| 1 | package telemetry |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "io" |
| 8 | "net/http" |
| 9 | "os" |
| 10 | "path/filepath" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "testing" |
| 14 | "time" |
| 15 | ) |
| 16 | |
| 17 | type roundTripFunc func(*http.Request) (*http.Response, error) |
| 18 | |
| 19 | func TestMain(m *testing.M) { |
| 20 | endpoint = "http://127.0.0.1:0/v1" |
| 21 | os.Exit(m.Run()) |
| 22 | } |
| 23 | |
| 24 | func (f roundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) { return f(req) } |
| 25 | |
| 26 | func telemetryResponse(status int) *http.Response { |
| 27 | return &http.Response{ |
| 28 | StatusCode: status, |
| 29 | Body: io.NopCloser(strings.NewReader("")), |
| 30 | Header: make(http.Header), |
| 31 | } |
| 32 | } |
| 33 | |
| 34 | func testClient(home string, transport http.RoundTripper) *Client { |
| 35 | return &Client{ |
| 36 | home: home, |
| 37 | version: "v1.20.0", |
| 38 | installID: strings.Repeat("a", 32), |
| 39 | http: &http.Client{Transport: transport}, |
| 40 | } |
| 41 | } |
| 42 | |
| 43 | func TestInstallIDRepairsMalformedOwnedFile(t *testing.T) { |
| 44 | home := t.TempDir() |
| 45 | path := filepath.Join(home, "cli-telemetry-install-id") |
| 46 | if err := os.WriteFile(path, []byte("truncated\n"), 0o600); err != nil { |
| 47 | t.Fatal(err) |
| 48 | } |
| 49 | |
| 50 | id, err := installID(home) |
| 51 | if err != nil { |
| 52 | t.Fatalf("installID: %v", err) |
| 53 | } |
| 54 | if !validInstallID(id) { |
| 55 | t.Fatalf("repaired install id = %q", id) |
| 56 | } |
| 57 | b, err := os.ReadFile(path) |
| 58 | if err != nil { |
| 59 | t.Fatal(err) |
| 60 | } |
| 61 | if got := strings.TrimSpace(string(b)); got != id { |
| 62 | t.Fatalf("persisted install id = %q, want %q", got, id) |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | func TestDailyPingSendsOnceWithCLISurface(t *testing.T) { |
| 67 | home := t.TempDir() |
| 68 | var mu sync.Mutex |
| 69 | var payloads []pingPayload |
| 70 | client := testClient(home, roundTripFunc(func(req *http.Request) (*http.Response, error) { |
| 71 | if req.URL.String() != endpoint+"/ping" { |
| 72 | t.Fatalf("request URL = %q", req.URL) |
| 73 | } |
| 74 | var payload pingPayload |
| 75 | if err := json.NewDecoder(req.Body).Decode(&payload); err != nil { |
| 76 | t.Fatal(err) |
| 77 | } |
| 78 | mu.Lock() |
| 79 | payloads = append(payloads, payload) |
| 80 | mu.Unlock() |
| 81 | return telemetryResponse(http.StatusAccepted), nil |
| 82 | })) |
| 83 | |
| 84 | if err := client.sendDailyPing(context.Background()); err != nil { |
| 85 | t.Fatal(err) |
| 86 | } |
| 87 | if err := client.sendDailyPing(context.Background()); err != nil { |
| 88 | t.Fatal(err) |
| 89 | } |
| 90 | mu.Lock() |
| 91 | defer mu.Unlock() |
| 92 | if len(payloads) != 1 { |
| 93 | t.Fatalf("ping requests = %d, want 1", len(payloads)) |
| 94 | } |
| 95 | if payloads[0].Surface != "cli" || payloads[0].InstallID != client.installID { |
| 96 | t.Fatalf("ping payload = %+v", payloads[0]) |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | func TestFailedDailyPingRemovesClaimAndRetries(t *testing.T) { |
| 101 | home := t.TempDir() |
| 102 | calls := 0 |
| 103 | client := testClient(home, roundTripFunc(func(*http.Request) (*http.Response, error) { |
| 104 | calls++ |
| 105 | if calls == 1 { |
| 106 | return nil, errors.New("offline") |
| 107 | } |
| 108 | return telemetryResponse(http.StatusAccepted), nil |
| 109 | })) |
| 110 | |
| 111 | if err := client.sendDailyPing(context.Background()); err == nil { |
| 112 | t.Fatal("first ping unexpectedly succeeded") |
| 113 | } |
| 114 | claim := filepath.Join(home, "cli-telemetry-ping-"+time.Now().UTC().Format("2006-01-02")) |
| 115 | if _, err := os.Stat(claim); !errors.Is(err, os.ErrNotExist) { |
| 116 | t.Fatalf("failed ping claim remains: %v", err) |
| 117 | } |
| 118 | if err := client.sendDailyPing(context.Background()); err != nil { |
| 119 | t.Fatalf("retry ping: %v", err) |
| 120 | } |
| 121 | if calls != 2 { |
| 122 | t.Fatalf("ping calls = %d, want 2", calls) |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | func TestFlushPendingAggregatesAndDeletesOnlyAfterSuccess(t *testing.T) { |
| 127 | home := t.TempDir() |
| 128 | for _, counters := range [][]Counter{ |
| 129 | {{Signal: "turns", Bucket: "count", Count: 2}}, |
| 130 | {{Signal: "turns", Bucket: "count", Count: 3}, {Signal: "cli_exit", Bucket: "success", Count: 1}}, |
| 131 | } { |
| 132 | if err := appendPending(home, pendingPayload{Version: "v1.20.0", OS: "android", Counters: counters}); err != nil { |
| 133 | t.Fatal(err) |
| 134 | } |
| 135 | } |
| 136 | requests := 0 |
| 137 | client := testClient(home, roundTripFunc(func(req *http.Request) (*http.Response, error) { |
| 138 | requests++ |
| 139 | var payload metricsPayload |
| 140 | if err := json.NewDecoder(req.Body).Decode(&payload); err != nil { |
| 141 | t.Fatal(err) |
| 142 | } |
| 143 | if payload.Surface != "cli" || payload.OS != "android" { |
| 144 | t.Fatalf("metrics payload = %+v", payload) |
| 145 | } |
| 146 | got := map[string]int{} |
| 147 | for _, counter := range payload.Counters { |
| 148 | got[counter.Signal+"/"+counter.Bucket] = counter.Count |
| 149 | } |
| 150 | if got["turns/count"] != 5 || got["cli_exit/success"] != 1 { |
| 151 | t.Fatalf("aggregated counters = %#v", got) |
| 152 | } |
| 153 | return telemetryResponse(http.StatusAccepted), nil |
| 154 | })) |
| 155 | |
| 156 | if err := client.flushPending(context.Background()); err != nil { |
| 157 | t.Fatal(err) |
| 158 | } |
| 159 | if requests != 1 { |
| 160 | t.Fatalf("metrics requests = %d, want 1", requests) |
| 161 | } |
| 162 | entries, err := os.ReadDir(filepath.Join(home, pendingDirName)) |
| 163 | if err != nil || len(entries) != 0 { |
| 164 | t.Fatalf("pending entries after success = %d, err = %v", len(entries), err) |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | func TestFlushPendingUploadsCompletionMetricsWithoutContent(t *testing.T) { |
| 169 | home := t.TempDir() |
| 170 | const secret = "PRIVATE_TASK_ANSWER_REASON_PATH_MODEL" |
| 171 | want := map[string]string{ |
| 172 | "completion_validation_outcome": "enforce_continue", |
| 173 | "completion_validation_latency": "s_5_15", |
| 174 | "completion_validation_error": "timeout", |
| 175 | "completion_validation_attempt": "repair", |
| 176 | "completion_evaluator_finish_reason": "stop", |
| 177 | "completion_evaluator_cache_hit": "90_100", |
| 178 | } |
| 179 | counters := make([]Counter, 0, len(want)+1) |
| 180 | for signal, bucket := range want { |
| 181 | counters = append(counters, Counter{Signal: signal, Bucket: bucket, Count: 1}) |
| 182 | } |
| 183 | // Even a syntactically safe bucket must be discarded when its signal could |
| 184 | // carry user content. |
| 185 | counters = append(counters, Counter{Signal: "task_text", Bucket: strings.ToLower(secret), Count: 1}) |
| 186 | if err := appendPending(home, pendingPayload{Version: "v1.34.0", OS: "linux", Counters: counters}); err != nil { |
| 187 | t.Fatal(err) |
| 188 | } |
| 189 | |
| 190 | client := testClient(home, roundTripFunc(func(req *http.Request) (*http.Response, error) { |
| 191 | body, err := io.ReadAll(req.Body) |
| 192 | if err != nil { |
| 193 | t.Fatal(err) |
| 194 | } |
| 195 | if strings.Contains(strings.ToUpper(string(body)), secret) { |
| 196 | t.Fatalf("completion metrics upload leaked private content: %s", body) |
| 197 | } |
| 198 | var payload metricsPayload |
| 199 | if err := json.Unmarshal(body, &payload); err != nil { |
| 200 | t.Fatal(err) |
| 201 | } |
| 202 | got := map[string]string{} |
| 203 | for _, counter := range payload.Counters { |
| 204 | got[counter.Signal] = counter.Bucket |
| 205 | } |
| 206 | if len(got) != len(want) { |
| 207 | t.Fatalf("uploaded completion signals = %#v, want %#v", got, want) |
| 208 | } |
| 209 | for signal, bucket := range want { |
| 210 | if got[signal] != bucket { |
| 211 | t.Errorf("%s bucket = %q, want %q", signal, got[signal], bucket) |
| 212 | } |
| 213 | } |
| 214 | return telemetryResponse(http.StatusAccepted), nil |
| 215 | })) |
| 216 | client.version = "v1.34.0" |
| 217 | if err := client.flushPending(context.Background()); err != nil { |
| 218 | t.Fatal(err) |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | func TestFailedFlushRestoresClaimsForRetry(t *testing.T) { |
| 223 | home := t.TempDir() |
| 224 | if err := appendPending(home, pendingPayload{ |
| 225 | Version: "v1.20.0", OS: "linux", Counters: []Counter{{Signal: "turns", Bucket: "count", Count: 1}}, |
| 226 | }); err != nil { |
| 227 | t.Fatal(err) |
| 228 | } |
| 229 | calls := 0 |
| 230 | client := testClient(home, roundTripFunc(func(*http.Request) (*http.Response, error) { |
| 231 | calls++ |
| 232 | if calls == 1 { |
| 233 | return telemetryResponse(http.StatusServiceUnavailable), nil |
| 234 | } |
| 235 | return telemetryResponse(http.StatusAccepted), nil |
| 236 | })) |
| 237 | |
| 238 | if err := client.flushPending(context.Background()); err == nil { |
| 239 | t.Fatal("first flush unexpectedly succeeded") |
| 240 | } |
| 241 | entries, err := os.ReadDir(filepath.Join(home, pendingDirName)) |
| 242 | if err != nil || len(entries) != 1 || !strings.HasSuffix(entries[0].Name(), ".json") { |
| 243 | t.Fatalf("failed flush entries = %v, err = %v", entries, err) |
| 244 | } |
| 245 | if err := client.flushPending(context.Background()); err != nil { |
| 246 | t.Fatalf("retry flush: %v", err) |
| 247 | } |
| 248 | entries, err = os.ReadDir(filepath.Join(home, pendingDirName)) |
| 249 | if err != nil || len(entries) != 0 { |
| 250 | t.Fatalf("pending entries after retry = %d, err = %v", len(entries), err) |
| 251 | } |
| 252 | } |
| 253 | |
| 254 | func TestPendingClaimsAreExclusiveAcrossFlushers(t *testing.T) { |
| 255 | home := t.TempDir() |
| 256 | if err := appendPending(home, pendingPayload{ |
| 257 | Version: "v1.20.0", OS: "linux", Counters: []Counter{{Signal: "turns", Bucket: "count", Count: 1}}, |
| 258 | }); err != nil { |
| 259 | t.Fatal(err) |
| 260 | } |
| 261 | dir := filepath.Join(home, pendingDirName) |
| 262 | first, err := claimPendingFiles(dir, time.Now()) |
| 263 | if err != nil { |
| 264 | t.Fatal(err) |
| 265 | } |
| 266 | second, err := claimPendingFiles(dir, time.Now()) |
| 267 | if err != nil { |
| 268 | t.Fatal(err) |
| 269 | } |
| 270 | if len(first) != 1 || len(second) != 0 { |
| 271 | t.Fatalf("claims: first=%v second=%v", first, second) |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | func TestPendingValidationAcceptsAndroid(t *testing.T) { |
| 276 | if !validPendingPayload(pendingPayload{ |
| 277 | Version: "v1.20.0", OS: "android", Counters: []Counter{{Signal: "turns", Bucket: "count", Count: 1}}, |
| 278 | }) { |
| 279 | t.Fatal("Android CLI payload was rejected") |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | func TestPendingQueueCountsActiveAndRecoversStaleClaims(t *testing.T) { |
| 284 | dir := filepath.Join(t.TempDir(), pendingDirName) |
| 285 | if err := os.MkdirAll(dir, 0o700); err != nil { |
| 286 | t.Fatal(err) |
| 287 | } |
| 288 | for i := range maxPending { |
| 289 | path := filepath.Join(dir, strings.Repeat("a", 16)+"-"+time.Unix(int64(i), 0).Format("150405")+".json.uploading") |
| 290 | if err := os.WriteFile(path, []byte("{}"), 0o600); err != nil { |
| 291 | t.Fatal(err) |
| 292 | } |
| 293 | } |
| 294 | if err := appendPending(filepath.Dir(dir), pendingPayload{ |
| 295 | Version: "v1.20.0", OS: "linux", Counters: []Counter{{Signal: "turns", Bucket: "count", Count: 1}}, |
| 296 | }); err != nil { |
| 297 | t.Fatal(err) |
| 298 | } |
| 299 | entries, err := os.ReadDir(dir) |
| 300 | if err != nil || len(entries) != maxPending { |
| 301 | t.Fatalf("bounded queue entries = %d, err = %v", len(entries), err) |
| 302 | } |
| 303 | |
| 304 | staleDir := filepath.Join(t.TempDir(), pendingDirName) |
| 305 | if err := os.MkdirAll(staleDir, 0o700); err != nil { |
| 306 | t.Fatal(err) |
| 307 | } |
| 308 | staleClaim := filepath.Join(staleDir, "sample.json.uploading") |
| 309 | if err := os.WriteFile(staleClaim, []byte("{}"), 0o600); err != nil { |
| 310 | t.Fatal(err) |
| 311 | } |
| 312 | stale := time.Now().Add(-3 * time.Minute) |
| 313 | if err := os.Chtimes(staleClaim, stale, stale); err != nil { |
| 314 | t.Fatal(err) |
| 315 | } |
| 316 | if !prunePending(staleDir, time.Now()) { |
| 317 | t.Fatal("stale claim recovery did not make a queue slot") |
| 318 | } |
| 319 | if _, err := os.Stat(strings.TrimSuffix(staleClaim, ".uploading")); err != nil { |
| 320 | t.Fatalf("stale claim was not recovered: %v", err) |
| 321 | } |
| 322 | } |
| 323 |