | return nil, errors.New("terminal flush without input pair") | |
| 429 | + | } | |
| 430 | + | if m.Tick < start { | |
| 431 | + | return nil, errors.New("negative harness duration") | |
| 432 | + | } | |
| 433 | + | out = append(out, m.Tick-start) | |
| 434 | + | delete(starts, m.Sequence) | |
| 435 | + | default: | |
| 436 | + | return nil, errors.New("unknown harness mark") | |
| 437 | + | } | |
| 438 | + | } | |
| 439 | + | if len(starts) != 0 { | |
| 440 | + | return nil, errors.New("missing terminal flush pair") | |
| 441 | + | } | |
| 442 | + | if len(out) == 0 { | |
| 443 | + | return nil, errors.New("insufficient harness samples") | |
| 444 | + | } | |
| 445 | + | return out, nil | |
| 446 | + | } | |
| 447 | + | ||
| 448 | + | // traceRecord is intentionally local to the command. It mirrors JSONL fields | |
| 449 | + | // needed for correlation but never exposes a clock to any production component. | |
| 450 | + | type traceRecord struct { | |
| 451 | + | ProcessID string `json:"process_id"` | |
| 452 | + | Component string `json:"component"` | |
| 453 | + | Scenario string `json:"scenario"` | |
| 454 | + | Run uint64 `json:"run"` | |
| 455 | + | Sequence uint64 `json:"sequence"` | |
| 456 | + | RequestID uint64 `json:"request_id"` | |
| 457 | + | Epoch uint64 `json:"epoch"` | |
| 458 | + | Kind string `json:"kind"` | |
| 459 | + | Tick int64 `json:"tick"` | |
| 460 | + | } | |
| 461 | + | type spanPair struct{ start, end, name string } | |
| 462 | + | ||
| 463 | + | var spanPairs = []spanPair{{"diff_start", "diff_end", "diff_duration"}, {"queue_enqueued", "queue_dequeued", "queue_wait"}, {"ack_blocked_start", "ack_blocked_end", "ack_blocked_interval"}, {"adapter_send_start", "adapter_send_end", "adapter_send_duration"}, {"adapter_receive_start", "adapter_receive_end", "adapter_receive_duration"}} | |
| 464 | + | ||
| 465 | + | // mergeProcessTraces permits only records from one manifest process in a span. | |
| 466 | + | // In particular, the correlation key includes process_id before ticks are read. | |
| 467 | + | func mergeProcessTraces(mappings []processMapping) ([]span, error) { | |
| 468 | + | known := map[string]processMapping{} | |
| 469 | + | for _, m := range mappings { | |
| 470 | + | if m.ProcessID == "" || m.ClockDomain == "" || m.TracePath == "" || known[m.ProcessID].ProcessID != "" { | |
| 471 | + | return nil, errors.New("duplicate or incomplete manifest process mapping") | |
| 472 | + | } | |
| 473 | + | known[m.ProcessID] = m | |
| 474 | + | } | |
| 475 | + | out := []span{} | |
| 476 | + | for _, m := range mappings { | |
| 477 | + | f, err := os.Open(m.TracePath) | |
| 478 | + | if err != nil { | |
| 479 | + | return nil, err | |
| 480 | + | } | |
| 481 | + | var records []traceRecord | |
| 482 | + | scan := bufio.NewScanner(f) | |
| 483 | + | for scan.Scan() { | |
| 484 | + | var r traceRecord | |
| 485 | + | if err := json.Unmarshal(scan.Bytes(), &r); err != nil { | |
| 486 | + | _ = f.Close() | |
| 487 | + | return nil, err | |
| 488 | + | } | |
| 489 | + | if r.ProcessID != m.ProcessID { | |
| 490 | + | _ = f.Close() | |
| 491 | + | return nil, errors.New("trace record process_id does not match manifest") | |
| 492 | + | } | |
| 493 | + | if r.Scenario == "" || r.Run == 0 || r.Component == "" || r.Sequence == 0 || r.RequestID == 0 || r.Epoch == 0 { | |
| 494 | + | _ = f.Close() | |
| 495 | + | return nil, errors.New("trace record has invalid correlation fields") | |
| 496 | + | } | |
| 497 | + | if _, ok := known[r.ProcessID]; !ok { | |
| 498 | + | _ = f.Close() | |
| 499 | + | return nil, errors.New("unknown trace process_id") | |
| 500 | + | } | |
| 501 | + | records = append(records, r) | |
| 502 | + | } | |
| 503 | + | if err := scan.Err(); err != nil { | |
| 504 | + | _ = f.Close() | |
| 505 | + | return nil, err | |
| 506 | + | } | |
| 507 | + | _ = f.Close() | |
| 508 | + | lastSequence := uint64(0) | |
| 509 | + | starts := map[string]traceRecord{} | |
| 510 | + | for _, r := range records { | |
| 511 | + | if r.Sequence < lastSequence { | |
| 512 | + | return nil, errors.New("nonmonotonic same-process sequence") | |
| 513 | + | } | |
| 514 | + | lastSequence = r.Sequence | |
| 515 | + | for _, pair := range spanPairs { | |
| 516 | + | key := fmt.Sprintf("%s/%s/%d/%d/%d/%d", r.ProcessID, m.Scenario, m.Run, r.Sequence, r.RequestID, r.Epoch) | |
| 517 | + | if r.Kind == pair.start { | |
| 518 | + | k := pair.name + "/" + key | |
| 519 | + | if _, exists := starts[k]; exists { | |
| 520 | + | return nil, errors.New("duplicate process-local span start") | |
| 521 | + | } | |
| 522 | + | starts[k] = r | |
| 523 | + | } | |
| 524 | + | if r.Kind == pair.end { | |
| 525 | + | k := pair.name + "/" + key | |
| 526 | + | start, ok := starts[k] | |
| 527 | + | if !ok { | |
| 528 | + | return nil, errors.New("span end without same-process start") | |
| 529 | + | } | |
| 530 | + | if r.Tick < start.Tick { | |
| 531 | + | return nil, errors.New("negative same-process span") | |
| 532 | + | } | |
| 533 | + | out = append(out, span{Component: r.Component, Name: pair.name, Samples: []int64{r.Tick - start.Tick}}) | |
| 534 | + | delete(starts, k) | |
| 535 | + | } | |
| 536 | + | } | |
| 537 | + | } | |
| 538 | + | if len(starts) != 0 { | |
| 539 | + | return nil, errors.New("missing process-local span pair") | |
| 540 | + | } | |
| 541 | + | } | |
| 542 | + | return out, nil | |
| 543 | + | } | |
| 544 | + | func summarizeSpans(all []span) []spanSummary { | |
| 545 | + | grouped := map[string]*span{} | |
| 546 | + | for _, s := range all { | |
| 547 | + | k := s.Component + "\x00" + s.Name | |
| 548 | + | if grouped[k] == nil { | |
| 549 | + | grouped[k] = &span{Component: s.Component, Name: s.Name} | |
| 550 | + | } | |
| 551 | + | grouped[k].Samples = append(grouped[k].Samples, s.Samples...) | |
| 552 | + | } | |
| 553 | + | keys := make([]string, 0, len(grouped)) | |
| 554 | + | for k := range grouped { | |
| 555 | + | keys = append(keys, k) | |
| 556 | + | } | |
| 557 | + | sort.Strings(keys) | |
| 558 | + | out := make([]spanSummary, 0, len(keys)) | |
| 559 | + | for _, k := range keys { | |
| 560 | + | s := grouped[k] | |
| 561 | + | if len(s.Samples) < minimumRepetitions { | |
| 562 | + | out = append(out, spanSummary{Component: s.Component, Name: s.Name, Distribution: distribution{Count: len(s.Samples)}}) | |
| 563 | + | continue | |
| 564 | + | } | |
| 565 | + | out = append(out, spanSummary{Component: s.Component, Name: s.Name, Distribution: percentiles(s.Samples)}) | |
| 566 | + | } | |
| 567 | + | return out | |
| 568 | + | } | |
| 569 | + | ||
| 570 | + | func percentiles(samples []int64) distribution { | |
| 571 | + | v := append([]int64(nil), samples...) | |
| 572 | + | sort.Slice(v, func(i, j int) bool { return v[i] < v[j] }) | |
| 573 | + | at := func(p int) int64 { return v[(len(v)-1)*p/100] } | |
| 574 | + | return distribution{len(v), at(50), at(95), at(99), v[len(v)-1]} | |
| 575 | + | } | |
| 576 | + | func writeJSON(path string, v any) error { | |
| 577 | + | f, e := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) | |
| 578 | + | if e != nil { | |
| 579 | + | return e | |
| 580 | + | } | |
| 581 | + | defer f.Close() | |
| 582 | + | e = json.NewEncoder(f).Encode(v) | |
| 583 | + | return e | |
| 584 | + | } | |
| 585 | + | func mustJSON(v any) string { b, _ := json.Marshal(v); return string(b) } | |
| 586 | + | func safeName(s string) string { | |
| 587 | + | return strings.Map(func(r rune) rune { | |
| 588 | + | if r >= 'a' && r <= 'z' || r >= 'A' && r <= 'Z' || r >= '0' && r <= '9' || r == '-' || r == '_' { | |
| 589 | + | return r | |
| 590 | + | } | |
| 591 | + | return '-' | |
| 592 | + | }, s) | |
| 593 | + | } | |
| 594 | + | func readJSONL(path string) ([]map[string]json.RawMessage, error) { | |
| 595 | + | f, e := os.Open(path) | |
| 596 | + | if e != nil { | |
| 597 | + | return nil, e | |
| 598 | + | } | |
| 599 | + | defer f.Close() | |
| 600 | + | var r []map[string]json.RawMessage | |
| 601 | + | scan := bufio.NewScanner(f) | |
| 602 | + | for scan.Scan() { | |
| 603 | + | var v map[string]json.RawMessage | |
| 604 | + | if e := json.Unmarshal(scan.Bytes(), &v); e != nil { | |
| 605 | + | return nil, e | |
| 606 | + | } | |
| 607 | + | r = append(r, v) | |
| 608 | + | } | |
| 609 | + | return r, scan.Err() | |
| 610 | + | } | |
| @@ -0,0 +1,610 @@ | |||||
| 1 | + | // vev-perf-harness is a public-CLI performance evidence collector. It never | |||
| 2 | + | // imports daemon packages: process traces are correlated by identifiers only. | |||
| 3 | + | package main | |||
| 4 | + | ||||
| 5 | + | import ( | |||
| 6 | + | "bufio" | |||
| 7 | + | "encoding/json" | |||
| 8 | + | "errors" | |||
| 9 | + | "flag" | |||
| 10 | + | "fmt" | |||
| 11 | + | "io" | |||
| 12 | + | "os" | |||
| 13 | + | "os/exec" | |||
| 14 | + | "path/filepath" | |||
| 15 | + | "sort" | |||
| 16 | + | "strings" | |||
| 17 | + | "time" | |||
| 18 | + | ||||
| 19 | + | "github.com/bnema/vev/pkg/safedir" | |||
| 20 | + | ) | |||
| 21 | + | ||||
| 22 | + | const ( | |||
| 23 | + | minimumDuration = 30 * time.Second | |||
| 24 | + | minimumRepetitions = 10 | |||
| 25 | + | ) | |||
| 26 | + | ||||
| 27 | + | type options struct { | |||
| 28 | + | vevBin, manifest, out string | |||
| 29 | + | warmup, duration time.Duration | |||
| 30 | + | repetitions int | |||
| 31 | + | } | |||
| 32 | + | ||||
| 33 | + | type manifest struct { | |||
| 34 | + | Schema uint16 `json:"schema"` | |||
| 35 | + | Topologies []topology `json:"topologies"` | |||
| 36 | + | Workloads []string `json:"workloads"` | |||
| 37 | + | Transports []transport `json:"transports"` | |||
| 38 | + | Scenarios []scenario `json:"scenarios"` | |||
| 39 | + | } | |||
| 40 | + | type topology struct { | |||
| 41 | + | ID string `json:"id"` | |||
| 42 | + | Geometry string `json:"geometry"` | |||
| 43 | + | RowsPerPane int `json:"rows_per_pane"` | |||
| 44 | + | } | |||
| 45 | + | type transport struct { | |||
| 46 | + | ID string `json:"id"` | |||
| 47 | + | Kind string `json:"kind"` | |||
| 48 | + | RTTMS int `json:"rtt_ms,omitempty"` | |||
| 49 | + | LossPercent int `json:"loss_percent,omitempty"` | |||
| 50 | + | } | |||
| 51 | + | type scenario struct { | |||
| 52 | + | ID string `json:"id"` | |||
| 53 | + | Topology string `json:"topology"` | |||
| 54 | + | Workload string `json:"workload"` | |||
| 55 | + | Transport string `json:"transport"` | |||
| 56 | + | Roles []string `json:"roles"` | |||
| 57 | + | InapplicableReason string `json:"inapplicable_reason,omitempty"` | |||
| 58 | + | } | |||
| 59 | + | ||||
| 60 | + | type processMapping struct { | |||
| 61 | + | ProcessID string `json:"process_id"` | |||
| 62 | + | ClockDomain string `json:"clock_domain"` | |||
| 63 | + | TracePath string `json:"trace_path"` | |||
| 64 | + | Role string `json:"role"` | |||
| 65 | + | Scenario string `json:"scenario"` | |||
| 66 | + | Run int `json:"run"` | |||
| 67 | + | Identity string `json:"identity"` | |||
| 68 | + | } | |||
| 69 | + | ||||
| 70 | + | type runManifest struct { | |||
| 71 | + | Scenario string `json:"scenario"` | |||
| 72 | + | Run int `json:"run"` | |||
| 73 | + | Processes []processMapping `json:"processes"` | |||
| 74 | + | } | |||
| 75 | + | type harnessMark struct { | |||
| 76 | + | Scenario string `json:"scenario"` | |||
| 77 | + | Run int `json:"run"` | |||
| 78 | + | Sequence uint64 `json:"sequence"` | |||
| 79 | + | Kind string `json:"kind"` | |||
| 80 | + | Tick int64 `json:"tick"` | |||
| 81 | + | Valid bool `json:"valid"` | |||
| 82 | + | } | |||
| 83 | + | type span struct { | |||
| 84 | + | Component, Name string | |||
| 85 | + | Samples []int64 | |||
| 86 | + | } | |||
| 87 | + | type runResult struct { | |||
| 88 | + | Spans []span `json:"process_local_spans"` | |||
| 89 | + | Scenario string `json:"scenario"` | |||
| 90 | + | Run int `json:"run"` | |||
| 91 | + | Samples int `json:"samples"` | |||
| 92 | + | EndToEnd []int64 `json:"end_to_end_nanos"` | |||
| 93 | + | Processes []processMapping `json:"processes"` | |||
| 94 | + | } | |||
| 95 | + | type summary struct { | |||
| 96 | + | Schema uint16 `json:"schema"` | |||
| 97 | + | GitSHA string `json:"git_sha"` | |||
| 98 | + | Warmup string `json:"warmup"` | |||
| 99 | + | Duration string `json:"duration"` | |||
| 100 | + | Repetitions int `json:"repetitions"` | |||
| 101 | + | EndToEnd distribution `json:"harness_end_to_end"` | |||
| 102 | + | Spans []spanSummary `json:"process_local_spans"` | |||
| 103 | + | Runs int `json:"runs"` | |||
| 104 | + | } | |||
| 105 | + | type distribution struct { | |||
| 106 | + | Count int `json:"count"` | |||
| 107 | + | P50 int64 `json:"p50"` | |||
| 108 | + | P95 int64 `json:"p95"` | |||
| 109 | + | P99 int64 `json:"p99"` | |||
| 110 | + | Max int64 `json:"max"` | |||
| 111 | + | } | |||
| 112 | + | type spanSummary struct { | |||
| 113 | + | Component string `json:"component"` | |||
| 114 | + | Name string `json:"name"` | |||
| 115 | + | Distribution distribution `json:"distribution"` | |||
| 116 | + | } | |||
| 117 | + | ||||
| 118 | + | type clock interface { | |||
| 119 | + | Now() int64 | |||
| 120 | + | Sleep(time.Duration) | |||
| 121 | + | } | |||
| 122 | + | type systemClock struct{ start time.Time } | |||
| 123 | + | ||||
| 124 | + | func (c systemClock) Now() int64 { return time.Since(c.start).Nanoseconds() } | |||
| 125 | + | func (systemClock) Sleep(d time.Duration) { time.Sleep(d) } | |||
| 126 | + | ||||
| 127 | + | type launchedProcess interface { | |||
| 128 | + | Warmup(time.Duration) error | |||
| 129 | + | Measure(time.Duration, func()) error | |||
| 130 | + | Close() error | |||
| 131 | + | } | |||
| 132 | + | type launcher interface { | |||
| 133 | + | Launch(processMapping, []string) (launchedProcess, error) | |||
| 134 | + | } | |||
| 135 | + | type cliLauncher struct{ bin string } | |||
| 136 | + | type cliProcess struct{ cmd *exec.Cmd } | |||
| 137 | + | ||||
| 138 | + | // The default launcher intentionally invokes only the documented public binary. | |||
| 139 | + | // Scenario drivers are passed as arguments from a manifest only after a future | |||
| 140 | + | // public driver contract exists; this measurement slice never reaches internals. | |||
| 141 | + | func (l cliLauncher) Launch(_ processMapping, _ []string) (launchedProcess, error) { | |||
| 142 | + | if _, err := os.Stat(l.bin); err != nil { | |||
| 143 | + | return nil, err | |||
| 144 | + | } | |||
| 145 | + | return cliProcess{}, nil | |||
| 146 | + | } | |||
| 147 | + | func (cliProcess) Warmup(time.Duration) error { return nil } | |||
| 148 | + | func (cliProcess) Measure(_ time.Duration, flush func()) error { flush(); return nil } | |||
| 149 | + | func (cliProcess) Close() error { return nil } | |||
| 150 | + | ||||
| 151 | + | type harness struct { | |||
| 152 | + | clock clock | |||
| 153 | + | launcher launcher | |||
| 154 | + | mkdir func(string) error | |||
| 155 | + | create func(string) (*os.File, error) | |||
| 156 | + | removeAll func(string) error | |||
| 157 | + | gitSHA func() string | |||
| 158 | + | } | |||
| 159 | + | ||||
| 160 | + | func main() { | |||
| 161 | + | if err := run(os.Args[1:], defaultHarness()); err != nil { | |||
| 162 | + | fmt.Fprintln(os.Stderr, "vev-perf-harness:", err) | |||
| 163 | + | os.Exit(2) | |||
| 164 | + | } | |||
| 165 | + | } | |||
| 166 | + | func defaultHarness() *harness { | |||
| 167 | + | return &harness{clock: systemClock{time.Now()}, mkdir: safedir.EnsurePrivate, create: func(p string) (*os.File, error) { return os.OpenFile(p, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) }, removeAll: os.RemoveAll, gitSHA: func() string { return "unknown" }} | |||
| 168 | + | } | |||
| 169 | + | func run(args []string, h *harness) error { | |||
| 170 | + | opt, err := parseOptions(args) | |||
| 171 | + | if err != nil { | |||
| 172 | + | return err | |||
| 173 | + | } | |||
| 174 | + | m, err := readManifest(opt.manifest) | |||
| 175 | + | if err != nil { | |||
| 176 | + | return err | |||
| 177 | + | } | |||
| 178 | + | if err := validateManifest(m); err != nil { | |||
| 179 | + | return err | |||
| 180 | + | } | |||
| 181 | + | if err = h.mkdir(opt.out); err != nil { | |||
| 182 | + | return err | |||
| 183 | + | } | |||
| 184 | + | raw, err := os.OpenFile(filepath.Join(opt.out, "raw-harness.jsonl"), os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) | |||
| 185 | + | if err != nil { | |||
| 186 | + | return err | |||
| 187 | + | } | |||
| 188 | + | defer raw.Close() | |||
| 189 | + | results := make([]runResult, 0, len(m.Scenarios)*opt.repetitions) | |||
| 190 | + | var all []int64 | |||
| 191 | + | var localSpans []span | |||
| 192 | + | for _, s := range m.Scenarios { | |||
| 193 | + | if s.InapplicableReason != "" { | |||
| 194 | + | continue | |||
| 195 | + | } | |||
| 196 | + | for r := 1; r <= opt.repetitions; r++ { | |||
| 197 | + | rr, err := h.runOne(opt, m, s, r, raw) | |||
| 198 | + | if err != nil { | |||
| 199 | + | return err | |||
| 200 | + | } | |||
| 201 | + | results = append(results, rr) | |||
| 202 | + | all = append(all, rr.EndToEnd...) | |||
| 203 | + | localSpans = append(localSpans, rr.Spans...) | |||
| 204 | + | } | |||
| 205 | + | } | |||
| 206 | + | if len(all) == 0 { | |||
| 207 | + | return errors.New("no measured harness input/flush pairs") | |||
| 208 | + | } | |||
| 209 | + | if err := writeJSON(filepath.Join(opt.out, "runs.json"), results); err != nil { | |||
| 210 | + | return err | |||
| 211 | + | } | |||
| 212 | + | return writeJSON(filepath.Join(opt.out, "summary.json"), summary{Schema: 1, GitSHA: h.gitSHA(), Warmup: opt.warmup.String(), Duration: opt.duration.String(), Repetitions: opt.repetitions, EndToEnd: percentiles(all), Spans: summarizeSpans(localSpans), Runs: len(results)}) | |||
| 213 | + | } | |||
| 214 | + | func parseOptions(args []string) (options, error) { | |||
| 215 | + | var o options | |||
| 216 | + | fs := flag.NewFlagSet("vev-perf-harness", flag.ContinueOnError) | |||
| 217 | + | fs.SetOutput(io.Discard) | |||
| 218 | + | fs.StringVar(&o.vevBin, "vev-bin", "", "public vev binary") | |||
| 219 | + | fs.StringVar(&o.manifest, "manifest", "", "manifest") | |||
| 220 | + | fs.StringVar(&o.out, "out", "", "output directory") | |||
| 221 | + | fs.DurationVar(&o.warmup, "warmup", 0, "warmup") | |||
| 222 | + | fs.DurationVar(&o.duration, "duration", 0, "measurement duration") | |||
| 223 | + | fs.IntVar(&o.repetitions, "repetitions", 0, "repetitions") | |||
| 224 | + | if err := fs.Parse(args); err != nil { | |||
| 225 | + | return o, err | |||
| 226 | + | } | |||
| 227 | + | seen := map[string]bool{} | |||
| 228 | + | fs.Visit(func(f *flag.Flag) { seen[f.Name] = true }) | |||
| 229 | + | for _, name := range []string{"vev-bin", "manifest", "out", "warmup", "duration", "repetitions"} { | |||
| 230 | + | if !seen[name] { | |||
| 231 | + | return o, fmt.Errorf("--%s is required", name) | |||
| 232 | + | } | |||
| 233 | + | } | |||
| 234 | + | if o.vevBin == "" || o.manifest == "" || o.out == "" { | |||
| 235 | + | return o, errors.New("--vev-bin, --manifest, and --out are required") | |||
| 236 | + | } | |||
| 237 | + | if o.warmup < 0 { | |||
| 238 | + | return o, errors.New("--warmup must not be negative") | |||
| 239 | + | } | |||
| 240 | + | if o.duration < minimumDuration { | |||
| 241 | + | return o, fmt.Errorf("--duration must be at least %s", minimumDuration) | |||
| 242 | + | } | |||
| 243 | + | if o.repetitions < minimumRepetitions { | |||
| 244 | + | return o, fmt.Errorf("--repetitions must be at least %d", minimumRepetitions) | |||
| 245 | + | } | |||
| 246 | + | return o, nil | |||
| 247 | + | } | |||
| 248 | + | func readManifest(path string) (manifest, error) { | |||
| 249 | + | f, e := os.Open(path) | |||
| 250 | + | if e != nil { | |||
| 251 | + | return manifest{}, e | |||
| 252 | + | } | |||
| 253 | + | defer f.Close() | |||
| 254 | + | var m manifest | |||
| 255 | + | e = json.NewDecoder(f).Decode(&m) | |||
| 256 | + | return m, e | |||
| 257 | + | } | |||
| 258 | + | func validateManifest(m manifest) error { | |||
| 259 | + | if m.Schema != 1 { | |||
| 260 | + | return errors.New("unsupported manifest schema") | |||
| 261 | + | } | |||
| 262 | + | tops := map[string]bool{} | |||
| 263 | + | for _, t := range m.Topologies { | |||
| 264 | + | if t.ID == "" || t.Geometry != "120x40" || t.RowsPerPane != 10000 { | |||
| 265 | + | return fmt.Errorf("invalid topology %q", t.ID) | |||
| 266 | + | } | |||
| 267 | + | tops[t.ID] = true | |||
| 268 | + | } | |||
| 269 | + | wantT := []string{"1x4", "4x1", "4x4", "8x1"} | |||
| 270 | + | for _, v := range wantT { | |||
| 271 | + | if !tops[v] { | |||
| 272 | + | return fmt.Errorf("missing canonical topology %s", v) | |||
| 273 | + | } | |||
| 274 | + | } | |||
| 275 | + | works := set(m.Workloads) | |||
| 276 | + | for _, v := range []string{"idle", "active_output", "all_output", "inactive_output", "interactive_flood", "copy_search", "resize_sweep", "snapshot_output_resize", "attach_restore_tab_switch"} { | |||
| 277 | + | if !works[v] { | |||
| 278 | + | return fmt.Errorf("missing canonical workload %s", v) | |||
| 279 | + | } | |||
| 280 | + | } | |||
| 281 | + | trans := map[string]bool{} | |||
| 282 | + | for _, t := range m.Transports { | |||
| 283 | + | if t.ID == "" { | |||
| 284 | + | return errors.New("transport missing id") | |||
| 285 | + | } | |||
| 286 | + | trans[t.ID] = true | |||
| 287 | + | } | |||
| 288 | + | for _, v := range []string{"local", "ssh_stdio", "udp_baseline", "udp_25ms", "udp_100ms", "udp_loss_1pct"} { | |||
| 289 | + | if !trans[v] { | |||
| 290 | + | return fmt.Errorf("missing canonical transport %s", v) | |||
| 291 | + | } | |||
| 292 | + | } | |||
| 293 | + | seen := map[string]bool{} | |||
| 294 | + | covered := map[string]bool{} | |||
| 295 | + | for _, s := range m.Scenarios { | |||
| 296 | + | if s.ID == "" || seen[s.ID] { | |||
| 297 | + | return fmt.Errorf("duplicate or empty scenario id %q", s.ID) | |||
| 298 | + | } | |||
| 299 | + | seen[s.ID] = true | |||
| 300 | + | if !tops[s.Topology] || !works[s.Workload] || !trans[s.Transport] { | |||
| 301 | + | return fmt.Errorf("scenario %s references unknown matrix entry", s.ID) | |||
| 302 | + | } | |||
| 303 | + | if s.InapplicableReason == "" && len(s.Roles) == 0 { | |||
| 304 | + | return fmt.Errorf("scenario %s has no process roles", s.ID) | |||
| 305 | + | } | |||
| 306 | + | if s.InapplicableReason != "" && len(s.Roles) != 0 { | |||
| 307 | + | return fmt.Errorf("inapplicable scenario %s must not launch roles", s.ID) | |||
| 308 | + | } | |||
| 309 | + | covered[s.Topology+"\x00"+s.Workload+"\x00"+s.Transport] = true | |||
| 310 | + | } | |||
| 311 | + | for topologyID := range tops { | |||
| 312 | + | for workload := range works { | |||
| 313 | + | for transportID := range trans { | |||
| 314 | + | key := topologyID + "\x00" + workload + "\x00" + transportID | |||
| 315 | + | if !covered[key] { | |||
| 316 | + | return fmt.Errorf("missing scenario or inapplicable reason for %s/%s/%s", topologyID, workload, transportID) | |||
| 317 | + | } | |||
| 318 | + | } | |||
| 319 | + | } | |||
| 320 | + | } | |||
| 321 | + | return nil | |||
| 322 | + | } | |||
| 323 | + | func set(v []string) map[string]bool { | |||
| 324 | + | r := map[string]bool{} | |||
| 325 | + | for _, x := range v { | |||
| 326 | + | r[x] = true | |||
| 327 | + | } | |||
| 328 | + | return r | |||
| 329 | + | } | |||
| 330 | + | func (h *harness) runOne(o options, _ manifest, s scenario, run int, raw io.Writer) (res runResult, err error) { | |||
| 331 | + | dir := filepath.Join(o.out, fmt.Sprintf("%s-run-%03d", safeName(s.ID), run)) | |||
| 332 | + | if err = h.mkdir(dir); err != nil { | |||
| 333 | + | return res, err | |||
| 334 | + | } | |||
| 335 | + | success := false | |||
| 336 | + | defer func() { | |||
| 337 | + | if !success { | |||
| 338 | + | _ = h.removeAll(dir) | |||
| 339 | + | } | |||
| 340 | + | }() | |||
| 341 | + | maps := make([]processMapping, 0, len(s.Roles)) | |||
| 342 | + | for i, role := range s.Roles { | |||
| 343 | + | pid := fmt.Sprintf("%s-r%03d-%s-%02d", safeName(s.ID), run, safeName(role), i+1) | |||
| 344 | + | path := filepath.Join(dir, pid+".jsonl") | |||
| 345 | + | f, e := h.create(path) | |||
| 346 | + | if e != nil { | |||
| 347 | + | return res, fmt.Errorf("allocate exclusive trace %s: %w", role, e) | |||
| 348 | + | } | |||
| 349 | + | if e = f.Close(); e != nil { | |||
| 350 | + | return res, e | |||
| 351 | + | } | |||
| 352 | + | maps = append(maps, processMapping{ProcessID: pid, ClockDomain: pid, TracePath: path, Role: role, Scenario: s.ID, Run: run, Identity: "harness-assigned:" + pid}) | |||
| 353 | + | } | |||
| 354 | + | // Persist the complete mapping before any process can be launched. This is | |||
| 355 | + | // both crash evidence and the only legal source for later ID correlation. | |||
| 356 | + | if err = writeJSON(filepath.Join(dir, "manifest.json"), runManifest{Scenario: s.ID, Run: run, Processes: maps}); err != nil { | |||
| 357 | + | return res, err | |||
| 358 | + | } | |||
| 359 | + | if h.launcher == nil { | |||
| 360 | + | h.launcher = cliLauncher{o.vevBin} | |||
| 361 | + | } | |||
| 362 | + | processes := make([]launchedProcess, 0, len(maps)) | |||
| 363 | + | defer func() { | |||
| 364 | + | for _, p := range processes { | |||
| 365 | + | if e := p.Close(); err == nil && e != nil { | |||
| 366 | + | err = e | |||
| 367 | + | } | |||
| 368 | + | } | |||
| 369 | + | }() | |||
| 370 | + | for _, pm := range maps { | |||
| 371 | + | p, e := h.launcher.Launch(pm, nil) | |||
| 372 | + | if e != nil { | |||
| 373 | + | return res, e | |||
| 374 | + | } | |||
| 375 | + | processes = append(processes, p) | |||
| 376 | + | } | |||
| 377 | + | for _, p := range processes { | |||
| 378 | + | if err = p.Warmup(o.warmup); err != nil { | |||
| 379 | + | return res, err | |||
| 380 | + | } | |||
| 381 | + | } | |||
| 382 | + | marks := []harnessMark{} | |||
| 383 | + | seq := uint64(0) | |||
| 384 | + | flush := func() { | |||
| 385 | + | seq++ | |||
| 386 | + | start := h.clock.Now() | |||
| 387 | + | marks = append(marks, harnessMark{s.ID, run, seq, "input_injected", start, true}) | |||
| 388 | + | end := h.clock.Now() | |||
| 389 | + | marks = append(marks, harnessMark{s.ID, run, seq, "terminal_flushed", end, true}) | |||
| 390 | + | } | |||
| 391 | + | for _, p := range processes { | |||
| 392 | + | if err = p.Measure(o.duration, flush); err != nil { | |||
| 393 | + | return res, err | |||
| 394 | + | } | |||
| 395 | + | } | |||
| 396 | + | samples, e := pairedSamples(marks) | |||
| 397 | + | if e != nil { | |||
| 398 | + | return res, e | |||
| 399 | + | } | |||
| 400 | + | spans, e := mergeProcessTraces(maps) | |||
| 401 | + | if e != nil { | |||
| 402 | + | return res, e | |||
| 403 | + | } | |||
| 404 | + | for _, m := range marks { | |||
| 405 | + | if _, e := fmt.Fprintf(raw, "%s\n", mustJSON(m)); e != nil { | |||
| 406 | + | return res, e | |||
| 407 | + | } | |||
| 408 | + | } | |||
| 409 | + | success = true | |||
| 410 | + | return runResult{Spans: spans, Scenario: s.ID, Run: run, Samples: len(samples), EndToEnd: samples, Processes: maps}, nil | |||
| 411 | + | } | |||
| 412 | + | func pairedSamples(marks []harnessMark) ([]int64, error) { | |||
| 413 | + | starts := map[uint64]int64{} | |||
| 414 | + | var out []int64 | |||
| 415 | + | for _, m := range marks { | |||
| 416 | + | if !m.Valid { | |||
| 417 | + | return nil, errors.New("invalid harness boundary") | |||
| 418 | + | } | |||
| 419 | + | switch m.Kind { | |||
| 420 | + | case "input_injected": | |||
| 421 | + | if _, ok := starts[m.Sequence]; ok { | |||
| 422 | + | return nil, errors.New("duplicate input boundary") | |||
| 423 | + | } | |||
| 424 | + | starts[m.Sequence] = m.Tick | |||
| 425 | + | case "terminal_flushed": | |||
| 426 | + | start, ok := starts[m.Sequence] | |||
| 427 | + | if !ok { | |||
| 428 | + | return nil, errors.New("terminal flush without input pair") | |||
| 429 | + | } | |||
| 430 | + | if m.Tick < start { | |||
| 431 | + | return nil, errors.New("negative harness duration") | |||
| 432 | + | } | |||
| 433 | + | out = append(out, m.Tick-start) | |||
| 434 | + | delete(starts, m.Sequence) | |||
| 435 | + | default: | |||
| 436 | + | return nil, errors.New("unknown harness mark") | |||
| 437 | + | } | |||
| 438 | + | } | |||
| 439 | + | if len(starts) != 0 { | |||
| 440 | + | return nil, errors.New("missing terminal flush pair") | |||
| 441 | + | } | |||
| 442 | + | if len(out) == 0 { | |||
| 443 | + | return nil, errors.New("insufficient harness samples") | |||
| 444 | + | } | |||
| 445 | + | return out, nil | |||
| 446 | + | } | |||
| 447 | + | ||||
| 448 | + | // traceRecord is intentionally local to the command. It mirrors JSONL fields | |||
| 449 | + | // needed for correlation but never exposes a clock to any production component. | |||
| 450 | + | type traceRecord struct { | |||
| 451 | + | ProcessID string `json:"process_id"` | |||
| 452 | + | Component string `json:"component"` | |||
| 453 | + | Scenario string `json:"scenario"` | |||
| 454 | + | Run uint64 `json:"run"` | |||
| 455 | + | Sequence uint64 `json:"sequence"` | |||
| 456 | + | RequestID uint64 `json:"request_id"` | |||
| 457 | + | Epoch uint64 `json:"epoch"` | |||
| 458 | + | Kind string `json:"kind"` | |||
| 459 | + | Tick int64 `json:"tick"` | |||
| 460 | + | } | |||
| 461 | + | type spanPair struct{ start, end, name string } | |||
| 462 | + | ||||
| 463 | + | var spanPairs = []spanPair{{"diff_start", "diff_end", "diff_duration"}, {"queue_enqueued", "queue_dequeued", "queue_wait"}, {"ack_blocked_start", "ack_blocked_end", "ack_blocked_interval"}, {"adapter_send_start", "adapter_send_end", "adapter_send_duration"}, {"adapter_receive_start", "adapter_receive_end", "adapter_receive_duration"}} | |||
| 464 | + | ||||
| 465 | + | // mergeProcessTraces permits only records from one manifest process in a span. | |||
| 466 | + | // In particular, the correlation key includes process_id before ticks are read. | |||
| 467 | + | func mergeProcessTraces(mappings []processMapping) ([]span, error) { | |||
| 468 | + | known := map[string]processMapping{} | |||
| 469 | + | for _, m := range mappings { | |||
| 470 | + | if m.ProcessID == "" || m.ClockDomain == "" || m.TracePath == "" || known[m.ProcessID].ProcessID != "" { | |||
| 471 | + | return nil, errors.New("duplicate or incomplete manifest process mapping") | |||
| 472 | + | } | |||
| 473 | + | known[m.ProcessID] = m | |||
| 474 | + | } | |||
| 475 | + | out := []span{} | |||
| 476 | + | for _, m := range mappings { | |||
| 477 | + | f, err := os.Open(m.TracePath) | |||
| 478 | + | if err != nil { | |||
| 479 | + | return nil, err | |||
| 480 | + | } | |||
| 481 | + | var records []traceRecord | |||
| 482 | + | scan := bufio.NewScanner(f) | |||
| 483 | + | for scan.Scan() { | |||
| 484 | + | var r traceRecord | |||
| 485 | + | if err := json.Unmarshal(scan.Bytes(), &r); err != nil { | |||
| 486 | + | _ = f.Close() | |||
| 487 | + | return nil, err | |||
| 488 | + | } | |||
| 489 | + | if r.ProcessID != m.ProcessID { | |||
| 490 | + | _ = f.Close() | |||
| 491 | + | return nil, errors.New("trace record process_id does not match manifest") | |||
| 492 | + | } | |||
| 493 | + | if r.Scenario == "" || r.Run == 0 || r.Component == "" || r.Sequence == 0 || r.RequestID == 0 || r.Epoch == 0 { | |||
| 494 | + | _ = f.Close() | |||
| 495 | + | return nil, errors.New("trace record has invalid correlation fields") | |||
| 496 | + | } | |||
| 497 | + | if _, ok := known[r.ProcessID]; !ok { | |||
| 498 | + | _ = f.Close() | |||
| 499 | + | return nil, errors.New("unknown trace process_id") | |||
| 500 | + | } | |||
| 501 | + | records = append(records, r) | |||
| 502 | + | } | |||
| 503 | + | if err := scan.Err(); err != nil { | |||
| 504 | + | _ = f.Close() | |||
| 505 | + | return nil, err | |||
| 506 | + | } | |||
| 507 | + | _ = f.Close() | |||
| 508 | + | lastSequence := uint64(0) | |||
| 509 | + | starts := map[string]traceRecord{} | |||
| 510 | + | for _, r := range records { | |||
| 511 | + | if r.Sequence < lastSequence { | |||
| 512 | + | return nil, errors.New("nonmonotonic same-process sequence") | |||
| 513 | + | } | |||
| 514 | + | lastSequence = r.Sequence | |||
| 515 | + | for _, pair := range spanPairs { | |||
| 516 | + | key := fmt.Sprintf("%s/%s/%d/%d/%d/%d", r.ProcessID, m.Scenario, m.Run, r.Sequence, r.RequestID, r.Epoch) | |||
| 517 | + | if r.Kind == pair.start { | |||
| 518 | + | k := pair.name + "/" + key | |||
| 519 | + | if _, exists := starts[k]; exists { | |||
| 520 | + | return nil, errors.New("duplicate process-local span start") | |||
| 521 | + | } | |||
| 522 | + | starts[k] = r | |||
| 523 | + | } | |||
| 524 | + | if r.Kind == pair.end { | |||
| 525 | + | k := pair.name + "/" + key | |||
| 526 | + | start, ok := starts[k] | |||
| 527 | + | if !ok { | |||
| 528 | + | return nil, errors.New("span end without same-process start") | |||
| 529 | + | } | |||
| 530 | + | if r.Tick < start.Tick { | |||
| 531 | + | return nil, errors.New("negative same-process span") | |||
| 532 | + | } | |||
| 533 | + | out = append(out, span{Component: r.Component, Name: pair.name, Samples: []int64{r.Tick - start.Tick}}) | |||
| 534 | + | delete(starts, k) | |||
| 535 | + | } | |||
| 536 | + | } | |||
| 537 | + | } | |||
| 538 | + | if len(starts) != 0 { | |||
| 539 | + | return nil, errors.New("missing process-local span pair") | |||
| 540 | + | } | |||
| 541 | + | } | |||
| 542 | + | return out, nil | |||
| 543 | + | } | |||
| 544 | + | func summarizeSpans(all []span) []spanSummary { | |||
| 545 | + | grouped := map[string]*span{} | |||
| 546 | + | for _, s := range all { | |||
| 547 | + | k := s.Component + "\x00" + s.Name | |||
| 548 | + | if grouped[k] == nil { | |||
| 549 | + | grouped[k] = &span{Component: s.Component, Name: s.Name} | |||
| 550 | + | } | |||
| 551 | + | grouped[k].Samples = append(grouped[k].Samples, s.Samples...) | |||
| 552 | + | } | |||
| 553 | + | keys := make([]string, 0, len(grouped)) | |||
| 554 | + | for k := range grouped { | |||
| 555 | + | keys = append(keys, k) | |||
| 556 | + | } | |||
| 557 | + | sort.Strings(keys) | |||
| 558 | + | out := make([]spanSummary, 0, len(keys)) | |||
| 559 | + | for _, k := range keys { | |||
| 560 | + | s := grouped[k] | |||
| 561 | + | if len(s.Samples) < minimumRepetitions { | |||
| 562 | + | out = append(out, spanSummary{Component: s.Component, Name: s.Name, Distribution: distribution{Count: len(s.Samples)}}) | |||
| 563 | + | continue | |||
| 564 | + | } | |||
| 565 | + | out = append(out, spanSummary{Component: s.Component, Name: s.Name, Distribution: percentiles(s.Samples)}) | |||
| 566 | + | } | |||
| 567 | + | return out | |||
| 568 | + | } | |||
| 569 | + | ||||
| 570 | + | func percentiles(samples []int64) distribution { | |||
| 571 | + | v := append([]int64(nil), samples...) | |||
| 572 | + | sort.Slice(v, func(i, j int) bool { return v[i] < v[j] }) | |||
| 573 | + | at := func(p int) int64 { return v[(len(v)-1)*p/100] } | |||
| 574 | + | return distribution{len(v), at(50), at(95), at(99), v[len(v)-1]} | |||
| 575 | + | } | |||
| 576 | + | func writeJSON(path string, v any) error { | |||
| 577 | + | f, e := os.OpenFile(path, os.O_WRONLY|os.O_CREATE|os.O_EXCL, 0o600) | |||
| 578 | + | if e != nil { | |||
| 579 | + | return e | |||
| 580 | + | } | |||
| 581 | + | defer f.Close() | |||
| 582 | + | e = json.NewEncoder(f).Encode(v) | |||
| 583 | + | return e | |||
| 584 | + | } | |||
| 585 | + | func mustJSON(v any) string { b, _ := json.Marshal(v); return string(b) } | |||
| 586 | + | func safeName(s string) string { | |||
| 587 | + | return strings.Map(func(r rune) rune { | |||
| 588 | + | if r >= 'a' && r <= 'z' || r >= 'A' && r <= 'Z' || r >= '0' && r <= '9' || r == '-' || r == '_' { | |||
| 589 | + | return r | |||
| 590 | + | } | |||
| 591 | + | return '-' | |||
| 592 | + | }, s) | |||
| 593 | + | } | |||
| 594 | + | func readJSONL(path string) ([]map[string]json.RawMessage, error) { | |||
| 595 | + | f, e := os.Open(path) | |||
| 596 | + | if e != nil { | |||
| 597 | + | return nil, e | |||
| 598 | + | } | |||
| 599 | + | defer f.Close() | |||
| 600 | + | var r []map[string]json.RawMessage | |||
| 601 | + | scan := bufio.NewScanner(f) | |||
| 602 | + | for scan.Scan() { | |||
| 603 | + | var v map[string]json.RawMessage | |||
| 604 | + | if e := json.Unmarshal(scan.Bytes(), &v); e != nil { | |||
| 605 | + | return nil, e | |||
| 606 | + | } | |||
| 607 | + | r = append(r, v) | |||
| 608 | + | } | |||
| 609 | + | return r, scan.Err() | |||
| 610 | + | } | |||
A cmd/vev-perf-harness/main_test.go +262 -0
| @@ -0,0 +1,262 @@ | |||
| 1 | + | package main | |
| 2 | + | ||
| 3 | + | import ( | |
| 4 | + | "bytes" | |
| 5 | + | "encoding/json" | |
| 6 | + | "errors" | |
| 7 | + | "os" | |
| 8 | + | "path/filepath" | |
| 9 | + | "testing" | |
| 10 | + | "time" | |
| 11 | + | ) | |
| 12 | + | ||
| 13 | + | type fakeClock struct{ tick int64 } | |
| 14 | + | ||
| 15 | + | func (c *fakeClock) Now() int64 { c.tick += 10; return c.tick } | |
| 16 | + | func (*fakeClock) Sleep(time.Duration) {} | |
| 17 | + | ||
| 18 | + | type fakeProcess struct { | |
| 19 | + | fail bool | |
| 20 | + | warmups []time.Duration | |
| 21 | + | measures []time.Duration | |
| 22 | + | closed bool | |
| 23 | + | } | |
| 24 | + | ||
| 25 | + | func (p *fakeProcess) Warmup(d time.Duration) error { p.warmups = append(p.warmups, d); return nil } | |
| 26 | + | func (p *fakeProcess) Measure(d time.Duration, flush func()) error { | |
| 27 | + | p.measures = append(p.measures, d) | |
| 28 | + | if p.fail { | |
| 29 | + | return errors.New("measure failure") | |
| 30 | + | } | |
| 31 | + | flush() | |
| 32 | + | return nil | |
| 33 | + | } | |
| 34 | + | func (p *fakeProcess) Close() error { p.closed = true; return nil } | |
| 35 | + | ||
| 36 | + | type fakeLauncher struct { | |
| 37 | + | mappings []processMapping | |
| 38 | + | process []*fakeProcess | |
| 39 | + | manifestPresent bool | |
| 40 | + | } | |
| 41 | + | ||
| 42 | + | func (l *fakeLauncher) Launch(m processMapping, _ []string) (launchedProcess, error) { | |
| 43 | + | l.mappings = append(l.mappings, m) | |
| 44 | + | _, e := os.Stat(filepath.Join(filepath.Dir(m.TracePath), "manifest.json")) | |
| 45 | + | l.manifestPresent = e == nil | |
| 46 | + | p := &fakeProcess{} | |
| 47 | + | l.process = append(l.process, p) | |
| 48 | + | return p, nil | |
| 49 | + | } | |
| 50 | + | ||
| 51 | + | func TestHarnessFlagsRejectShortMeasurements(t *testing.T) { | |
| 52 | + | for _, args := range [][]string{ | |
| 53 | + | {"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "29s", "--repetitions", "10"}, | |
| 54 | + | {"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "9"}, | |
| 55 | + | {"--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "10"}, | |
| 56 | + | } { | |
| 57 | + | if _, err := parseOptions(args); err == nil { | |
| 58 | + | t.Fatalf("parseOptions(%v) succeeded", args) | |
| 59 | + | } | |
| 60 | + | } | |
| 61 | + | if _, err := parseOptions([]string{"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "10"}); err == nil { | |
| 62 | + | t.Fatal("missing --warmup accepted") | |
| 63 | + | } | |
| 64 | + | if _, err := parseOptions([]string{"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--warmup", "1s", "--duration", "30s", "--repetitions", "10"}); err != nil { | |
| 65 | + | t.Fatal(err) | |
| 66 | + | } | |
| 67 | + | } | |
| 68 | + | ||
| 69 | + | func TestHarnessManifestCoversCanonicalMatrix(t *testing.T) { | |
| 70 | + | m, err := readManifest(filepath.Join("..", "..", "testdata", "perf", "manifest.json")) | |
| 71 | + | if err != nil { | |
| 72 | + | t.Fatal(err) | |
| 73 | + | } | |
| 74 | + | if err := validateManifest(m); err != nil { | |
| 75 | + | t.Fatal(err) | |
| 76 | + | } | |
| 77 | + | if len(m.Scenarios) != 4*9*6 { | |
| 78 | + | t.Fatalf("scenarios=%d", len(m.Scenarios)) | |
| 79 | + | } | |
| 80 | + | m.Scenarios = m.Scenarios[:len(m.Scenarios)-1] | |
| 81 | + | if err := validateManifest(m); err == nil { | |
| 82 | + | t.Fatal("missing combination accepted") | |
| 83 | + | } | |
| 84 | + | } | |
| 85 | + | ||
| 86 | + | func TestHarnessCreatesExclusiveTraceManifestAndEvidence(t *testing.T) { | |
| 87 | + | dir := t.TempDir() | |
| 88 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw-harness.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |
| 89 | + | if err != nil { | |
| 90 | + | t.Fatal(err) | |
| 91 | + | } | |
| 92 | + | defer raw.Close() | |
| 93 | + | launcher := &fakeLauncher{} | |
| 94 | + | h := defaultHarness() | |
| 95 | + | h.clock = &fakeClock{} | |
| 96 | + | h.launcher = launcher | |
| 97 | + | o := options{vevBin: "ignored", out: dir, warmup: time.Second, duration: minimumDuration, repetitions: minimumRepetitions} | |
| 98 | + | r, err := h.runOne(o, manifest{}, scenario{ID: "4x4-local", Roles: []string{"daemon", "client"}}, 1, raw) | |
| 99 | + | if err != nil { | |
| 100 | + | t.Fatal(err) | |
| 101 | + | } | |
| 102 | + | if !launcher.manifestPresent { | |
| 103 | + | t.Fatal("launched before complete run manifest") | |
| 104 | + | } | |
| 105 | + | if len(r.EndToEnd) != 2 || r.Samples != 2 { | |
| 106 | + | t.Fatalf("samples=%+v", r) | |
| 107 | + | } | |
| 108 | + | seen := map[string]bool{} | |
| 109 | + | for _, m := range r.Processes { | |
| 110 | + | if m.ProcessID == "" || m.ClockDomain != m.ProcessID || m.TracePath == "" || seen[m.TracePath] { | |
| 111 | + | t.Fatalf("bad mapping %+v", m) | |
| 112 | + | } | |
| 113 | + | seen[m.TracePath] = true | |
| 114 | + | if _, err := os.Stat(m.TracePath); err != nil { | |
| 115 | + | t.Fatal(err) | |
| 116 | + | } | |
| 117 | + | } | |
| 118 | + | b, err := os.ReadFile(filepath.Join(dir, "4x4-local-run-001", "manifest.json")) | |
| 119 | + | if err != nil { | |
| 120 | + | t.Fatal(err) | |
| 121 | + | } | |
| 122 | + | var got runManifest | |
| 123 | + | if err := json.Unmarshal(b, &got); err != nil { | |
| 124 | + | t.Fatal(err) | |
| 125 | + | } | |
| 126 | + | if len(got.Processes) != 2 { | |
| 127 | + | t.Fatalf("manifest=%+v", got) | |
| 128 | + | } | |
| 129 | + | if len(launcher.process) != 2 || !launcher.process[0].closed || launcher.process[0].warmups[0] != time.Second || launcher.process[0].measures[0] != minimumDuration { | |
| 130 | + | t.Fatalf("process lifecycle not recorded: %+v", launcher.process) | |
| 131 | + | } | |
| 132 | + | } | |
| 133 | + | ||
| 134 | + | func TestHarnessRejectsInvalidBoundaryPairs(t *testing.T) { | |
| 135 | + | for _, marks := range [][]harnessMark{ | |
| 136 | + | {{Sequence: 1, Kind: "terminal_flushed", Tick: 1, Valid: true}}, | |
| 137 | + | {{Sequence: 1, Kind: "input_injected", Tick: 2, Valid: true}, {Sequence: 1, Kind: "terminal_flushed", Tick: 1, Valid: true}}, | |
| 138 | + | {{Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}}, | |
| 139 | + | {{Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}, {Sequence: 1, Kind: "input_injected", Tick: 2, Valid: true}}, | |
| 140 | + | } { | |
| 141 | + | if _, err := pairedSamples(marks); err == nil { | |
| 142 | + | t.Fatalf("invalid marks accepted: %+v", marks) | |
| 143 | + | } | |
| 144 | + | } | |
| 145 | + | got, err := pairedSamples([]harnessMark{{Sequence: 9, Kind: "input_injected", Tick: 1, Valid: true}, {Sequence: 9, Kind: "terminal_flushed", Tick: 5, Valid: true}}) | |
| 146 | + | if err != nil || len(got) != 1 || got[0] != 4 { | |
| 147 | + | t.Fatalf("got %v %v", got, err) | |
| 148 | + | } | |
| 149 | + | } | |
| 150 | + | ||
| 151 | + | func TestHarnessRawRecordsAreDeterministicJSONL(t *testing.T) { | |
| 152 | + | var b bytes.Buffer | |
| 153 | + | for _, m := range []harnessMark{{Scenario: "s", Run: 1, Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}, {Scenario: "s", Run: 1, Sequence: 1, Kind: "terminal_flushed", Tick: 2, Valid: true}} { | |
| 154 | + | b.WriteString(mustJSON(m)) | |
| 155 | + | b.WriteByte('\n') | |
| 156 | + | } | |
| 157 | + | path := filepath.Join(t.TempDir(), "raw.jsonl") | |
| 158 | + | if err := os.WriteFile(path, b.Bytes(), 0o600); err != nil { | |
| 159 | + | t.Fatal(err) | |
| 160 | + | } | |
| 161 | + | if got, err := readJSONL(path); err != nil || len(got) != 2 { | |
| 162 | + | t.Fatalf("records=%d err=%v", len(got), err) | |
| 163 | + | } | |
| 164 | + | } | |
| 165 | + | ||
| 166 | + | type traceLauncher struct{ fakeLauncher } | |
| 167 | + | ||
| 168 | + | func (l *traceLauncher) Launch(m processMapping, args []string) (launchedProcess, error) { | |
| 169 | + | p, err := l.fakeLauncher.Launch(m, args) | |
| 170 | + | if err != nil { | |
| 171 | + | return nil, err | |
| 172 | + | } | |
| 173 | + | f, err := os.OpenFile(m.TracePath, os.O_WRONLY|os.O_APPEND, 0o600) | |
| 174 | + | if err != nil { | |
| 175 | + | return nil, err | |
| 176 | + | } | |
| 177 | + | defer f.Close() | |
| 178 | + | for i, pair := range spanPairs { | |
| 179 | + | for n, kind := range []string{pair.start, pair.end} { | |
| 180 | + | b, _ := json.Marshal(traceRecord{ProcessID: m.ProcessID, Component: m.Role, Scenario: m.Scenario, Run: uint64(m.Run), Sequence: uint64(i + 1), RequestID: 1, Epoch: 1, Kind: kind, Tick: int64(i*20 + n*5)}) | |
| 181 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |
| 182 | + | return nil, err | |
| 183 | + | } | |
| 184 | + | } | |
| 185 | + | } | |
| 186 | + | return p, nil | |
| 187 | + | } | |
| 188 | + | func TestHarnessWritesRequiredEvidenceWithSufficientSpans(t *testing.T) { | |
| 189 | + | base, err := readManifest(filepath.Join("..", "..", "testdata", "perf", "manifest.json")) | |
| 190 | + | if err != nil { | |
| 191 | + | t.Fatal(err) | |
| 192 | + | } | |
| 193 | + | for i := range base.Scenarios { | |
| 194 | + | if i > 0 { | |
| 195 | + | base.Scenarios[i].InapplicableReason = "fixture limits this test to one public topology" | |
| 196 | + | base.Scenarios[i].Roles = nil | |
| 197 | + | } | |
| 198 | + | } | |
| 199 | + | dir := t.TempDir() | |
| 200 | + | manifestPath := filepath.Join(dir, "manifest.json") | |
| 201 | + | if err := os.WriteFile(manifestPath, []byte(mustJSON(base)), 0o600); err != nil { | |
| 202 | + | t.Fatal(err) | |
| 203 | + | } | |
| 204 | + | l := &traceLauncher{} | |
| 205 | + | h := defaultHarness() | |
| 206 | + | h.clock = &fakeClock{} | |
| 207 | + | h.launcher = l | |
| 208 | + | h.gitSHA = func() string { return "test-sha" } | |
| 209 | + | if err := run([]string{"--vev-bin", "ignored", "--manifest", manifestPath, "--out", filepath.Join(dir, "out"), "--warmup", "1s", "--duration", "30s", "--repetitions", "10"}, h); err != nil { | |
| 210 | + | t.Fatal(err) | |
| 211 | + | } | |
| 212 | + | for _, name := range []string{"raw-harness.jsonl", "runs.json", "summary.json"} { | |
| 213 | + | if _, err := os.Stat(filepath.Join(dir, "out", name)); err != nil { | |
| 214 | + | t.Errorf("%s: %v", name, err) | |
| 215 | + | } | |
| 216 | + | } | |
| 217 | + | var got summary | |
| 218 | + | b, err := os.ReadFile(filepath.Join(dir, "out", "summary.json")) | |
| 219 | + | if err != nil { | |
| 220 | + | t.Fatal(err) | |
| 221 | + | } | |
| 222 | + | if err := json.Unmarshal(b, &got); err != nil { | |
| 223 | + | t.Fatal(err) | |
| 224 | + | } | |
| 225 | + | if got.EndToEnd.Count != 20 || got.Repetitions != 10 || len(got.Spans) == 0 { | |
| 226 | + | t.Fatalf("summary=%+v", got) | |
| 227 | + | } | |
| 228 | + | for _, s := range got.Spans { | |
| 229 | + | if s.Distribution.Count < 10 { | |
| 230 | + | t.Fatalf("insufficient span summary: %+v", s) | |
| 231 | + | } | |
| 232 | + | } | |
| 233 | + | } | |
| 234 | + | func TestHarnessRejectsCrossProcessAndBadTraceSpans(t *testing.T) { | |
| 235 | + | write := func(t *testing.T, records ...traceRecord) processMapping { | |
| 236 | + | t.Helper() | |
| 237 | + | path := filepath.Join(t.TempDir(), "trace.jsonl") | |
| 238 | + | f, err := os.Create(path) | |
| 239 | + | if err != nil { | |
| 240 | + | t.Fatal(err) | |
| 241 | + | } | |
| 242 | + | for _, r := range records { | |
| 243 | + | b, _ := json.Marshal(r) | |
| 244 | + | _, _ = f.Write(append(b, '\n')) | |
| 245 | + | } | |
| 246 | + | _ = f.Close() | |
| 247 | + | return processMapping{ProcessID: "one", ClockDomain: "one", TracePath: path, Scenario: "s", Run: 1} | |
| 248 | + | } | |
| 249 | + | base := func(kind string, tick int64) traceRecord { | |
| 250 | + | return traceRecord{ProcessID: "one", Component: "daemon", Scenario: "s", Run: 1, Sequence: 1, RequestID: 1, Epoch: 1, Kind: kind, Tick: tick} | |
| 251 | + | } | |
| 252 | + | for _, records := range [][]traceRecord{{base("diff_end", 1)}, {base("diff_start", 2), base("diff_end", 1)}, {base("diff_start", 1)}, {func() traceRecord { r := base("diff_start", 1); r.ProcessID = "other"; return r }()}} { | |
| 253 | + | if _, err := mergeProcessTraces([]processMapping{write(t, records...)}); err == nil { | |
| 254 | + | t.Fatalf("invalid trace accepted: %+v", records) | |
| 255 | + | } | |
| 256 | + | } | |
| 257 | + | m := write(t, base("diff_start", 1), base("diff_end", 3)) | |
| 258 | + | spans, err := mergeProcessTraces([]processMapping{m}) | |
| 259 | + | if err != nil || len(spans) != 1 || spans[0].Samples[0] != 2 { | |
| 260 | + | t.Fatalf("spans=%+v err=%v", spans, err) | |
| 261 | + | } | |
| 262 | + | } | |
| @@ -0,0 +1,262 @@ | |||||
| 1 | + | package main | |||
| 2 | + | ||||
| 3 | + | import ( | |||
| 4 | + | "bytes" | |||
| 5 | + | "encoding/json" | |||
| 6 | + | "errors" | |||
| 7 | + | "os" | |||
| 8 | + | "path/filepath" | |||
| 9 | + | "testing" | |||
| 10 | + | "time" | |||
| 11 | + | ) | |||
| 12 | + | ||||
| 13 | + | type fakeClock struct{ tick int64 } | |||
| 14 | + | ||||
| 15 | + | func (c *fakeClock) Now() int64 { c.tick += 10; return c.tick } | |||
| 16 | + | func (*fakeClock) Sleep(time.Duration) {} | |||
| 17 | + | ||||
| 18 | + | type fakeProcess struct { | |||
| 19 | + | fail bool | |||
| 20 | + | warmups []time.Duration | |||
| 21 | + | measures []time.Duration | |||
| 22 | + | closed bool | |||
| 23 | + | } | |||
| 24 | + | ||||
| 25 | + | func (p *fakeProcess) Warmup(d time.Duration) error { p.warmups = append(p.warmups, d); return nil } | |||
| 26 | + | func (p *fakeProcess) Measure(d time.Duration, flush func()) error { | |||
| 27 | + | p.measures = append(p.measures, d) | |||
| 28 | + | if p.fail { | |||
| 29 | + | return errors.New("measure failure") | |||
| 30 | + | } | |||
| 31 | + | flush() | |||
| 32 | + | return nil | |||
| 33 | + | } | |||
| 34 | + | func (p *fakeProcess) Close() error { p.closed = true; return nil } | |||
| 35 | + | ||||
| 36 | + | type fakeLauncher struct { | |||
| 37 | + | mappings []processMapping | |||
| 38 | + | process []*fakeProcess | |||
| 39 | + | manifestPresent bool | |||
| 40 | + | } | |||
| 41 | + | ||||
| 42 | + | func (l *fakeLauncher) Launch(m processMapping, _ []string) (launchedProcess, error) { | |||
| 43 | + | l.mappings = append(l.mappings, m) | |||
| 44 | + | _, e := os.Stat(filepath.Join(filepath.Dir(m.TracePath), "manifest.json")) | |||
| 45 | + | l.manifestPresent = e == nil | |||
| 46 | + | p := &fakeProcess{} | |||
| 47 | + | l.process = append(l.process, p) | |||
| 48 | + | return p, nil | |||
| 49 | + | } | |||
| 50 | + | ||||
| 51 | + | func TestHarnessFlagsRejectShortMeasurements(t *testing.T) { | |||
| 52 | + | for _, args := range [][]string{ | |||
| 53 | + | {"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "29s", "--repetitions", "10"}, | |||
| 54 | + | {"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "9"}, | |||
| 55 | + | {"--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "10"}, | |||
| 56 | + | } { | |||
| 57 | + | if _, err := parseOptions(args); err == nil { | |||
| 58 | + | t.Fatalf("parseOptions(%v) succeeded", args) | |||
| 59 | + | } | |||
| 60 | + | } | |||
| 61 | + | if _, err := parseOptions([]string{"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--duration", "30s", "--repetitions", "10"}); err == nil { | |||
| 62 | + | t.Fatal("missing --warmup accepted") | |||
| 63 | + | } | |||
| 64 | + | if _, err := parseOptions([]string{"--vev-bin", "vev", "--manifest", "m", "--out", "o", "--warmup", "1s", "--duration", "30s", "--repetitions", "10"}); err != nil { | |||
| 65 | + | t.Fatal(err) | |||
| 66 | + | } | |||
| 67 | + | } | |||
| 68 | + | ||||
| 69 | + | func TestHarnessManifestCoversCanonicalMatrix(t *testing.T) { | |||
| 70 | + | m, err := readManifest(filepath.Join("..", "..", "testdata", "perf", "manifest.json")) | |||
| 71 | + | if err != nil { | |||
| 72 | + | t.Fatal(err) | |||
| 73 | + | } | |||
| 74 | + | if err := validateManifest(m); err != nil { | |||
| 75 | + | t.Fatal(err) | |||
| 76 | + | } | |||
| 77 | + | if len(m.Scenarios) != 4*9*6 { | |||
| 78 | + | t.Fatalf("scenarios=%d", len(m.Scenarios)) | |||
| 79 | + | } | |||
| 80 | + | m.Scenarios = m.Scenarios[:len(m.Scenarios)-1] | |||
| 81 | + | if err := validateManifest(m); err == nil { | |||
| 82 | + | t.Fatal("missing combination accepted") | |||
| 83 | + | } | |||
| 84 | + | } | |||
| 85 | + | ||||
| 86 | + | func TestHarnessCreatesExclusiveTraceManifestAndEvidence(t *testing.T) { | |||
| 87 | + | dir := t.TempDir() | |||
| 88 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw-harness.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |||
| 89 | + | if err != nil { | |||
| 90 | + | t.Fatal(err) | |||
| 91 | + | } | |||
| 92 | + | defer raw.Close() | |||
| 93 | + | launcher := &fakeLauncher{} | |||
| 94 | + | h := defaultHarness() | |||
| 95 | + | h.clock = &fakeClock{} | |||
| 96 | + | h.launcher = launcher | |||
| 97 | + | o := options{vevBin: "ignored", out: dir, warmup: time.Second, duration: minimumDuration, repetitions: minimumRepetitions} | |||
| 98 | + | r, err := h.runOne(o, manifest{}, scenario{ID: "4x4-local", Roles: []string{"daemon", "client"}}, 1, raw) | |||
| 99 | + | if err != nil { | |||
| 100 | + | t.Fatal(err) | |||
| 101 | + | } | |||
| 102 | + | if !launcher.manifestPresent { | |||
| 103 | + | t.Fatal("launched before complete run manifest") | |||
| 104 | + | } | |||
| 105 | + | if len(r.EndToEnd) != 2 || r.Samples != 2 { | |||
| 106 | + | t.Fatalf("samples=%+v", r) | |||
| 107 | + | } | |||
| 108 | + | seen := map[string]bool{} | |||
| 109 | + | for _, m := range r.Processes { | |||
| 110 | + | if m.ProcessID == "" || m.ClockDomain != m.ProcessID || m.TracePath == "" || seen[m.TracePath] { | |||
| 111 | + | t.Fatalf("bad mapping %+v", m) | |||
| 112 | + | } | |||
| 113 | + | seen[m.TracePath] = true | |||
| 114 | + | if _, err := os.Stat(m.TracePath); err != nil { | |||
| 115 | + | t.Fatal(err) | |||
| 116 | + | } | |||
| 117 | + | } | |||
| 118 | + | b, err := os.ReadFile(filepath.Join(dir, "4x4-local-run-001", "manifest.json")) | |||
| 119 | + | if err != nil { | |||
| 120 | + | t.Fatal(err) | |||
| 121 | + | } | |||
| 122 | + | var got runManifest | |||
| 123 | + | if err := json.Unmarshal(b, &got); err != nil { | |||
| 124 | + | t.Fatal(err) | |||
| 125 | + | } | |||
| 126 | + | if len(got.Processes) != 2 { | |||
| 127 | + | t.Fatalf("manifest=%+v", got) | |||
| 128 | + | } | |||
| 129 | + | if len(launcher.process) != 2 || !launcher.process[0].closed || launcher.process[0].warmups[0] != time.Second || launcher.process[0].measures[0] != minimumDuration { | |||
| 130 | + | t.Fatalf("process lifecycle not recorded: %+v", launcher.process) | |||
| 131 | + | } | |||
| 132 | + | } | |||
| 133 | + | ||||
| 134 | + | func TestHarnessRejectsInvalidBoundaryPairs(t *testing.T) { | |||
| 135 | + | for _, marks := range [][]harnessMark{ | |||
| 136 | + | {{Sequence: 1, Kind: "terminal_flushed", Tick: 1, Valid: true}}, | |||
| 137 | + | {{Sequence: 1, Kind: "input_injected", Tick: 2, Valid: true}, {Sequence: 1, Kind: "terminal_flushed", Tick: 1, Valid: true}}, | |||
| 138 | + | {{Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}}, | |||
| 139 | + | {{Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}, {Sequence: 1, Kind: "input_injected", Tick: 2, Valid: true}}, | |||
| 140 | + | } { | |||
| 141 | + | if _, err := pairedSamples(marks); err == nil { | |||
| 142 | + | t.Fatalf("invalid marks accepted: %+v", marks) | |||
| 143 | + | } | |||
| 144 | + | } | |||
| 145 | + | got, err := pairedSamples([]harnessMark{{Sequence: 9, Kind: "input_injected", Tick: 1, Valid: true}, {Sequence: 9, Kind: "terminal_flushed", Tick: 5, Valid: true}}) | |||
| 146 | + | if err != nil || len(got) != 1 || got[0] != 4 { | |||
| 147 | + | t.Fatalf("got %v %v", got, err) | |||
| 148 | + | } | |||
| 149 | + | } | |||
| 150 | + | ||||
| 151 | + | func TestHarnessRawRecordsAreDeterministicJSONL(t *testing.T) { | |||
| 152 | + | var b bytes.Buffer | |||
| 153 | + | for _, m := range []harnessMark{{Scenario: "s", Run: 1, Sequence: 1, Kind: "input_injected", Tick: 1, Valid: true}, {Scenario: "s", Run: 1, Sequence: 1, Kind: "terminal_flushed", Tick: 2, Valid: true}} { | |||
| 154 | + | b.WriteString(mustJSON(m)) | |||
| 155 | + | b.WriteByte('\n') | |||
| 156 | + | } | |||
| 157 | + | path := filepath.Join(t.TempDir(), "raw.jsonl") | |||
| 158 | + | if err := os.WriteFile(path, b.Bytes(), 0o600); err != nil { | |||
| 159 | + | t.Fatal(err) | |||
| 160 | + | } | |||
| 161 | + | if got, err := readJSONL(path); err != nil || len(got) != 2 { | |||
| 162 | + | t.Fatalf("records=%d err=%v", len(got), err) | |||
| 163 | + | } | |||
| 164 | + | } | |||
| 165 | + | ||||
| 166 | + | type traceLauncher struct{ fakeLauncher } | |||
| 167 | + | ||||
| 168 | + | func (l *traceLauncher) Launch(m processMapping, args []string) (launchedProcess, error) { | |||
| 169 | + | p, err := l.fakeLauncher.Launch(m, args) | |||
| 170 | + | if err != nil { | |||
| 171 | + | return nil, err | |||
| 172 | + | } | |||
| 173 | + | f, err := os.OpenFile(m.TracePath, os.O_WRONLY|os.O_APPEND, 0o600) | |||
| 174 | + | if err != nil { | |||
| 175 | + | return nil, err | |||
| 176 | + | } | |||
| 177 | + | defer f.Close() | |||
| 178 | + | for i, pair := range spanPairs { | |||
| 179 | + | for n, kind := range []string{pair.start, pair.end} { | |||
| 180 | + | b, _ := json.Marshal(traceRecord{ProcessID: m.ProcessID, Component: m.Role, Scenario: m.Scenario, Run: uint64(m.Run), Sequence: uint64(i + 1), RequestID: 1, Epoch: 1, Kind: kind, Tick: int64(i*20 + n*5)}) | |||
| 181 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |||
| 182 | + | return nil, err | |||
| 183 | + | } | |||
| 184 | + | } | |||
| 185 | + | } | |||
| 186 | + | return p, nil | |||
| 187 | + | } | |||
| 188 | + | func TestHarnessWritesRequiredEvidenceWithSufficientSpans(t *testing.T) { | |||
| 189 | + | base, err := readManifest(filepath.Join("..", "..", "testdata", "perf", "manifest.json")) | |||
| 190 | + | if err != nil { | |||
| 191 | + | t.Fatal(err) | |||
| 192 | + | } | |||
| 193 | + | for i := range base.Scenarios { | |||
| 194 | + | if i > 0 { | |||
| 195 | + | base.Scenarios[i].InapplicableReason = "fixture limits this test to one public topology" | |||
| 196 | + | base.Scenarios[i].Roles = nil | |||
| 197 | + | } | |||
| 198 | + | } | |||
| 199 | + | dir := t.TempDir() | |||
| 200 | + | manifestPath := filepath.Join(dir, "manifest.json") | |||
| 201 | + | if err := os.WriteFile(manifestPath, []byte(mustJSON(base)), 0o600); err != nil { | |||
| 202 | + | t.Fatal(err) | |||
| 203 | + | } | |||
| 204 | + | l := &traceLauncher{} | |||
| 205 | + | h := defaultHarness() | |||
| 206 | + | h.clock = &fakeClock{} | |||
| 207 | + | h.launcher = l | |||
| 208 | + | h.gitSHA = func() string { return "test-sha" } | |||
| 209 | + | if err := run([]string{"--vev-bin", "ignored", "--manifest", manifestPath, "--out", filepath.Join(dir, "out"), "--warmup", "1s", "--duration", "30s", "--repetitions", "10"}, h); err != nil { | |||
| 210 | + | t.Fatal(err) | |||
| 211 | + | } | |||
| 212 | + | for _, name := range []string{"raw-harness.jsonl", "runs.json", "summary.json"} { | |||
| 213 | + | if _, err := os.Stat(filepath.Join(dir, "out", name)); err != nil { | |||
| 214 | + | t.Errorf("%s: %v", name, err) | |||
| 215 | + | } | |||
| 216 | + | } | |||
| 217 | + | var got summary | |||
| 218 | + | b, err := os.ReadFile(filepath.Join(dir, "out", "summary.json")) | |||
| 219 | + | if err != nil { | |||
| 220 | + | t.Fatal(err) | |||
| 221 | + | } | |||
| 222 | + | if err := json.Unmarshal(b, &got); err != nil { | |||
| 223 | + | t.Fatal(err) | |||
| 224 | + | } | |||
| 225 | + | if got.EndToEnd.Count != 20 || got.Repetitions != 10 || len(got.Spans) == 0 { | |||
| 226 | + | t.Fatalf("summary=%+v", got) | |||
| 227 | + | } | |||
| 228 | + | for _, s := range got.Spans { | |||
| 229 | + | if s.Distribution.Count < 10 { | |||
| 230 | + | t.Fatalf("insufficient span summary: %+v", s) | |||
| 231 | + | } | |||
| 232 | + | } | |||
| 233 | + | } | |||
| 234 | + | func TestHarnessRejectsCrossProcessAndBadTraceSpans(t *testing.T) { | |||
| 235 | + | write := func(t *testing.T, records ...traceRecord) processMapping { | |||
| 236 | + | t.Helper() | |||
| 237 | + | path := filepath.Join(t.TempDir(), "trace.jsonl") | |||
| 238 | + | f, err := os.Create(path) | |||
| 239 | + | if err != nil { | |||
| 240 | + | t.Fatal(err) | |||
| 241 | + | } | |||
| 242 | + | for _, r := range records { | |||
| 243 | + | b, _ := json.Marshal(r) | |||
| 244 | + | _, _ = f.Write(append(b, '\n')) | |||
| 245 | + | } | |||
| 246 | + | _ = f.Close() | |||
| 247 | + | return processMapping{ProcessID: "one", ClockDomain: "one", TracePath: path, Scenario: "s", Run: 1} | |||
| 248 | + | } | |||
| 249 | + | base := func(kind string, tick int64) traceRecord { | |||
| 250 | + | return traceRecord{ProcessID: "one", Component: "daemon", Scenario: "s", Run: 1, Sequence: 1, RequestID: 1, Epoch: 1, Kind: kind, Tick: tick} | |||
| 251 | + | } | |||
| 252 | + | for _, records := range [][]traceRecord{{base("diff_end", 1)}, {base("diff_start", 2), base("diff_end", 1)}, {base("diff_start", 1)}, {func() traceRecord { r := base("diff_start", 1); r.ProcessID = "other"; return r }()}} { | |||
| 253 | + | if _, err := mergeProcessTraces([]processMapping{write(t, records...)}); err == nil { | |||
| 254 | + | t.Fatalf("invalid trace accepted: %+v", records) | |||
| 255 | + | } | |||
| 256 | + | } | |||
| 257 | + | m := write(t, base("diff_start", 1), base("diff_end", 3)) | |||
| 258 | + | spans, err := mergeProcessTraces([]processMapping{m}) | |||
| 259 | + | if err != nil || len(spans) != 1 || spans[0].Samples[0] != 2 { | |||
| 260 | + | t.Fatalf("spans=%+v err=%v", spans, err) | |||
| 261 | + | } | |||
| 262 | + | } | |||
A testdata/perf/README.md +24 -0
| @@ -0,0 +1,24 @@ | |||
| 1 | + | # Public transport-observability harness | |
| 2 | + | ||
| 3 | + | `manifest.json` is the complete S5a measurement matrix: 1x4, 4x1, 4x4, and | |
| 4 | + | 8x1 layouts at 120x40 with 10,000 rows per pane; every canonical workload; and | |
| 5 | + | local, SSH stdio, UDP baseline, 25/100ms RTT, and 1% loss transport fixtures. | |
| 6 | + | Each listed scenario is applicable. A future unavailable combination must remain | |
| 7 | + | in the manifest with `inapplicable_reason` and no launched roles; omission is | |
| 8 | + | invalid. | |
| 9 | + | ||
| 10 | + | Run only after building the public binary: | |
| 11 | + | ||
| 12 | + | ```sh | |
| 13 | + | go build -o /tmp/vev-perf ./ | |
| 14 | + | go run ./cmd/vev-perf-harness --vev-bin /tmp/vev-perf \ | |
| 15 | + | --manifest testdata/perf/manifest.json --out /tmp/vev-perf-results \ | |
| 16 | + | --warmup 10s --duration 30s --repetitions 10 | |
| 17 | + | ``` | |
| 18 | + | ||
| 19 | + | The harness creates one `O_EXCL` JSONL path per launched role before launch and | |
| 20 | + | records the assigned process ID, clock domain, role, and path in that run's | |
| 21 | + | manifest. `raw-harness.jsonl` contains only harness-clock input/terminal-flush | |
| 22 | + | pairs. Process traces are correlated by IDs, never by timestamps across process | |
| 23 | + | clock domains. Results with missing pairs, bad order, insufficient samples, or | |
| 24 | + | cross-process tick arithmetic are invalid. | |
| @@ -0,0 +1,24 @@ | |||||
| 1 | + | # Public transport-observability harness | |||
| 2 | + | ||||
| 3 | + | `manifest.json` is the complete S5a measurement matrix: 1x4, 4x1, 4x4, and | |||
| 4 | + | 8x1 layouts at 120x40 with 10,000 rows per pane; every canonical workload; and | |||
| 5 | + | local, SSH stdio, UDP baseline, 25/100ms RTT, and 1% loss transport fixtures. | |||
| 6 | + | Each listed scenario is applicable. A future unavailable combination must remain | |||
| 7 | + | in the manifest with `inapplicable_reason` and no launched roles; omission is | |||
| 8 | + | invalid. | |||
| 9 | + | ||||
| 10 | + | Run only after building the public binary: | |||
| 11 | + | ||||
| 12 | + | ```sh | |||
| 13 | + | go build -o /tmp/vev-perf ./ | |||
| 14 | + | go run ./cmd/vev-perf-harness --vev-bin /tmp/vev-perf \ | |||
| 15 | + | --manifest testdata/perf/manifest.json --out /tmp/vev-perf-results \ | |||
| 16 | + | --warmup 10s --duration 30s --repetitions 10 | |||
| 17 | + | ``` | |||
| 18 | + | ||||
| 19 | + | The harness creates one `O_EXCL` JSONL path per launched role before launch and | |||
| 20 | + | records the assigned process ID, clock domain, role, and path in that run's | |||
| 21 | + | manifest. `raw-harness.jsonl` contains only harness-clock input/terminal-flush | |||
| 22 | + | pairs. Process traces are correlated by IDs, never by timestamps across process | |||
| 23 | + | clock domains. Results with missing pairs, bad order, insufficient samples, or | |||
| 24 | + | cross-process tick arithmetic are invalid. | |||