fix(harness): preserve complete diagnostic evidence
- commit
- 1f305e95de96bf19eb4453042da1f8e0de4d0ee7
- author
- brice · · 2026-07-13 09:27 UTC
- parent
- ba3714c
- signature
- GPG Verified
CD88F9E564E0102646FB6077A1A07C7C7C3E349D· arch-gpg Verified for the committer email
7 files · +304 -66
M cmd/vev-perf-harness/evidence.go +8 -2
| @@ -35,7 +35,7 @@ func measuredEventSamples(marks []harnessMark, interval measuredInterval) ([]eve | |||
| 35 | 35 | return nil, errors.New("negative harness duration") | |
| 36 | 36 | } | |
| 37 | 37 | delete(starts, m.Sequence) | |
| 38 | - | if start >= interval.Start && m.Tick <= interval.End { | |
| 38 | + | if start >= interval.Start && m.Tick < interval.End { | |
| 39 | 39 | events = append(events, eventSample{Sequence: m.Sequence, Injected: start, Flushed: m.Tick, Latency: m.Tick - start}) | |
| 40 | 40 | } | |
| 41 | 41 | default: | |
| @@ -283,7 +283,13 @@ func writeJSON(path string, v any) error { | |||
| 283 | 283 | return f.Close() | |
| 284 | 284 | } | |
| 285 | 285 | ||
| 286 | - | func mustJSON(v any) string { b, _ := json.Marshal(v); return string(b) } | |
| 286 | + | func mustJSON(v any) string { | |
| 287 | + | b, err := json.Marshal(v) | |
| 288 | + | if err != nil { | |
| 289 | + | panic(fmt.Sprintf("marshal JSON: %v", err)) | |
| 290 | + | } | |
| 291 | + | return string(b) | |
| 292 | + | } | |
| 287 | 293 | ||
| 288 | 294 | func safeName(s string) string { | |
| 289 | 295 | return strings.Map(func(r rune) rune { | |
| @@ -35,7 +35,7 @@ func measuredEventSamples(marks []harnessMark, interval measuredInterval) ([]eve | |||||
| 35 | return nil, errors.New("negative harness duration") | 35 | return nil, errors.New("negative harness duration") | ||
| 36 | } | 36 | } | ||
| 37 | delete(starts, m.Sequence) | 37 | delete(starts, m.Sequence) | ||
| 38 | - | if start >= interval.Start && m.Tick <= interval.End { | 38 | + | if start >= interval.Start && m.Tick < interval.End { |
| 39 | events = append(events, eventSample{Sequence: m.Sequence, Injected: start, Flushed: m.Tick, Latency: m.Tick - start}) | 39 | events = append(events, eventSample{Sequence: m.Sequence, Injected: start, Flushed: m.Tick, Latency: m.Tick - start}) | ||
| 40 | } | 40 | } | ||
| 41 | default: | 41 | default: | ||
| @@ -283,7 +283,13 @@ func writeJSON(path string, v any) error { | |||||
| 283 | return f.Close() | 283 | return f.Close() | ||
| 284 | } | 284 | } | ||
| 285 | 285 | ||||
| 286 | - | func mustJSON(v any) string { b, _ := json.Marshal(v); return string(b) } | 286 | + | func mustJSON(v any) string { |
| 287 | + | b, err := json.Marshal(v) | |||
| 288 | + | if err != nil { | |||
| 289 | + | panic(fmt.Sprintf("marshal JSON: %v", err)) | |||
| 290 | + | } | |||
| 291 | + | return string(b) | |||
| 292 | + | } | |||
| 287 | 293 | ||||
| 288 | func safeName(s string) string { | 294 | func safeName(s string) string { | ||
| 289 | return strings.Map(func(r rune) rune { | 295 | return strings.Map(func(r rune) rune { | ||
M cmd/vev-perf-harness/evidence_test.go +33 -5
| @@ -3,6 +3,7 @@ package main | |||
| 3 | 3 | import ( | |
| 4 | 4 | "bytes" | |
| 5 | 5 | "encoding/json" | |
| 6 | + | "errors" | |
| 6 | 7 | "os" | |
| 7 | 8 | "path/filepath" | |
| 8 | 9 | "strings" | |
| @@ -99,6 +100,8 @@ func TestHarnessCollectsOnlyCompleteMeasuredIntervalEvents(t *testing.T) { | |||
| 99 | 100 | {Sequence: 3, Kind: "terminal_flushed", Tick: 201, Valid: true}, // exits interval | |
| 100 | 101 | {Sequence: 4, Kind: "input_injected", Tick: 99, Valid: true}, | |
| 101 | 102 | {Sequence: 4, Kind: "terminal_flushed", Tick: 110, Valid: true}, // enters interval | |
| 103 | + | {Sequence: 5, Kind: "input_injected", Tick: 180, Valid: true}, | |
| 104 | + | {Sequence: 5, Kind: "terminal_flushed", Tick: 200, Valid: true}, // end is excluded | |
| 102 | 105 | } | |
| 103 | 106 | events, err := measuredEventSamples(marks, interval) | |
| 104 | 107 | if err != nil { | |
| @@ -132,7 +135,11 @@ func TestHarnessRunRejectsStraddlingEventsAsInsufficient(t *testing.T) { | |||
| 132 | 135 | if err != nil { | |
| 133 | 136 | t.Fatal(err) | |
| 134 | 137 | } | |
| 135 | - | defer raw.Close() | |
| 138 | + | t.Cleanup(func() { | |
| 139 | + | if err := raw.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |
| 140 | + | t.Error(err) | |
| 141 | + | } | |
| 142 | + | }) | |
| 136 | 143 | clock := &fakeClock{} | |
| 137 | 144 | h := defaultHarness() | |
| 138 | 145 | h.clock, h.launcher = clock, insufficientEventLauncher{clock: clock} | |
| @@ -140,6 +147,16 @@ func TestHarnessRunRejectsStraddlingEventsAsInsufficient(t *testing.T) { | |||
| 140 | 147 | if err == nil || !strings.Contains(err.Error(), "insufficient in-interval event samples") { | |
| 141 | 148 | t.Fatalf("straddling run error=%v, want insufficient in-interval sample rejection", err) | |
| 142 | 149 | } | |
| 150 | + | if err := raw.Close(); err != nil { | |
| 151 | + | t.Fatal(err) | |
| 152 | + | } | |
| 153 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |
| 154 | + | if got, want := len(marks), 4; got != want { | |
| 155 | + | t.Fatalf("raw marks=%d, want warmup and straddling pair=%d: %+v", got, want, marks) | |
| 156 | + | } | |
| 157 | + | if straddling := marks[len(marks)-1]; straddling.Sequence != 2 || straddling.Kind != "terminal_flushed" || !straddling.Valid { | |
| 158 | + | t.Fatalf("straddling boundary=%+v", straddling) | |
| 159 | + | } | |
| 143 | 160 | } | |
| 144 | 161 | ||
| 145 | 162 | func TestHarnessCalculatesEventCadenceAndMaxGapFromRawPairs(t *testing.T) { | |
| @@ -186,7 +203,11 @@ func TestHarnessClosesRolesBeforeMergingReceiveSpans(t *testing.T) { | |||
| 186 | 203 | if err != nil { | |
| 187 | 204 | t.Fatal(err) | |
| 188 | 205 | } | |
| 189 | - | defer raw.Close() | |
| 206 | + | t.Cleanup(func() { | |
| 207 | + | if err := raw.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |
| 208 | + | t.Error(err) | |
| 209 | + | } | |
| 210 | + | }) | |
| 190 | 211 | launcher := &closeTraceLauncher{} | |
| 191 | 212 | h := defaultHarness() | |
| 192 | 213 | h.clock, h.launcher = &fakeClock{}, launcher | |
| @@ -282,10 +303,17 @@ func TestHarnessRejectsCrossProcessAndBadTraceSpans(t *testing.T) { | |||
| 282 | 303 | t.Fatal(err) | |
| 283 | 304 | } | |
| 284 | 305 | for _, r := range records { | |
| 285 | - | b, _ := json.Marshal(r) | |
| 286 | - | _, _ = f.Write(append(b, '\n')) | |
| 306 | + | b, err := json.Marshal(r) | |
| 307 | + | if err != nil { | |
| 308 | + | t.Fatal(err) | |
| 309 | + | } | |
| 310 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |
| 311 | + | t.Fatal(err) | |
| 312 | + | } | |
| 313 | + | } | |
| 314 | + | if err := f.Close(); err != nil { | |
| 315 | + | t.Fatal(err) | |
| 287 | 316 | } | |
| 288 | - | _ = f.Close() | |
| 289 | 317 | return processMapping{ProcessID: "one", ClockDomain: "one", TracePath: path, Scenario: "s", Run: 1} | |
| 290 | 318 | } | |
| 291 | 319 | base := func(kind string, tick int64) traceRecord { | |
| @@ -3,6 +3,7 @@ package main | |||||
| 3 | import ( | 3 | import ( | ||
| 4 | "bytes" | 4 | "bytes" | ||
| 5 | "encoding/json" | 5 | "encoding/json" | ||
| 6 | + | "errors" | |||
| 6 | "os" | 7 | "os" | ||
| 7 | "path/filepath" | 8 | "path/filepath" | ||
| 8 | "strings" | 9 | "strings" | ||
| @@ -99,6 +100,8 @@ func TestHarnessCollectsOnlyCompleteMeasuredIntervalEvents(t *testing.T) { | |||||
| 99 | {Sequence: 3, Kind: "terminal_flushed", Tick: 201, Valid: true}, // exits interval | 100 | {Sequence: 3, Kind: "terminal_flushed", Tick: 201, Valid: true}, // exits interval | ||
| 100 | {Sequence: 4, Kind: "input_injected", Tick: 99, Valid: true}, | 101 | {Sequence: 4, Kind: "input_injected", Tick: 99, Valid: true}, | ||
| 101 | {Sequence: 4, Kind: "terminal_flushed", Tick: 110, Valid: true}, // enters interval | 102 | {Sequence: 4, Kind: "terminal_flushed", Tick: 110, Valid: true}, // enters interval | ||
| 103 | + | {Sequence: 5, Kind: "input_injected", Tick: 180, Valid: true}, | |||
| 104 | + | {Sequence: 5, Kind: "terminal_flushed", Tick: 200, Valid: true}, // end is excluded | |||
| 102 | } | 105 | } | ||
| 103 | events, err := measuredEventSamples(marks, interval) | 106 | events, err := measuredEventSamples(marks, interval) | ||
| 104 | if err != nil { | 107 | if err != nil { | ||
| @@ -132,7 +135,11 @@ func TestHarnessRunRejectsStraddlingEventsAsInsufficient(t *testing.T) { | |||||
| 132 | if err != nil { | 135 | if err != nil { | ||
| 133 | t.Fatal(err) | 136 | t.Fatal(err) | ||
| 134 | } | 137 | } | ||
| 135 | - | defer raw.Close() | 138 | + | t.Cleanup(func() { |
| 139 | + | if err := raw.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |||
| 140 | + | t.Error(err) | |||
| 141 | + | } | |||
| 142 | + | }) | |||
| 136 | clock := &fakeClock{} | 143 | clock := &fakeClock{} | ||
| 137 | h := defaultHarness() | 144 | h := defaultHarness() | ||
| 138 | h.clock, h.launcher = clock, insufficientEventLauncher{clock: clock} | 145 | h.clock, h.launcher = clock, insufficientEventLauncher{clock: clock} | ||
| @@ -140,6 +147,16 @@ func TestHarnessRunRejectsStraddlingEventsAsInsufficient(t *testing.T) { | |||||
| 140 | if err == nil || !strings.Contains(err.Error(), "insufficient in-interval event samples") { | 147 | if err == nil || !strings.Contains(err.Error(), "insufficient in-interval event samples") { | ||
| 141 | t.Fatalf("straddling run error=%v, want insufficient in-interval sample rejection", err) | 148 | t.Fatalf("straddling run error=%v, want insufficient in-interval sample rejection", err) | ||
| 142 | } | 149 | } | ||
| 150 | + | if err := raw.Close(); err != nil { | |||
| 151 | + | t.Fatal(err) | |||
| 152 | + | } | |||
| 153 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |||
| 154 | + | if got, want := len(marks), 4; got != want { | |||
| 155 | + | t.Fatalf("raw marks=%d, want warmup and straddling pair=%d: %+v", got, want, marks) | |||
| 156 | + | } | |||
| 157 | + | if straddling := marks[len(marks)-1]; straddling.Sequence != 2 || straddling.Kind != "terminal_flushed" || !straddling.Valid { | |||
| 158 | + | t.Fatalf("straddling boundary=%+v", straddling) | |||
| 159 | + | } | |||
| 143 | } | 160 | } | ||
| 144 | 161 | ||||
| 145 | func TestHarnessCalculatesEventCadenceAndMaxGapFromRawPairs(t *testing.T) { | 162 | func TestHarnessCalculatesEventCadenceAndMaxGapFromRawPairs(t *testing.T) { | ||
| @@ -186,7 +203,11 @@ func TestHarnessClosesRolesBeforeMergingReceiveSpans(t *testing.T) { | |||||
| 186 | if err != nil { | 203 | if err != nil { | ||
| 187 | t.Fatal(err) | 204 | t.Fatal(err) | ||
| 188 | } | 205 | } | ||
| 189 | - | defer raw.Close() | 206 | + | t.Cleanup(func() { |
| 207 | + | if err := raw.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |||
| 208 | + | t.Error(err) | |||
| 209 | + | } | |||
| 210 | + | }) | |||
| 190 | launcher := &closeTraceLauncher{} | 211 | launcher := &closeTraceLauncher{} | ||
| 191 | h := defaultHarness() | 212 | h := defaultHarness() | ||
| 192 | h.clock, h.launcher = &fakeClock{}, launcher | 213 | h.clock, h.launcher = &fakeClock{}, launcher | ||
| @@ -282,10 +303,17 @@ func TestHarnessRejectsCrossProcessAndBadTraceSpans(t *testing.T) { | |||||
| 282 | t.Fatal(err) | 303 | t.Fatal(err) | ||
| 283 | } | 304 | } | ||
| 284 | for _, r := range records { | 305 | for _, r := range records { | ||
| 285 | - | b, _ := json.Marshal(r) | 306 | + | b, err := json.Marshal(r) |
| 286 | - | _, _ = f.Write(append(b, '\n')) | 307 | + | if err != nil { |
| 308 | + | t.Fatal(err) | |||
| 309 | + | } | |||
| 310 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |||
| 311 | + | t.Fatal(err) | |||
| 312 | + | } | |||
| 313 | + | } | |||
| 314 | + | if err := f.Close(); err != nil { | |||
| 315 | + | t.Fatal(err) | |||
| 287 | } | 316 | } | ||
| 288 | - | _ = f.Close() | |||
| 289 | return processMapping{ProcessID: "one", ClockDomain: "one", TracePath: path, Scenario: "s", Run: 1} | 317 | return processMapping{ProcessID: "one", ClockDomain: "one", TracePath: path, Scenario: "s", Run: 1} | ||
| 290 | } | 318 | } | ||
| 291 | base := func(kind string, tick int64) traceRecord { | 319 | base := func(kind string, tick int64) traceRecord { | ||
M cmd/vev-perf-harness/runner.go +42 -24
| @@ -2,6 +2,7 @@ package main | |||
| 2 | 2 | ||
| 3 | 3 | import ( | |
| 4 | 4 | "context" | |
| 5 | + | "encoding/json" | |
| 5 | 6 | "errors" | |
| 6 | 7 | "fmt" | |
| 7 | 8 | "io" | |
| @@ -58,8 +59,7 @@ func createExclusiveRunDir(path string) error { | |||
| 58 | 59 | return err | |
| 59 | 60 | } | |
| 60 | 61 | if err := safedir.EnsurePrivate(path); err != nil { | |
| 61 | - | _ = os.Remove(path) | |
| 62 | - | return err | |
| 62 | + | return errors.Join(err, os.Remove(path)) | |
| 63 | 63 | } | |
| 64 | 64 | return nil | |
| 65 | 65 | } | |
| @@ -175,7 +175,9 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||
| 175 | 175 | success := false | |
| 176 | 176 | defer func() { | |
| 177 | 177 | if !success { | |
| 178 | - | _ = h.removeAll(dir) | |
| 178 | + | if cleanupErr := h.removeAll(dir); cleanupErr != nil { | |
| 179 | + | err = errors.Join(err, fmt.Errorf("remove failed run directory: %w", cleanupErr)) | |
| 180 | + | } | |
| 179 | 181 | } | |
| 180 | 182 | }() | |
| 181 | 183 | maps := make([]processMapping, 0, len(s.Roles)) | |
| @@ -239,22 +241,38 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||
| 239 | 241 | } | |
| 240 | 242 | } | |
| 241 | 243 | } | |
| 244 | + | marks := []harnessMark{} | |
| 245 | + | recordMark := func(mark harnessMark) error { | |
| 246 | + | marks = append(marks, mark) | |
| 247 | + | if err := json.NewEncoder(raw).Encode(mark); err != nil { | |
| 248 | + | return fmt.Errorf("write raw harness mark: %w", err) | |
| 249 | + | } | |
| 250 | + | return nil | |
| 251 | + | } | |
| 252 | + | seq := uint64(0) | |
| 242 | 253 | for _, p := range processes { | |
| 243 | 254 | if p.mapping.Role != "client" { | |
| 244 | 255 | continue | |
| 245 | 256 | } | |
| 246 | - | if err = p.process.Warmup(workloadInput(s, run, "warmup")); err != nil { | |
| 257 | + | seq++ | |
| 258 | + | sequence := seq | |
| 259 | + | if err = recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "input_injected", Tick: h.clock.Now(), Valid: true}); err != nil { | |
| 247 | 260 | return res, err | |
| 248 | 261 | } | |
| 262 | + | warmupErr := p.process.Warmup(workloadInput(s, run, "warmup")) | |
| 263 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: warmupErr == nil}); markErr != nil { | |
| 264 | + | return res, errors.Join(warmupErr, markErr) | |
| 265 | + | } | |
| 266 | + | if warmupErr != nil { | |
| 267 | + | return res, warmupErr | |
| 268 | + | } | |
| 249 | 269 | } | |
| 250 | 270 | // The harness clock, rather than a process clock, owns the explicit warmup | |
| 251 | - | // and measured intervals. Warmup boundaries are intentionally never added to | |
| 252 | - | // marks, and each measured event must complete before interval.End to count. | |
| 271 | + | // and measured intervals. Raw evidence retains every boundary, while only | |
| 272 | + | // complete valid pairs within [Start, End) contribute measurement samples. | |
| 253 | 273 | h.clock.Sleep(o.warmup) | |
| 254 | 274 | interval := measuredInterval{Start: h.clock.Now()} | |
| 255 | 275 | interval.End = interval.Start + o.duration.Nanoseconds() | |
| 256 | - | marks := []harnessMark{} | |
| 257 | - | seq := uint64(0) | |
| 258 | 276 | nextInjection := interval.Start | |
| 259 | 277 | for { | |
| 260 | 278 | now := h.clock.Now() | |
| @@ -271,16 +289,28 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||
| 271 | 289 | } | |
| 272 | 290 | seq++ | |
| 273 | 291 | sequence := seq | |
| 292 | + | injectedRecorded := false | |
| 293 | + | flushedRecorded := false | |
| 274 | 294 | injected := func() error { | |
| 275 | - | marks = append(marks, harnessMark{s.ID, run, sequence, "input_injected", h.clock.Now(), true}) | |
| 295 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "input_injected", Tick: h.clock.Now(), Valid: true}); markErr != nil { | |
| 296 | + | return markErr | |
| 297 | + | } | |
| 298 | + | injectedRecorded = true | |
| 276 | 299 | return nil | |
| 277 | 300 | } | |
| 278 | 301 | flushed := func() error { | |
| 279 | - | marks = append(marks, harnessMark{s.ID, run, sequence, "terminal_flushed", h.clock.Now(), true}) | |
| 302 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: true}); markErr != nil { | |
| 303 | + | return markErr | |
| 304 | + | } | |
| 305 | + | flushedRecorded = true | |
| 280 | 306 | return nil | |
| 281 | 307 | } | |
| 282 | - | if err = p.process.Measure(workloadInput(s, run, fmt.Sprintf("measured-%d", sequence)), injected, flushed); err != nil { | |
| 283 | - | return res, err | |
| 308 | + | if measureErr := p.process.Measure(workloadInput(s, run, fmt.Sprintf("measured-%d", sequence)), injected, flushed); measureErr != nil { | |
| 309 | + | if injectedRecorded && !flushedRecorded { | |
| 310 | + | diagnosticErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: false}) | |
| 311 | + | return res, errors.Join(measureErr, diagnosticErr) | |
| 312 | + | } | |
| 313 | + | return res, measureErr | |
| 284 | 314 | } | |
| 285 | 315 | } | |
| 286 | 316 | // Cap injection rate rather than bursting events after a slow terminal | |
| @@ -308,18 +338,6 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||
| 308 | 338 | if e != nil { | |
| 309 | 339 | return res, e | |
| 310 | 340 | } | |
| 311 | - | inInterval := make(map[uint64]bool, len(events)) | |
| 312 | - | for _, event := range events { | |
| 313 | - | inInterval[event.Sequence] = true | |
| 314 | - | } | |
| 315 | - | for _, m := range marks { | |
| 316 | - | if !inInterval[m.Sequence] { | |
| 317 | - | continue | |
| 318 | - | } | |
| 319 | - | if _, e := fmt.Fprintf(raw, "%s\n", mustJSON(m)); e != nil { | |
| 320 | - | return res, e | |
| 321 | - | } | |
| 322 | - | } | |
| 323 | 341 | success = true | |
| 324 | 342 | return runResult{Spans: spans, Scenario: s.ID, Run: run, Samples: len(samples), EndToEnd: samples, Event: percentiles(samples), Cadence: percentiles(cadenceSamples), CadenceSamples: cadenceSamples, MaxGap: maxGap, Processes: maps}, nil | |
| 325 | 343 | } | |
| @@ -2,6 +2,7 @@ package main | |||||
| 2 | 2 | ||||
| 3 | import ( | 3 | import ( | ||
| 4 | "context" | 4 | "context" | ||
| 5 | + | "encoding/json" | |||
| 5 | "errors" | 6 | "errors" | ||
| 6 | "fmt" | 7 | "fmt" | ||
| 7 | "io" | 8 | "io" | ||
| @@ -58,8 +59,7 @@ func createExclusiveRunDir(path string) error { | |||||
| 58 | return err | 59 | return err | ||
| 59 | } | 60 | } | ||
| 60 | if err := safedir.EnsurePrivate(path); err != nil { | 61 | if err := safedir.EnsurePrivate(path); err != nil { | ||
| 61 | - | _ = os.Remove(path) | 62 | + | return errors.Join(err, os.Remove(path)) |
| 62 | - | return err | |||
| 63 | } | 63 | } | ||
| 64 | return nil | 64 | return nil | ||
| 65 | } | 65 | } | ||
| @@ -175,7 +175,9 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||||
| 175 | success := false | 175 | success := false | ||
| 176 | defer func() { | 176 | defer func() { | ||
| 177 | if !success { | 177 | if !success { | ||
| 178 | - | _ = h.removeAll(dir) | 178 | + | if cleanupErr := h.removeAll(dir); cleanupErr != nil { |
| 179 | + | err = errors.Join(err, fmt.Errorf("remove failed run directory: %w", cleanupErr)) | |||
| 180 | + | } | |||
| 179 | } | 181 | } | ||
| 180 | }() | 182 | }() | ||
| 181 | maps := make([]processMapping, 0, len(s.Roles)) | 183 | maps := make([]processMapping, 0, len(s.Roles)) | ||
| @@ -239,22 +241,38 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||||
| 239 | } | 241 | } | ||
| 240 | } | 242 | } | ||
| 241 | } | 243 | } | ||
| 244 | + | marks := []harnessMark{} | |||
| 245 | + | recordMark := func(mark harnessMark) error { | |||
| 246 | + | marks = append(marks, mark) | |||
| 247 | + | if err := json.NewEncoder(raw).Encode(mark); err != nil { | |||
| 248 | + | return fmt.Errorf("write raw harness mark: %w", err) | |||
| 249 | + | } | |||
| 250 | + | return nil | |||
| 251 | + | } | |||
| 252 | + | seq := uint64(0) | |||
| 242 | for _, p := range processes { | 253 | for _, p := range processes { | ||
| 243 | if p.mapping.Role != "client" { | 254 | if p.mapping.Role != "client" { | ||
| 244 | continue | 255 | continue | ||
| 245 | } | 256 | } | ||
| 246 | - | if err = p.process.Warmup(workloadInput(s, run, "warmup")); err != nil { | 257 | + | seq++ |
| 258 | + | sequence := seq | |||
| 259 | + | if err = recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "input_injected", Tick: h.clock.Now(), Valid: true}); err != nil { | |||
| 247 | return res, err | 260 | return res, err | ||
| 248 | } | 261 | } | ||
| 262 | + | warmupErr := p.process.Warmup(workloadInput(s, run, "warmup")) | |||
| 263 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: warmupErr == nil}); markErr != nil { | |||
| 264 | + | return res, errors.Join(warmupErr, markErr) | |||
| 265 | + | } | |||
| 266 | + | if warmupErr != nil { | |||
| 267 | + | return res, warmupErr | |||
| 268 | + | } | |||
| 249 | } | 269 | } | ||
| 250 | // The harness clock, rather than a process clock, owns the explicit warmup | 270 | // The harness clock, rather than a process clock, owns the explicit warmup | ||
| 251 | - | // and measured intervals. Warmup boundaries are intentionally never added to | 271 | + | // and measured intervals. Raw evidence retains every boundary, while only |
| 252 | - | // marks, and each measured event must complete before interval.End to count. | 272 | + | // complete valid pairs within [Start, End) contribute measurement samples. |
| 253 | h.clock.Sleep(o.warmup) | 273 | h.clock.Sleep(o.warmup) | ||
| 254 | interval := measuredInterval{Start: h.clock.Now()} | 274 | interval := measuredInterval{Start: h.clock.Now()} | ||
| 255 | interval.End = interval.Start + o.duration.Nanoseconds() | 275 | interval.End = interval.Start + o.duration.Nanoseconds() | ||
| 256 | - | marks := []harnessMark{} | |||
| 257 | - | seq := uint64(0) | |||
| 258 | nextInjection := interval.Start | 276 | nextInjection := interval.Start | ||
| 259 | for { | 277 | for { | ||
| 260 | now := h.clock.Now() | 278 | now := h.clock.Now() | ||
| @@ -271,16 +289,28 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||||
| 271 | } | 289 | } | ||
| 272 | seq++ | 290 | seq++ | ||
| 273 | sequence := seq | 291 | sequence := seq | ||
| 292 | + | injectedRecorded := false | |||
| 293 | + | flushedRecorded := false | |||
| 274 | injected := func() error { | 294 | injected := func() error { | ||
| 275 | - | marks = append(marks, harnessMark{s.ID, run, sequence, "input_injected", h.clock.Now(), true}) | 295 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "input_injected", Tick: h.clock.Now(), Valid: true}); markErr != nil { |
| 296 | + | return markErr | |||
| 297 | + | } | |||
| 298 | + | injectedRecorded = true | |||
| 276 | return nil | 299 | return nil | ||
| 277 | } | 300 | } | ||
| 278 | flushed := func() error { | 301 | flushed := func() error { | ||
| 279 | - | marks = append(marks, harnessMark{s.ID, run, sequence, "terminal_flushed", h.clock.Now(), true}) | 302 | + | if markErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: true}); markErr != nil { |
| 303 | + | return markErr | |||
| 304 | + | } | |||
| 305 | + | flushedRecorded = true | |||
| 280 | return nil | 306 | return nil | ||
| 281 | } | 307 | } | ||
| 282 | - | if err = p.process.Measure(workloadInput(s, run, fmt.Sprintf("measured-%d", sequence)), injected, flushed); err != nil { | 308 | + | if measureErr := p.process.Measure(workloadInput(s, run, fmt.Sprintf("measured-%d", sequence)), injected, flushed); measureErr != nil { |
| 283 | - | return res, err | 309 | + | if injectedRecorded && !flushedRecorded { |
| 310 | + | diagnosticErr := recordMark(harnessMark{Scenario: s.ID, Run: run, Sequence: sequence, Kind: "terminal_flushed", Tick: h.clock.Now(), Valid: false}) | |||
| 311 | + | return res, errors.Join(measureErr, diagnosticErr) | |||
| 312 | + | } | |||
| 313 | + | return res, measureErr | |||
| 284 | } | 314 | } | ||
| 285 | } | 315 | } | ||
| 286 | // Cap injection rate rather than bursting events after a slow terminal | 316 | // Cap injection rate rather than bursting events after a slow terminal | ||
| @@ -308,18 +338,6 @@ func (h *harness) runOne(o options, mat manifest, s scenario, run int, raw io.Wr | |||||
| 308 | if e != nil { | 338 | if e != nil { | ||
| 309 | return res, e | 339 | return res, e | ||
| 310 | } | 340 | } | ||
| 311 | - | inInterval := make(map[uint64]bool, len(events)) | |||
| 312 | - | for _, event := range events { | |||
| 313 | - | inInterval[event.Sequence] = true | |||
| 314 | - | } | |||
| 315 | - | for _, m := range marks { | |||
| 316 | - | if !inInterval[m.Sequence] { | |||
| 317 | - | continue | |||
| 318 | - | } | |||
| 319 | - | if _, e := fmt.Fprintf(raw, "%s\n", mustJSON(m)); e != nil { | |||
| 320 | - | return res, e | |||
| 321 | - | } | |||
| 322 | - | } | |||
| 323 | success = true | 341 | success = true | ||
| 324 | return runResult{Spans: spans, Scenario: s.ID, Run: run, Samples: len(samples), EndToEnd: samples, Event: percentiles(samples), Cadence: percentiles(cadenceSamples), CadenceSamples: cadenceSamples, MaxGap: maxGap, Processes: maps}, nil | 342 | return runResult{Spans: spans, Scenario: s.ID, Run: run, Samples: len(samples), EndToEnd: samples, Event: percentiles(samples), Cadence: percentiles(cadenceSamples), CadenceSamples: cadenceSamples, MaxGap: maxGap, Processes: maps}, nil | ||
| 325 | } | 343 | } | ||
M cmd/vev-perf-harness/runner_test.go +67 -10
| @@ -45,12 +45,12 @@ func TestHarnessCanonicalLocalRolesAreIsolatedAcrossRepetitions(t *testing.T) { | |||
| 45 | 45 | if err != nil { | |
| 46 | 46 | t.Fatal(err) | |
| 47 | 47 | } | |
| 48 | - | defer os.RemoveAll(out) | |
| 48 | + | removeTestTree(t, out) | |
| 49 | 49 | raw, err := os.OpenFile(filepath.Join(out, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |
| 50 | 50 | if err != nil { | |
| 51 | 51 | t.Fatal(err) | |
| 52 | 52 | } | |
| 53 | - | defer raw.Close() | |
| 53 | + | closeTestFile(t, raw) | |
| 54 | 54 | h := defaultHarness() | |
| 55 | 55 | h.clock = &fakeClock{} | |
| 56 | 56 | h.launcher = &cliLauncher{bin: bin} | |
| @@ -118,7 +118,7 @@ func TestHarnessResolvesCanonicalRelativePathsBeforeRoleCWDChanges(t *testing.T) | |||
| 118 | 118 | if err != nil { | |
| 119 | 119 | t.Fatal(err) | |
| 120 | 120 | } | |
| 121 | - | t.Cleanup(func() { _ = os.RemoveAll(workspace) }) | |
| 121 | + | removeTestTree(t, workspace) | |
| 122 | 122 | bin := filepath.Join(workspace, "vev") | |
| 123 | 123 | build := exec.Command("go", "build", "-o", bin, "./") | |
| 124 | 124 | build.Dir = root | |
| @@ -198,7 +198,7 @@ func TestHarnessRejectsExistingRunDirectoryBeforeLaunchingRoles(t *testing.T) { | |||
| 198 | 198 | if err != nil { | |
| 199 | 199 | t.Fatal(err) | |
| 200 | 200 | } | |
| 201 | - | defer raw.Close() | |
| 201 | + | closeTestFile(t, raw) | |
| 202 | 202 | s := scenario{ID: "isolated", Roles: []string{"daemon", "client"}} | |
| 203 | 203 | stale := filepath.Join(out, "isolated-run-001") | |
| 204 | 204 | if err := os.Mkdir(stale, 0o700); err != nil { | |
| @@ -282,7 +282,7 @@ func TestHarnessFakeRunnerRoutesClientToPeerAndCleansEveryRole(t *testing.T) { | |||
| 282 | 282 | if err != nil { | |
| 283 | 283 | t.Fatal(err) | |
| 284 | 284 | } | |
| 285 | - | defer raw.Close() | |
| 285 | + | closeTestFile(t, raw) | |
| 286 | 286 | l := &fakeLauncher{} | |
| 287 | 287 | h := defaultHarness() | |
| 288 | 288 | h.clock, h.launcher = &fakeClock{}, l | |
| @@ -320,13 +320,70 @@ func TestHarnessAggregatesRawEventsSeparatelyFromRunDispersion(t *testing.T) { | |||
| 320 | 320 | } | |
| 321 | 321 | } | |
| 322 | 322 | ||
| 323 | + | func TestHarnessRawRecordsContainEveryMarkWhileSamplesRemainFiltered(t *testing.T) { | |
| 324 | + | dir := t.TempDir() | |
| 325 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |
| 326 | + | if err != nil { | |
| 327 | + | t.Fatal(err) | |
| 328 | + | } | |
| 329 | + | h := defaultHarness() | |
| 330 | + | h.clock, h.launcher = &fakeClock{}, &fakeLauncher{} | |
| 331 | + | result, err := h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "raw-complete", Roles: []string{"daemon", "client"}}, 1, raw) | |
| 332 | + | if err != nil { | |
| 333 | + | t.Fatal(err) | |
| 334 | + | } | |
| 335 | + | if err := raw.Close(); err != nil { | |
| 336 | + | t.Fatal(err) | |
| 337 | + | } | |
| 338 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |
| 339 | + | if got, want := len(marks), 2*(result.Samples+1); got != want { | |
| 340 | + | t.Fatalf("raw marks=%d, want warmup plus every measured boundary=%d", got, want) | |
| 341 | + | } | |
| 342 | + | for i, mark := range marks[:2] { | |
| 343 | + | if mark.Sequence != 1 || !mark.Valid || []string{"input_injected", "terminal_flushed"}[i] != mark.Kind { | |
| 344 | + | t.Fatalf("warmup mark %d=%+v", i, mark) | |
| 345 | + | } | |
| 346 | + | } | |
| 347 | + | for sequence := uint64(2); sequence <= uint64(result.Samples+1); sequence++ { | |
| 348 | + | if marks[2*(sequence-1)].Sequence != sequence || marks[2*(sequence-1)+1].Sequence != sequence { | |
| 349 | + | t.Fatalf("raw marks did not retain both boundaries for sequence %d: %+v", sequence, marks) | |
| 350 | + | } | |
| 351 | + | } | |
| 352 | + | } | |
| 353 | + | ||
| 354 | + | func TestHarnessPersistsFailedFlushDiagnostics(t *testing.T) { | |
| 355 | + | dir := t.TempDir() | |
| 356 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |
| 357 | + | if err != nil { | |
| 358 | + | t.Fatal(err) | |
| 359 | + | } | |
| 360 | + | failure := errors.New("flush failed") | |
| 361 | + | h := defaultHarness() | |
| 362 | + | h.clock, h.launcher = &fakeClock{}, failedFlushLauncher{err: failure} | |
| 363 | + | _, err = h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "failed-flush", Roles: []string{"daemon", "client"}}, 1, raw) | |
| 364 | + | if !errors.Is(err, failure) { | |
| 365 | + | t.Fatalf("run error=%v, want failed flush", err) | |
| 366 | + | } | |
| 367 | + | if err := raw.Close(); err != nil { | |
| 368 | + | t.Fatal(err) | |
| 369 | + | } | |
| 370 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |
| 371 | + | if got, want := len(marks), 4; got != want { | |
| 372 | + | t.Fatalf("raw marks=%d, want warmup and failed measured pair=%d: %+v", got, want, marks) | |
| 373 | + | } | |
| 374 | + | failed := marks[len(marks)-1] | |
| 375 | + | if failed.Sequence != 2 || failed.Kind != "terminal_flushed" || failed.Valid { | |
| 376 | + | t.Fatalf("failed flush diagnostic=%+v", failed) | |
| 377 | + | } | |
| 378 | + | } | |
| 379 | + | ||
| 323 | 380 | func TestHarnessSchedulesMeasuredEventsAcrossFullInterval(t *testing.T) { | |
| 324 | 381 | dir := t.TempDir() | |
| 325 | 382 | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |
| 326 | 383 | if err != nil { | |
| 327 | 384 | t.Fatal(err) | |
| 328 | 385 | } | |
| 329 | - | defer raw.Close() | |
| 386 | + | closeTestFile(t, raw) | |
| 330 | 387 | h := defaultHarness() | |
| 331 | 388 | h.clock, h.launcher = &fakeClock{}, &fakeLauncher{} | |
| 332 | 389 | result, err := h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "schedule", Roles: []string{"daemon", "client"}}, 1, raw) | |
| @@ -343,8 +400,8 @@ func TestHarnessSchedulesMeasuredEventsAcrossFullInterval(t *testing.T) { | |||
| 343 | 400 | if err != nil { | |
| 344 | 401 | t.Fatal(err) | |
| 345 | 402 | } | |
| 346 | - | if len(records) != result.Samples*2 { | |
| 347 | - | t.Fatalf("raw records=%d, want %d complete measured boundaries", len(records), result.Samples*2) | |
| 403 | + | if want := result.Samples*2 + 2; len(records) != want { | |
| 404 | + | t.Fatalf("raw records=%d, want %d warmup and measured boundaries", len(records), want) | |
| 348 | 405 | } | |
| 349 | 406 | } | |
| 350 | 407 | ||
| @@ -354,7 +411,7 @@ func TestHarnessPreservesWorkloadAndCleanupErrors(t *testing.T) { | |||
| 354 | 411 | if err != nil { | |
| 355 | 412 | t.Fatal(err) | |
| 356 | 413 | } | |
| 357 | - | defer raw.Close() | |
| 414 | + | closeTestFile(t, raw) | |
| 358 | 415 | primary := errors.New("primary workload failure") | |
| 359 | 416 | cleanup := errors.New("daemon cleanup failure") | |
| 360 | 417 | launcher := &fakeLauncher{ | |
| @@ -383,7 +440,7 @@ func TestHarnessCleansRunDirectoryOnProcessFailure(t *testing.T) { | |||
| 383 | 440 | if err != nil { | |
| 384 | 441 | t.Fatal(err) | |
| 385 | 442 | } | |
| 386 | - | defer raw.Close() | |
| 443 | + | closeTestFile(t, raw) | |
| 387 | 444 | launcher := &fakeLauncher{} | |
| 388 | 445 | h := defaultHarness() | |
| 389 | 446 | h.clock = &fakeClock{} | |
| @@ -45,12 +45,12 @@ func TestHarnessCanonicalLocalRolesAreIsolatedAcrossRepetitions(t *testing.T) { | |||||
| 45 | if err != nil { | 45 | if err != nil { | ||
| 46 | t.Fatal(err) | 46 | t.Fatal(err) | ||
| 47 | } | 47 | } | ||
| 48 | - | defer os.RemoveAll(out) | 48 | + | removeTestTree(t, out) |
| 49 | raw, err := os.OpenFile(filepath.Join(out, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | 49 | raw, err := os.OpenFile(filepath.Join(out, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | ||
| 50 | if err != nil { | 50 | if err != nil { | ||
| 51 | t.Fatal(err) | 51 | t.Fatal(err) | ||
| 52 | } | 52 | } | ||
| 53 | - | defer raw.Close() | 53 | + | closeTestFile(t, raw) |
| 54 | h := defaultHarness() | 54 | h := defaultHarness() | ||
| 55 | h.clock = &fakeClock{} | 55 | h.clock = &fakeClock{} | ||
| 56 | h.launcher = &cliLauncher{bin: bin} | 56 | h.launcher = &cliLauncher{bin: bin} | ||
| @@ -118,7 +118,7 @@ func TestHarnessResolvesCanonicalRelativePathsBeforeRoleCWDChanges(t *testing.T) | |||||
| 118 | if err != nil { | 118 | if err != nil { | ||
| 119 | t.Fatal(err) | 119 | t.Fatal(err) | ||
| 120 | } | 120 | } | ||
| 121 | - | t.Cleanup(func() { _ = os.RemoveAll(workspace) }) | 121 | + | removeTestTree(t, workspace) |
| 122 | bin := filepath.Join(workspace, "vev") | 122 | bin := filepath.Join(workspace, "vev") | ||
| 123 | build := exec.Command("go", "build", "-o", bin, "./") | 123 | build := exec.Command("go", "build", "-o", bin, "./") | ||
| 124 | build.Dir = root | 124 | build.Dir = root | ||
| @@ -198,7 +198,7 @@ func TestHarnessRejectsExistingRunDirectoryBeforeLaunchingRoles(t *testing.T) { | |||||
| 198 | if err != nil { | 198 | if err != nil { | ||
| 199 | t.Fatal(err) | 199 | t.Fatal(err) | ||
| 200 | } | 200 | } | ||
| 201 | - | defer raw.Close() | 201 | + | closeTestFile(t, raw) |
| 202 | s := scenario{ID: "isolated", Roles: []string{"daemon", "client"}} | 202 | s := scenario{ID: "isolated", Roles: []string{"daemon", "client"}} | ||
| 203 | stale := filepath.Join(out, "isolated-run-001") | 203 | stale := filepath.Join(out, "isolated-run-001") | ||
| 204 | if err := os.Mkdir(stale, 0o700); err != nil { | 204 | if err := os.Mkdir(stale, 0o700); err != nil { | ||
| @@ -282,7 +282,7 @@ func TestHarnessFakeRunnerRoutesClientToPeerAndCleansEveryRole(t *testing.T) { | |||||
| 282 | if err != nil { | 282 | if err != nil { | ||
| 283 | t.Fatal(err) | 283 | t.Fatal(err) | ||
| 284 | } | 284 | } | ||
| 285 | - | defer raw.Close() | 285 | + | closeTestFile(t, raw) |
| 286 | l := &fakeLauncher{} | 286 | l := &fakeLauncher{} | ||
| 287 | h := defaultHarness() | 287 | h := defaultHarness() | ||
| 288 | h.clock, h.launcher = &fakeClock{}, l | 288 | h.clock, h.launcher = &fakeClock{}, l | ||
| @@ -320,13 +320,70 @@ func TestHarnessAggregatesRawEventsSeparatelyFromRunDispersion(t *testing.T) { | |||||
| 320 | } | 320 | } | ||
| 321 | } | 321 | } | ||
| 322 | 322 | ||||
| 323 | + | func TestHarnessRawRecordsContainEveryMarkWhileSamplesRemainFiltered(t *testing.T) { | |||
| 324 | + | dir := t.TempDir() | |||
| 325 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |||
| 326 | + | if err != nil { | |||
| 327 | + | t.Fatal(err) | |||
| 328 | + | } | |||
| 329 | + | h := defaultHarness() | |||
| 330 | + | h.clock, h.launcher = &fakeClock{}, &fakeLauncher{} | |||
| 331 | + | result, err := h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "raw-complete", Roles: []string{"daemon", "client"}}, 1, raw) | |||
| 332 | + | if err != nil { | |||
| 333 | + | t.Fatal(err) | |||
| 334 | + | } | |||
| 335 | + | if err := raw.Close(); err != nil { | |||
| 336 | + | t.Fatal(err) | |||
| 337 | + | } | |||
| 338 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |||
| 339 | + | if got, want := len(marks), 2*(result.Samples+1); got != want { | |||
| 340 | + | t.Fatalf("raw marks=%d, want warmup plus every measured boundary=%d", got, want) | |||
| 341 | + | } | |||
| 342 | + | for i, mark := range marks[:2] { | |||
| 343 | + | if mark.Sequence != 1 || !mark.Valid || []string{"input_injected", "terminal_flushed"}[i] != mark.Kind { | |||
| 344 | + | t.Fatalf("warmup mark %d=%+v", i, mark) | |||
| 345 | + | } | |||
| 346 | + | } | |||
| 347 | + | for sequence := uint64(2); sequence <= uint64(result.Samples+1); sequence++ { | |||
| 348 | + | if marks[2*(sequence-1)].Sequence != sequence || marks[2*(sequence-1)+1].Sequence != sequence { | |||
| 349 | + | t.Fatalf("raw marks did not retain both boundaries for sequence %d: %+v", sequence, marks) | |||
| 350 | + | } | |||
| 351 | + | } | |||
| 352 | + | } | |||
| 353 | + | ||||
| 354 | + | func TestHarnessPersistsFailedFlushDiagnostics(t *testing.T) { | |||
| 355 | + | dir := t.TempDir() | |||
| 356 | + | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | |||
| 357 | + | if err != nil { | |||
| 358 | + | t.Fatal(err) | |||
| 359 | + | } | |||
| 360 | + | failure := errors.New("flush failed") | |||
| 361 | + | h := defaultHarness() | |||
| 362 | + | h.clock, h.launcher = &fakeClock{}, failedFlushLauncher{err: failure} | |||
| 363 | + | _, err = h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "failed-flush", Roles: []string{"daemon", "client"}}, 1, raw) | |||
| 364 | + | if !errors.Is(err, failure) { | |||
| 365 | + | t.Fatalf("run error=%v, want failed flush", err) | |||
| 366 | + | } | |||
| 367 | + | if err := raw.Close(); err != nil { | |||
| 368 | + | t.Fatal(err) | |||
| 369 | + | } | |||
| 370 | + | marks := readHarnessMarks(t, filepath.Join(dir, "raw.jsonl")) | |||
| 371 | + | if got, want := len(marks), 4; got != want { | |||
| 372 | + | t.Fatalf("raw marks=%d, want warmup and failed measured pair=%d: %+v", got, want, marks) | |||
| 373 | + | } | |||
| 374 | + | failed := marks[len(marks)-1] | |||
| 375 | + | if failed.Sequence != 2 || failed.Kind != "terminal_flushed" || failed.Valid { | |||
| 376 | + | t.Fatalf("failed flush diagnostic=%+v", failed) | |||
| 377 | + | } | |||
| 378 | + | } | |||
| 379 | + | ||||
| 323 | func TestHarnessSchedulesMeasuredEventsAcrossFullInterval(t *testing.T) { | 380 | func TestHarnessSchedulesMeasuredEventsAcrossFullInterval(t *testing.T) { | ||
| 324 | dir := t.TempDir() | 381 | dir := t.TempDir() | ||
| 325 | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | 382 | raw, err := os.OpenFile(filepath.Join(dir, "raw.jsonl"), os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600) | ||
| 326 | if err != nil { | 383 | if err != nil { | ||
| 327 | t.Fatal(err) | 384 | t.Fatal(err) | ||
| 328 | } | 385 | } | ||
| 329 | - | defer raw.Close() | 386 | + | closeTestFile(t, raw) |
| 330 | h := defaultHarness() | 387 | h := defaultHarness() | ||
| 331 | h.clock, h.launcher = &fakeClock{}, &fakeLauncher{} | 388 | h.clock, h.launcher = &fakeClock{}, &fakeLauncher{} | ||
| 332 | result, err := h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "schedule", Roles: []string{"daemon", "client"}}, 1, raw) | 389 | result, err := h.runOne(options{out: dir, warmup: time.Second, duration: minimumDuration}, manifest{}, scenario{ID: "schedule", Roles: []string{"daemon", "client"}}, 1, raw) | ||
| @@ -343,8 +400,8 @@ func TestHarnessSchedulesMeasuredEventsAcrossFullInterval(t *testing.T) { | |||||
| 343 | if err != nil { | 400 | if err != nil { | ||
| 344 | t.Fatal(err) | 401 | t.Fatal(err) | ||
| 345 | } | 402 | } | ||
| 346 | - | if len(records) != result.Samples*2 { | 403 | + | if want := result.Samples*2 + 2; len(records) != want { |
| 347 | - | t.Fatalf("raw records=%d, want %d complete measured boundaries", len(records), result.Samples*2) | 404 | + | t.Fatalf("raw records=%d, want %d warmup and measured boundaries", len(records), want) |
| 348 | } | 405 | } | ||
| 349 | } | 406 | } | ||
| 350 | 407 | ||||
| @@ -354,7 +411,7 @@ func TestHarnessPreservesWorkloadAndCleanupErrors(t *testing.T) { | |||||
| 354 | if err != nil { | 411 | if err != nil { | ||
| 355 | t.Fatal(err) | 412 | t.Fatal(err) | ||
| 356 | } | 413 | } | ||
| 357 | - | defer raw.Close() | 414 | + | closeTestFile(t, raw) |
| 358 | primary := errors.New("primary workload failure") | 415 | primary := errors.New("primary workload failure") | ||
| 359 | cleanup := errors.New("daemon cleanup failure") | 416 | cleanup := errors.New("daemon cleanup failure") | ||
| 360 | launcher := &fakeLauncher{ | 417 | launcher := &fakeLauncher{ | ||
| @@ -383,7 +440,7 @@ func TestHarnessCleansRunDirectoryOnProcessFailure(t *testing.T) { | |||||
| 383 | if err != nil { | 440 | if err != nil { | ||
| 384 | t.Fatal(err) | 441 | t.Fatal(err) | ||
| 385 | } | 442 | } | ||
| 386 | - | defer raw.Close() | 443 | + | closeTestFile(t, raw) |
| 387 | launcher := &fakeLauncher{} | 444 | launcher := &fakeLauncher{} | ||
| 388 | h := defaultHarness() | 445 | h := defaultHarness() | ||
| 389 | h.clock = &fakeClock{} | 446 | h.clock = &fakeClock{} | ||
M cmd/vev-perf-harness/test_helpers_test.go +84 -9
| @@ -4,8 +4,10 @@ import ( | |||
| 4 | 4 | "bytes" | |
| 5 | 5 | "encoding/json" | |
| 6 | 6 | "errors" | |
| 7 | + | "fmt" | |
| 7 | 8 | "os" | |
| 8 | 9 | "path/filepath" | |
| 10 | + | "testing" | |
| 9 | 11 | "time" | |
| 10 | 12 | ) | |
| 11 | 13 | ||
| @@ -100,6 +102,65 @@ func (l insufficientEventLauncher) Launch(m processMapping, _ roleCommand) (laun | |||
| 100 | 102 | return &fakeProcess{}, nil | |
| 101 | 103 | } | |
| 102 | 104 | ||
| 105 | + | type failedFlushProcess struct{ err error } | |
| 106 | + | ||
| 107 | + | func (*failedFlushProcess) Warmup([]byte) error { return nil } | |
| 108 | + | ||
| 109 | + | func (p *failedFlushProcess) Measure(_ []byte, injected, _ func() error) error { | |
| 110 | + | if err := injected(); err != nil { | |
| 111 | + | return err | |
| 112 | + | } | |
| 113 | + | return p.err | |
| 114 | + | } | |
| 115 | + | ||
| 116 | + | func (*failedFlushProcess) Close() error { return nil } | |
| 117 | + | ||
| 118 | + | type failedFlushLauncher struct{ err error } | |
| 119 | + | ||
| 120 | + | func (l failedFlushLauncher) Launch(m processMapping, _ roleCommand) (launchedProcess, error) { | |
| 121 | + | if m.Role == "client" { | |
| 122 | + | return &failedFlushProcess{err: l.err}, nil | |
| 123 | + | } | |
| 124 | + | return &fakeProcess{}, nil | |
| 125 | + | } | |
| 126 | + | ||
| 127 | + | func closeTestFile(t *testing.T, f *os.File) { | |
| 128 | + | t.Helper() | |
| 129 | + | t.Cleanup(func() { | |
| 130 | + | if err := f.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |
| 131 | + | t.Error(err) | |
| 132 | + | } | |
| 133 | + | }) | |
| 134 | + | } | |
| 135 | + | ||
| 136 | + | func removeTestTree(t *testing.T, path string) { | |
| 137 | + | t.Helper() | |
| 138 | + | t.Cleanup(func() { | |
| 139 | + | if err := os.RemoveAll(path); err != nil { | |
| 140 | + | t.Error(err) | |
| 141 | + | } | |
| 142 | + | }) | |
| 143 | + | } | |
| 144 | + | ||
| 145 | + | func readHarnessMarks(t *testing.T, path string) []harnessMark { | |
| 146 | + | t.Helper() | |
| 147 | + | records, err := readJSONL(path) | |
| 148 | + | if err != nil { | |
| 149 | + | t.Fatal(err) | |
| 150 | + | } | |
| 151 | + | marks := make([]harnessMark, len(records)) | |
| 152 | + | for i, record := range records { | |
| 153 | + | encoded, err := json.Marshal(record) | |
| 154 | + | if err != nil { | |
| 155 | + | t.Fatal(err) | |
| 156 | + | } | |
| 157 | + | if err := json.Unmarshal(encoded, &marks[i]); err != nil { | |
| 158 | + | t.Fatal(err) | |
| 159 | + | } | |
| 160 | + | } | |
| 161 | + | return marks | |
| 162 | + | } | |
| 163 | + | ||
| 103 | 164 | type fakePTY struct{ writes [][]byte } | |
| 104 | 165 | ||
| 105 | 166 | func (p *fakePTY) Read([]byte) (int, error) { return 0, os.ErrClosed } | |
| @@ -175,7 +236,11 @@ func (l *failingLauncher) Launch(m processMapping, command roleCommand) (launche | |||
| 175 | 236 | return nil, err | |
| 176 | 237 | } | |
| 177 | 238 | if m.Role == "client" { | |
| 178 | - | p.(*fakeProcess).fail = true | |
| 239 | + | fake, ok := p.(*fakeProcess) | |
| 240 | + | if !ok { | |
| 241 | + | return nil, fmt.Errorf("fake launcher returned %T, want *fakeProcess", p) | |
| 242 | + | } | |
| 243 | + | fake.fail = true | |
| 179 | 244 | } | |
| 180 | 245 | return p, nil | |
| 181 | 246 | } | |
| @@ -203,15 +268,20 @@ func (l *traceLauncher) Launch(m processMapping, command roleCommand) (launchedP | |||
| 203 | 268 | if err != nil { | |
| 204 | 269 | return nil, err | |
| 205 | 270 | } | |
| 206 | - | defer f.Close() | |
| 207 | 271 | for i, pair := range spanPairs { | |
| 208 | 272 | for n, kind := range []string{pair.start, pair.end} { | |
| 209 | - | 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), Valid: true}) | |
| 273 | + | b, err := 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), Valid: true}) | |
| 274 | + | if err != nil { | |
| 275 | + | return nil, errors.Join(err, f.Close()) | |
| 276 | + | } | |
| 210 | 277 | if _, err := f.Write(append(b, '\n')); err != nil { | |
| 211 | - | return nil, err | |
| 278 | + | return nil, errors.Join(err, f.Close()) | |
| 212 | 279 | } | |
| 213 | 280 | } | |
| 214 | 281 | } | |
| 282 | + | if err := f.Close(); err != nil { | |
| 283 | + | return nil, err | |
| 284 | + | } | |
| 215 | 285 | return p, nil | |
| 216 | 286 | } | |
| 217 | 287 | ||
| @@ -236,7 +306,11 @@ func (l *closeTraceLauncher) Launch(m processMapping, command roleCommand) (laun | |||
| 236 | 306 | if err := appendTraceRecord(m.TracePath, start); err != nil { | |
| 237 | 307 | return nil, err | |
| 238 | 308 | } | |
| 239 | - | return &closeTraceProcess{fakeProcess: p.(*fakeProcess), traceEnd: func() error { | |
| 309 | + | fake, ok := p.(*fakeProcess) | |
| 310 | + | if !ok { | |
| 311 | + | return nil, fmt.Errorf("fake launcher returned %T, want *fakeProcess", p) | |
| 312 | + | } | |
| 313 | + | return &closeTraceProcess{fakeProcess: fake, traceEnd: func() error { | |
| 240 | 314 | return appendTraceRecord(m.TracePath, traceRecord{ProcessID: m.ProcessID, Component: m.Role, Scenario: m.Scenario, Run: uint64(m.Run), Sequence: 1, RequestID: 1, Epoch: 1, Kind: "adapter_receive_end", Tick: 20, Valid: true}) | |
| 241 | 315 | }}, nil | |
| 242 | 316 | } | |
| @@ -246,13 +320,14 @@ func appendTraceRecord(path string, record traceRecord) error { | |||
| 246 | 320 | if err != nil { | |
| 247 | 321 | return err | |
| 248 | 322 | } | |
| 249 | - | defer f.Close() | |
| 250 | 323 | b, err := json.Marshal(record) | |
| 251 | 324 | if err != nil { | |
| 252 | - | return err | |
| 325 | + | return errors.Join(err, f.Close()) | |
| 326 | + | } | |
| 327 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |
| 328 | + | return errors.Join(err, f.Close()) | |
| 253 | 329 | } | |
| 254 | - | _, err = f.Write(append(b, '\n')) | |
| 255 | - | return err | |
| 330 | + | return f.Close() | |
| 256 | 331 | } | |
| 257 | 332 | ||
| 258 | 333 | // boundedLocalConcurrentTrace is an excerpt from the second run of the | |
| @@ -4,8 +4,10 @@ import ( | |||||
| 4 | "bytes" | 4 | "bytes" | ||
| 5 | "encoding/json" | 5 | "encoding/json" | ||
| 6 | "errors" | 6 | "errors" | ||
| 7 | + | "fmt" | |||
| 7 | "os" | 8 | "os" | ||
| 8 | "path/filepath" | 9 | "path/filepath" | ||
| 10 | + | "testing" | |||
| 9 | "time" | 11 | "time" | ||
| 10 | ) | 12 | ) | ||
| 11 | 13 | ||||
| @@ -100,6 +102,65 @@ func (l insufficientEventLauncher) Launch(m processMapping, _ roleCommand) (laun | |||||
| 100 | return &fakeProcess{}, nil | 102 | return &fakeProcess{}, nil | ||
| 101 | } | 103 | } | ||
| 102 | 104 | ||||
| 105 | + | type failedFlushProcess struct{ err error } | |||
| 106 | + | ||||
| 107 | + | func (*failedFlushProcess) Warmup([]byte) error { return nil } | |||
| 108 | + | ||||
| 109 | + | func (p *failedFlushProcess) Measure(_ []byte, injected, _ func() error) error { | |||
| 110 | + | if err := injected(); err != nil { | |||
| 111 | + | return err | |||
| 112 | + | } | |||
| 113 | + | return p.err | |||
| 114 | + | } | |||
| 115 | + | ||||
| 116 | + | func (*failedFlushProcess) Close() error { return nil } | |||
| 117 | + | ||||
| 118 | + | type failedFlushLauncher struct{ err error } | |||
| 119 | + | ||||
| 120 | + | func (l failedFlushLauncher) Launch(m processMapping, _ roleCommand) (launchedProcess, error) { | |||
| 121 | + | if m.Role == "client" { | |||
| 122 | + | return &failedFlushProcess{err: l.err}, nil | |||
| 123 | + | } | |||
| 124 | + | return &fakeProcess{}, nil | |||
| 125 | + | } | |||
| 126 | + | ||||
| 127 | + | func closeTestFile(t *testing.T, f *os.File) { | |||
| 128 | + | t.Helper() | |||
| 129 | + | t.Cleanup(func() { | |||
| 130 | + | if err := f.Close(); err != nil && !errors.Is(err, os.ErrClosed) { | |||
| 131 | + | t.Error(err) | |||
| 132 | + | } | |||
| 133 | + | }) | |||
| 134 | + | } | |||
| 135 | + | ||||
| 136 | + | func removeTestTree(t *testing.T, path string) { | |||
| 137 | + | t.Helper() | |||
| 138 | + | t.Cleanup(func() { | |||
| 139 | + | if err := os.RemoveAll(path); err != nil { | |||
| 140 | + | t.Error(err) | |||
| 141 | + | } | |||
| 142 | + | }) | |||
| 143 | + | } | |||
| 144 | + | ||||
| 145 | + | func readHarnessMarks(t *testing.T, path string) []harnessMark { | |||
| 146 | + | t.Helper() | |||
| 147 | + | records, err := readJSONL(path) | |||
| 148 | + | if err != nil { | |||
| 149 | + | t.Fatal(err) | |||
| 150 | + | } | |||
| 151 | + | marks := make([]harnessMark, len(records)) | |||
| 152 | + | for i, record := range records { | |||
| 153 | + | encoded, err := json.Marshal(record) | |||
| 154 | + | if err != nil { | |||
| 155 | + | t.Fatal(err) | |||
| 156 | + | } | |||
| 157 | + | if err := json.Unmarshal(encoded, &marks[i]); err != nil { | |||
| 158 | + | t.Fatal(err) | |||
| 159 | + | } | |||
| 160 | + | } | |||
| 161 | + | return marks | |||
| 162 | + | } | |||
| 163 | + | ||||
| 103 | type fakePTY struct{ writes [][]byte } | 164 | type fakePTY struct{ writes [][]byte } | ||
| 104 | 165 | ||||
| 105 | func (p *fakePTY) Read([]byte) (int, error) { return 0, os.ErrClosed } | 166 | func (p *fakePTY) Read([]byte) (int, error) { return 0, os.ErrClosed } | ||
| @@ -175,7 +236,11 @@ func (l *failingLauncher) Launch(m processMapping, command roleCommand) (launche | |||||
| 175 | return nil, err | 236 | return nil, err | ||
| 176 | } | 237 | } | ||
| 177 | if m.Role == "client" { | 238 | if m.Role == "client" { | ||
| 178 | - | p.(*fakeProcess).fail = true | 239 | + | fake, ok := p.(*fakeProcess) |
| 240 | + | if !ok { | |||
| 241 | + | return nil, fmt.Errorf("fake launcher returned %T, want *fakeProcess", p) | |||
| 242 | + | } | |||
| 243 | + | fake.fail = true | |||
| 179 | } | 244 | } | ||
| 180 | return p, nil | 245 | return p, nil | ||
| 181 | } | 246 | } | ||
| @@ -203,15 +268,20 @@ func (l *traceLauncher) Launch(m processMapping, command roleCommand) (launchedP | |||||
| 203 | if err != nil { | 268 | if err != nil { | ||
| 204 | return nil, err | 269 | return nil, err | ||
| 205 | } | 270 | } | ||
| 206 | - | defer f.Close() | |||
| 207 | for i, pair := range spanPairs { | 271 | for i, pair := range spanPairs { | ||
| 208 | for n, kind := range []string{pair.start, pair.end} { | 272 | for n, kind := range []string{pair.start, pair.end} { | ||
| 209 | - | 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), Valid: true}) | 273 | + | b, err := 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), Valid: true}) |
| 274 | + | if err != nil { | |||
| 275 | + | return nil, errors.Join(err, f.Close()) | |||
| 276 | + | } | |||
| 210 | if _, err := f.Write(append(b, '\n')); err != nil { | 277 | if _, err := f.Write(append(b, '\n')); err != nil { | ||
| 211 | - | return nil, err | 278 | + | return nil, errors.Join(err, f.Close()) |
| 212 | } | 279 | } | ||
| 213 | } | 280 | } | ||
| 214 | } | 281 | } | ||
| 282 | + | if err := f.Close(); err != nil { | |||
| 283 | + | return nil, err | |||
| 284 | + | } | |||
| 215 | return p, nil | 285 | return p, nil | ||
| 216 | } | 286 | } | ||
| 217 | 287 | ||||
| @@ -236,7 +306,11 @@ func (l *closeTraceLauncher) Launch(m processMapping, command roleCommand) (laun | |||||
| 236 | if err := appendTraceRecord(m.TracePath, start); err != nil { | 306 | if err := appendTraceRecord(m.TracePath, start); err != nil { | ||
| 237 | return nil, err | 307 | return nil, err | ||
| 238 | } | 308 | } | ||
| 239 | - | return &closeTraceProcess{fakeProcess: p.(*fakeProcess), traceEnd: func() error { | 309 | + | fake, ok := p.(*fakeProcess) |
| 310 | + | if !ok { | |||
| 311 | + | return nil, fmt.Errorf("fake launcher returned %T, want *fakeProcess", p) | |||
| 312 | + | } | |||
| 313 | + | return &closeTraceProcess{fakeProcess: fake, traceEnd: func() error { | |||
| 240 | return appendTraceRecord(m.TracePath, traceRecord{ProcessID: m.ProcessID, Component: m.Role, Scenario: m.Scenario, Run: uint64(m.Run), Sequence: 1, RequestID: 1, Epoch: 1, Kind: "adapter_receive_end", Tick: 20, Valid: true}) | 314 | return appendTraceRecord(m.TracePath, traceRecord{ProcessID: m.ProcessID, Component: m.Role, Scenario: m.Scenario, Run: uint64(m.Run), Sequence: 1, RequestID: 1, Epoch: 1, Kind: "adapter_receive_end", Tick: 20, Valid: true}) | ||
| 241 | }}, nil | 315 | }}, nil | ||
| 242 | } | 316 | } | ||
| @@ -246,13 +320,14 @@ func appendTraceRecord(path string, record traceRecord) error { | |||||
| 246 | if err != nil { | 320 | if err != nil { | ||
| 247 | return err | 321 | return err | ||
| 248 | } | 322 | } | ||
| 249 | - | defer f.Close() | |||
| 250 | b, err := json.Marshal(record) | 323 | b, err := json.Marshal(record) | ||
| 251 | if err != nil { | 324 | if err != nil { | ||
| 252 | - | return err | 325 | + | return errors.Join(err, f.Close()) |
| 326 | + | } | |||
| 327 | + | if _, err := f.Write(append(b, '\n')); err != nil { | |||
| 328 | + | return errors.Join(err, f.Close()) | |||
| 253 | } | 329 | } | ||
| 254 | - | _, err = f.Write(append(b, '\n')) | 330 | + | return f.Close() |
| 255 | - | return err | |||
| 256 | } | 331 | } | ||
| 257 | 332 | ||||
| 258 | // boundedLocalConcurrentTrace is an excerpt from the second run of the | 333 | // boundedLocalConcurrentTrace is an excerpt from the second run of the | ||
M cmd/vev-perf-harness/udp_netem.go +38 -12
| @@ -46,7 +46,13 @@ type scheduledPackets []scheduledPacket | |||
| 46 | 46 | func (p scheduledPackets) Len() int { return len(p) } | |
| 47 | 47 | func (p scheduledPackets) Less(i, j int) bool { return p[i].due.Before(p[j].due) } | |
| 48 | 48 | func (p scheduledPackets) Swap(i, j int) { p[i], p[j] = p[j], p[i] } | |
| 49 | - | func (p *scheduledPackets) Push(x any) { *p = append(*p, x.(scheduledPacket)) } | |
| 49 | + | func (p *scheduledPackets) Push(x any) { | |
| 50 | + | packet, ok := x.(scheduledPacket) | |
| 51 | + | if !ok { | |
| 52 | + | panic(fmt.Sprintf("scheduled packet type %T", x)) | |
| 53 | + | } | |
| 54 | + | *p = append(*p, packet) | |
| 55 | + | } | |
| 50 | 56 | func (p *scheduledPackets) Pop() any { | |
| 51 | 57 | old := *p | |
| 52 | 58 | n := len(old) | |
| @@ -56,17 +62,19 @@ func (p *scheduledPackets) Pop() any { | |||
| 56 | 62 | } | |
| 57 | 63 | ||
| 58 | 64 | type udpNetemRelay struct { | |
| 59 | - | config udpNetemConfig | |
| 60 | - | conn net.PacketConn | |
| 61 | - | done chan struct{} | |
| 62 | - | wg sync.WaitGroup | |
| 63 | - | once sync.Once | |
| 65 | + | config udpNetemConfig | |
| 66 | + | conn net.PacketConn | |
| 67 | + | done chan struct{} | |
| 68 | + | wg sync.WaitGroup | |
| 69 | + | once sync.Once | |
| 70 | + | closeErr error | |
| 64 | 71 | ||
| 65 | 72 | // slots bounds queued and deadline-pending packets together. The reader and | |
| 66 | 73 | // the sole scheduler are the relay's only goroutines. | |
| 67 | 74 | slots chan struct{} | |
| 68 | 75 | queue chan scheduledPacket | |
| 69 | 76 | queueOverflowDrops atomic.Uint64 | |
| 77 | + | deliveryErrors atomic.Uint64 | |
| 70 | 78 | ||
| 71 | 79 | mu sync.Mutex | |
| 72 | 80 | client, dst net.Addr | |
| @@ -95,17 +103,26 @@ func newUDPNetem(config udpNetemConfig) (udpNetem, error) { | |||
| 95 | 103 | } | |
| 96 | 104 | ||
| 97 | 105 | func (r *udpNetemRelay) Port() int { | |
| 98 | - | return r.conn.LocalAddr().(*net.UDPAddr).Port | |
| 106 | + | addr, ok := r.conn.LocalAddr().(*net.UDPAddr) | |
| 107 | + | if !ok { | |
| 108 | + | return 0 | |
| 109 | + | } | |
| 110 | + | return addr.Port | |
| 99 | 111 | } | |
| 100 | 112 | ||
| 101 | 113 | func (r *udpNetemRelay) Close() error { | |
| 102 | - | var closeErr error | |
| 103 | 114 | r.once.Do(func() { | |
| 104 | 115 | close(r.done) | |
| 105 | - | closeErr = r.conn.Close() | |
| 116 | + | r.closeErr = r.conn.Close() | |
| 106 | 117 | r.wg.Wait() | |
| 118 | + | if drops := r.queueOverflowDrops.Load(); drops != 0 { | |
| 119 | + | r.closeErr = errors.Join(r.closeErr, fmt.Errorf("udp netem queue overflow drops: %d", drops)) | |
| 120 | + | } | |
| 121 | + | if writes := r.deliveryErrors.Load(); writes != 0 { | |
| 122 | + | r.closeErr = errors.Join(r.closeErr, fmt.Errorf("udp netem delivery errors: %d", writes)) | |
| 123 | + | } | |
| 107 | 124 | }) | |
| 108 | - | return closeErr | |
| 125 | + | return r.closeErr | |
| 109 | 126 | } | |
| 110 | 127 | ||
| 111 | 128 | func (r *udpNetemRelay) run() { | |
| @@ -183,12 +200,21 @@ func (r *udpNetemRelay) schedule() { | |||
| 183 | 200 | case <-timerC: | |
| 184 | 201 | now := time.Now() | |
| 185 | 202 | for pending.Len() > 0 && !pending[0].due.After(now) { | |
| 186 | - | packet := heap.Pop(&pending).(scheduledPacket) | |
| 203 | + | packet, ok := heap.Pop(&pending).(scheduledPacket) | |
| 204 | + | if !ok { | |
| 205 | + | panic("scheduled packet heap contained unexpected type") | |
| 206 | + | } | |
| 187 | 207 | select { | |
| 188 | 208 | case <-r.done: | |
| 189 | 209 | return | |
| 190 | 210 | default: | |
| 191 | - | _, _ = r.conn.WriteTo(packet.packet, packet.to) | |
| 211 | + | if _, err := r.conn.WriteTo(packet.packet, packet.to); err != nil { | |
| 212 | + | select { | |
| 213 | + | case <-r.done: | |
| 214 | + | default: | |
| 215 | + | r.deliveryErrors.Add(1) | |
| 216 | + | } | |
| 217 | + | } | |
| 192 | 218 | <-r.slots | |
| 193 | 219 | } | |
| 194 | 220 | } | |
| @@ -46,7 +46,13 @@ type scheduledPackets []scheduledPacket | |||||
| 46 | func (p scheduledPackets) Len() int { return len(p) } | 46 | func (p scheduledPackets) Len() int { return len(p) } | ||
| 47 | func (p scheduledPackets) Less(i, j int) bool { return p[i].due.Before(p[j].due) } | 47 | func (p scheduledPackets) Less(i, j int) bool { return p[i].due.Before(p[j].due) } | ||
| 48 | func (p scheduledPackets) Swap(i, j int) { p[i], p[j] = p[j], p[i] } | 48 | func (p scheduledPackets) Swap(i, j int) { p[i], p[j] = p[j], p[i] } | ||
| 49 | - | func (p *scheduledPackets) Push(x any) { *p = append(*p, x.(scheduledPacket)) } | 49 | + | func (p *scheduledPackets) Push(x any) { |
| 50 | + | packet, ok := x.(scheduledPacket) | |||
| 51 | + | if !ok { | |||
| 52 | + | panic(fmt.Sprintf("scheduled packet type %T", x)) | |||
| 53 | + | } | |||
| 54 | + | *p = append(*p, packet) | |||
| 55 | + | } | |||
| 50 | func (p *scheduledPackets) Pop() any { | 56 | func (p *scheduledPackets) Pop() any { | ||
| 51 | old := *p | 57 | old := *p | ||
| 52 | n := len(old) | 58 | n := len(old) | ||
| @@ -56,17 +62,19 @@ func (p *scheduledPackets) Pop() any { | |||||
| 56 | } | 62 | } | ||
| 57 | 63 | ||||
| 58 | type udpNetemRelay struct { | 64 | type udpNetemRelay struct { | ||
| 59 | - | config udpNetemConfig | 65 | + | config udpNetemConfig |
| 60 | - | conn net.PacketConn | 66 | + | conn net.PacketConn |
| 61 | - | done chan struct{} | 67 | + | done chan struct{} |
| 62 | - | wg sync.WaitGroup | 68 | + | wg sync.WaitGroup |
| 63 | - | once sync.Once | 69 | + | once sync.Once |
| 70 | + | closeErr error | |||
| 64 | 71 | ||||
| 65 | // slots bounds queued and deadline-pending packets together. The reader and | 72 | // slots bounds queued and deadline-pending packets together. The reader and | ||
| 66 | // the sole scheduler are the relay's only goroutines. | 73 | // the sole scheduler are the relay's only goroutines. | ||
| 67 | slots chan struct{} | 74 | slots chan struct{} | ||
| 68 | queue chan scheduledPacket | 75 | queue chan scheduledPacket | ||
| 69 | queueOverflowDrops atomic.Uint64 | 76 | queueOverflowDrops atomic.Uint64 | ||
| 77 | + | deliveryErrors atomic.Uint64 | |||
| 70 | 78 | ||||
| 71 | mu sync.Mutex | 79 | mu sync.Mutex | ||
| 72 | client, dst net.Addr | 80 | client, dst net.Addr | ||
| @@ -95,17 +103,26 @@ func newUDPNetem(config udpNetemConfig) (udpNetem, error) { | |||||
| 95 | } | 103 | } | ||
| 96 | 104 | ||||
| 97 | func (r *udpNetemRelay) Port() int { | 105 | func (r *udpNetemRelay) Port() int { | ||
| 98 | - | return r.conn.LocalAddr().(*net.UDPAddr).Port | 106 | + | addr, ok := r.conn.LocalAddr().(*net.UDPAddr) |
| 107 | + | if !ok { | |||
| 108 | + | return 0 | |||
| 109 | + | } | |||
| 110 | + | return addr.Port | |||
| 99 | } | 111 | } | ||
| 100 | 112 | ||||
| 101 | func (r *udpNetemRelay) Close() error { | 113 | func (r *udpNetemRelay) Close() error { | ||
| 102 | - | var closeErr error | |||
| 103 | r.once.Do(func() { | 114 | r.once.Do(func() { | ||
| 104 | close(r.done) | 115 | close(r.done) | ||
| 105 | - | closeErr = r.conn.Close() | 116 | + | r.closeErr = r.conn.Close() |
| 106 | r.wg.Wait() | 117 | r.wg.Wait() | ||
| 118 | + | if drops := r.queueOverflowDrops.Load(); drops != 0 { | |||
| 119 | + | r.closeErr = errors.Join(r.closeErr, fmt.Errorf("udp netem queue overflow drops: %d", drops)) | |||
| 120 | + | } | |||
| 121 | + | if writes := r.deliveryErrors.Load(); writes != 0 { | |||
| 122 | + | r.closeErr = errors.Join(r.closeErr, fmt.Errorf("udp netem delivery errors: %d", writes)) | |||
| 123 | + | } | |||
| 107 | }) | 124 | }) | ||
| 108 | - | return closeErr | 125 | + | return r.closeErr |
| 109 | } | 126 | } | ||
| 110 | 127 | ||||
| 111 | func (r *udpNetemRelay) run() { | 128 | func (r *udpNetemRelay) run() { | ||
| @@ -183,12 +200,21 @@ func (r *udpNetemRelay) schedule() { | |||||
| 183 | case <-timerC: | 200 | case <-timerC: | ||
| 184 | now := time.Now() | 201 | now := time.Now() | ||
| 185 | for pending.Len() > 0 && !pending[0].due.After(now) { | 202 | for pending.Len() > 0 && !pending[0].due.After(now) { | ||
| 186 | - | packet := heap.Pop(&pending).(scheduledPacket) | 203 | + | packet, ok := heap.Pop(&pending).(scheduledPacket) |
| 204 | + | if !ok { | |||
| 205 | + | panic("scheduled packet heap contained unexpected type") | |||
| 206 | + | } | |||
| 187 | select { | 207 | select { | ||
| 188 | case <-r.done: | 208 | case <-r.done: | ||
| 189 | return | 209 | return | ||
| 190 | default: | 210 | default: | ||
| 191 | - | _, _ = r.conn.WriteTo(packet.packet, packet.to) | 211 | + | if _, err := r.conn.WriteTo(packet.packet, packet.to); err != nil { |
| 212 | + | select { | |||
| 213 | + | case <-r.done: | |||
| 214 | + | default: | |||
| 215 | + | r.deliveryErrors.Add(1) | |||
| 216 | + | } | |||
| 217 | + | } | |||
| 192 | <-r.slots | 218 | <-r.slots | ||
| 193 | } | 219 | } | ||
| 194 | } | 220 | } | ||
M cmd/vev-perf-harness/udp_netem_test.go +32 -4
| @@ -9,6 +9,15 @@ import ( | |||
| 9 | 9 | "time" | |
| 10 | 10 | ) | |
| 11 | 11 | ||
| 12 | + | func closeUDPFixture(t *testing.T, c interface{ Close() error }) { | |
| 13 | + | t.Helper() | |
| 14 | + | t.Cleanup(func() { | |
| 15 | + | if err := c.Close(); err != nil { | |
| 16 | + | t.Error(err) | |
| 17 | + | } | |
| 18 | + | }) | |
| 19 | + | } | |
| 20 | + | ||
| 12 | 21 | func TestUDPNetemDropsWhenBoundedSchedulerIsFull(t *testing.T) { | |
| 13 | 22 | r := &udpNetemRelay{ | |
| 14 | 23 | done: make(chan struct{}), | |
| @@ -29,6 +38,21 @@ func TestUDPNetemDropsWhenBoundedSchedulerIsFull(t *testing.T) { | |||
| 29 | 38 | } | |
| 30 | 39 | } | |
| 31 | 40 | ||
| 41 | + | func TestUDPNetemCloseRejectsQueueOverflow(t *testing.T) { | |
| 42 | + | netem, err := newUDPNetem(udpNetemConfig{TargetPath: filepath.Join(t.TempDir(), "target")}) | |
| 43 | + | if err != nil { | |
| 44 | + | t.Fatal(err) | |
| 45 | + | } | |
| 46 | + | relay, ok := netem.(*udpNetemRelay) | |
| 47 | + | if !ok { | |
| 48 | + | t.Fatalf("netem type=%T, want *udpNetemRelay", netem) | |
| 49 | + | } | |
| 50 | + | relay.queueOverflowDrops.Store(3) | |
| 51 | + | if err := netem.Close(); err == nil || err.Error() != "udp netem queue overflow drops: 3" { | |
| 52 | + | t.Fatalf("Close error=%v, want overflow rejection", err) | |
| 53 | + | } | |
| 54 | + | } | |
| 55 | + | ||
| 32 | 56 | func TestUDPNetemExecutesRTTAndLossFixtures(t *testing.T) { | |
| 33 | 57 | fixtures := []struct { | |
| 34 | 58 | name string | |
| @@ -48,21 +72,25 @@ func TestUDPNetemExecutesRTTAndLossFixtures(t *testing.T) { | |||
| 48 | 72 | if err != nil { | |
| 49 | 73 | t.Fatal(err) | |
| 50 | 74 | } | |
| 51 | - | defer target.Close() | |
| 75 | + | closeUDPFixture(t, target) | |
| 76 | + | targetAddr, ok := target.LocalAddr().(*net.UDPAddr) | |
| 77 | + | if !ok { | |
| 78 | + | t.Fatalf("target address type=%T, want *net.UDPAddr", target.LocalAddr()) | |
| 79 | + | } | |
| 52 | 80 | path := filepath.Join(t.TempDir(), "target") | |
| 53 | - | if err := os.WriteFile(path, []byte("VEV-UDP "+strconv.Itoa(target.LocalAddr().(*net.UDPAddr).Port)+" key\\n"), 0o600); err != nil { | |
| 81 | + | if err := os.WriteFile(path, []byte("VEV-UDP "+strconv.Itoa(targetAddr.Port)+" key\\n"), 0o600); err != nil { | |
| 54 | 82 | t.Fatal(err) | |
| 55 | 83 | } | |
| 56 | 84 | netem, err := newUDPNetem(udpNetemConfig{RTT: tc.rtt, LossPercent: tc.loss, TargetPath: path}) | |
| 57 | 85 | if err != nil { | |
| 58 | 86 | t.Fatal(err) | |
| 59 | 87 | } | |
| 60 | - | defer netem.Close() | |
| 88 | + | closeUDPFixture(t, netem) | |
| 61 | 89 | client, err := net.ListenPacket("udp", "127.0.0.1:0") | |
| 62 | 90 | if err != nil { | |
| 63 | 91 | t.Fatal(err) | |
| 64 | 92 | } | |
| 65 | - | defer client.Close() | |
| 93 | + | closeUDPFixture(t, client) | |
| 66 | 94 | started := time.Now() | |
| 67 | 95 | for i := 0; i < tc.sent; i++ { | |
| 68 | 96 | if _, err := client.WriteTo([]byte{byte(i)}, &net.UDPAddr{IP: net.ParseIP("127.0.0.1"), Port: netem.Port()}); err != nil { | |
| @@ -9,6 +9,15 @@ import ( | |||||
| 9 | "time" | 9 | "time" | ||
| 10 | ) | 10 | ) | ||
| 11 | 11 | ||||
| 12 | + | func closeUDPFixture(t *testing.T, c interface{ Close() error }) { | |||
| 13 | + | t.Helper() | |||
| 14 | + | t.Cleanup(func() { | |||
| 15 | + | if err := c.Close(); err != nil { | |||
| 16 | + | t.Error(err) | |||
| 17 | + | } | |||
| 18 | + | }) | |||
| 19 | + | } | |||
| 20 | + | ||||
| 12 | func TestUDPNetemDropsWhenBoundedSchedulerIsFull(t *testing.T) { | 21 | func TestUDPNetemDropsWhenBoundedSchedulerIsFull(t *testing.T) { | ||
| 13 | r := &udpNetemRelay{ | 22 | r := &udpNetemRelay{ | ||
| 14 | done: make(chan struct{}), | 23 | done: make(chan struct{}), | ||
| @@ -29,6 +38,21 @@ func TestUDPNetemDropsWhenBoundedSchedulerIsFull(t *testing.T) { | |||||
| 29 | } | 38 | } | ||
| 30 | } | 39 | } | ||
| 31 | 40 | ||||
| 41 | + | func TestUDPNetemCloseRejectsQueueOverflow(t *testing.T) { | |||
| 42 | + | netem, err := newUDPNetem(udpNetemConfig{TargetPath: filepath.Join(t.TempDir(), "target")}) | |||
| 43 | + | if err != nil { | |||
| 44 | + | t.Fatal(err) | |||
| 45 | + | } | |||
| 46 | + | relay, ok := netem.(*udpNetemRelay) | |||
| 47 | + | if !ok { | |||
| 48 | + | t.Fatalf("netem type=%T, want *udpNetemRelay", netem) | |||
| 49 | + | } | |||
| 50 | + | relay.queueOverflowDrops.Store(3) | |||
| 51 | + | if err := netem.Close(); err == nil || err.Error() != "udp netem queue overflow drops: 3" { | |||
| 52 | + | t.Fatalf("Close error=%v, want overflow rejection", err) | |||
| 53 | + | } | |||
| 54 | + | } | |||
| 55 | + | ||||
| 32 | func TestUDPNetemExecutesRTTAndLossFixtures(t *testing.T) { | 56 | func TestUDPNetemExecutesRTTAndLossFixtures(t *testing.T) { | ||
| 33 | fixtures := []struct { | 57 | fixtures := []struct { | ||
| 34 | name string | 58 | name string | ||
| @@ -48,21 +72,25 @@ func TestUDPNetemExecutesRTTAndLossFixtures(t *testing.T) { | |||||
| 48 | if err != nil { | 72 | if err != nil { | ||
| 49 | t.Fatal(err) | 73 | t.Fatal(err) | ||
| 50 | } | 74 | } | ||
| 51 | - | defer target.Close() | 75 | + | closeUDPFixture(t, target) |
| 76 | + | targetAddr, ok := target.LocalAddr().(*net.UDPAddr) | |||
| 77 | + | if !ok { | |||
| 78 | + | t.Fatalf("target address type=%T, want *net.UDPAddr", target.LocalAddr()) | |||
| 79 | + | } | |||
| 52 | path := filepath.Join(t.TempDir(), "target") | 80 | path := filepath.Join(t.TempDir(), "target") | ||
| 53 | - | if err := os.WriteFile(path, []byte("VEV-UDP "+strconv.Itoa(target.LocalAddr().(*net.UDPAddr).Port)+" key\\n"), 0o600); err != nil { | 81 | + | if err := os.WriteFile(path, []byte("VEV-UDP "+strconv.Itoa(targetAddr.Port)+" key\\n"), 0o600); err != nil { |
| 54 | t.Fatal(err) | 82 | t.Fatal(err) | ||
| 55 | } | 83 | } | ||
| 56 | netem, err := newUDPNetem(udpNetemConfig{RTT: tc.rtt, LossPercent: tc.loss, TargetPath: path}) | 84 | netem, err := newUDPNetem(udpNetemConfig{RTT: tc.rtt, LossPercent: tc.loss, TargetPath: path}) | ||
| 57 | if err != nil { | 85 | if err != nil { | ||
| 58 | t.Fatal(err) | 86 | t.Fatal(err) | ||
| 59 | } | 87 | } | ||
| 60 | - | defer netem.Close() | 88 | + | closeUDPFixture(t, netem) |
| 61 | client, err := net.ListenPacket("udp", "127.0.0.1:0") | 89 | client, err := net.ListenPacket("udp", "127.0.0.1:0") | ||
| 62 | if err != nil { | 90 | if err != nil { | ||
| 63 | t.Fatal(err) | 91 | t.Fatal(err) | ||
| 64 | } | 92 | } | ||
| 65 | - | defer client.Close() | 93 | + | closeUDPFixture(t, client) |
| 66 | started := time.Now() | 94 | started := time.Now() | ||
| 67 | for i := 0; i < tc.sent; i++ { | 95 | for i := 0; i < tc.sent; i++ { | ||
| 68 | if _, err := client.WriteTo([]byte{byte(i)}, &net.UDPAddr{IP: net.ParseIP("127.0.0.1"), Port: netem.Port()}); err != nil { | 96 | if _, err := client.WriteTo([]byte{byte(i)}, &net.UDPAddr{IP: net.ParseIP("127.0.0.1"), Port: netem.Port()}); err != nil { | ||