Scott Baker | 2c1c482 | 2019-10-16 11:02:41 -0700 | [diff] [blame^] | 1 | // Copyright 2015 The Go Authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style |
| 3 | // license that can be found in the LICENSE file. |
| 4 | |
| 5 | /* |
| 6 | Package trace implements tracing of requests and long-lived objects. |
| 7 | It exports HTTP interfaces on /debug/requests and /debug/events. |
| 8 | |
| 9 | A trace.Trace provides tracing for short-lived objects, usually requests. |
| 10 | A request handler might be implemented like this: |
| 11 | |
| 12 | func fooHandler(w http.ResponseWriter, req *http.Request) { |
| 13 | tr := trace.New("mypkg.Foo", req.URL.Path) |
| 14 | defer tr.Finish() |
| 15 | ... |
| 16 | tr.LazyPrintf("some event %q happened", str) |
| 17 | ... |
| 18 | if err := somethingImportant(); err != nil { |
| 19 | tr.LazyPrintf("somethingImportant failed: %v", err) |
| 20 | tr.SetError() |
| 21 | } |
| 22 | } |
| 23 | |
| 24 | The /debug/requests HTTP endpoint organizes the traces by family, |
| 25 | errors, and duration. It also provides histogram of request duration |
| 26 | for each family. |
| 27 | |
| 28 | A trace.EventLog provides tracing for long-lived objects, such as RPC |
| 29 | connections. |
| 30 | |
| 31 | // A Fetcher fetches URL paths for a single domain. |
| 32 | type Fetcher struct { |
| 33 | domain string |
| 34 | events trace.EventLog |
| 35 | } |
| 36 | |
| 37 | func NewFetcher(domain string) *Fetcher { |
| 38 | return &Fetcher{ |
| 39 | domain, |
| 40 | trace.NewEventLog("mypkg.Fetcher", domain), |
| 41 | } |
| 42 | } |
| 43 | |
| 44 | func (f *Fetcher) Fetch(path string) (string, error) { |
| 45 | resp, err := http.Get("http://" + f.domain + "/" + path) |
| 46 | if err != nil { |
| 47 | f.events.Errorf("Get(%q) = %v", path, err) |
| 48 | return "", err |
| 49 | } |
| 50 | f.events.Printf("Get(%q) = %s", path, resp.Status) |
| 51 | ... |
| 52 | } |
| 53 | |
| 54 | func (f *Fetcher) Close() error { |
| 55 | f.events.Finish() |
| 56 | return nil |
| 57 | } |
| 58 | |
| 59 | The /debug/events HTTP endpoint organizes the event logs by family and |
| 60 | by time since the last error. The expanded view displays recent log |
| 61 | entries and the log's call stack. |
| 62 | */ |
| 63 | package trace // import "golang.org/x/net/trace" |
| 64 | |
| 65 | import ( |
| 66 | "bytes" |
| 67 | "context" |
| 68 | "fmt" |
| 69 | "html/template" |
| 70 | "io" |
| 71 | "log" |
| 72 | "net" |
| 73 | "net/http" |
| 74 | "net/url" |
| 75 | "runtime" |
| 76 | "sort" |
| 77 | "strconv" |
| 78 | "sync" |
| 79 | "sync/atomic" |
| 80 | "time" |
| 81 | |
| 82 | "golang.org/x/net/internal/timeseries" |
| 83 | ) |
| 84 | |
| 85 | // DebugUseAfterFinish controls whether to debug uses of Trace values after finishing. |
| 86 | // FOR DEBUGGING ONLY. This will slow down the program. |
| 87 | var DebugUseAfterFinish = false |
| 88 | |
| 89 | // HTTP ServeMux paths. |
| 90 | const ( |
| 91 | debugRequestsPath = "/debug/requests" |
| 92 | debugEventsPath = "/debug/events" |
| 93 | ) |
| 94 | |
| 95 | // AuthRequest determines whether a specific request is permitted to load the |
| 96 | // /debug/requests or /debug/events pages. |
| 97 | // |
| 98 | // It returns two bools; the first indicates whether the page may be viewed at all, |
| 99 | // and the second indicates whether sensitive events will be shown. |
| 100 | // |
| 101 | // AuthRequest may be replaced by a program to customize its authorization requirements. |
| 102 | // |
| 103 | // The default AuthRequest function returns (true, true) if and only if the request |
| 104 | // comes from localhost/127.0.0.1/[::1]. |
| 105 | var AuthRequest = func(req *http.Request) (any, sensitive bool) { |
| 106 | // RemoteAddr is commonly in the form "IP" or "IP:port". |
| 107 | // If it is in the form "IP:port", split off the port. |
| 108 | host, _, err := net.SplitHostPort(req.RemoteAddr) |
| 109 | if err != nil { |
| 110 | host = req.RemoteAddr |
| 111 | } |
| 112 | switch host { |
| 113 | case "localhost", "127.0.0.1", "::1": |
| 114 | return true, true |
| 115 | default: |
| 116 | return false, false |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | func init() { |
| 121 | _, pat := http.DefaultServeMux.Handler(&http.Request{URL: &url.URL{Path: debugRequestsPath}}) |
| 122 | if pat == debugRequestsPath { |
| 123 | panic("/debug/requests is already registered. You may have two independent copies of " + |
| 124 | "golang.org/x/net/trace in your binary, trying to maintain separate state. This may " + |
| 125 | "involve a vendored copy of golang.org/x/net/trace.") |
| 126 | } |
| 127 | |
| 128 | // TODO(jbd): Serve Traces from /debug/traces in the future? |
| 129 | // There is no requirement for a request to be present to have traces. |
| 130 | http.HandleFunc(debugRequestsPath, Traces) |
| 131 | http.HandleFunc(debugEventsPath, Events) |
| 132 | } |
| 133 | |
| 134 | // NewContext returns a copy of the parent context |
| 135 | // and associates it with a Trace. |
| 136 | func NewContext(ctx context.Context, tr Trace) context.Context { |
| 137 | return context.WithValue(ctx, contextKey, tr) |
| 138 | } |
| 139 | |
| 140 | // FromContext returns the Trace bound to the context, if any. |
| 141 | func FromContext(ctx context.Context) (tr Trace, ok bool) { |
| 142 | tr, ok = ctx.Value(contextKey).(Trace) |
| 143 | return |
| 144 | } |
| 145 | |
| 146 | // Traces responds with traces from the program. |
| 147 | // The package initialization registers it in http.DefaultServeMux |
| 148 | // at /debug/requests. |
| 149 | // |
| 150 | // It performs authorization by running AuthRequest. |
| 151 | func Traces(w http.ResponseWriter, req *http.Request) { |
| 152 | any, sensitive := AuthRequest(req) |
| 153 | if !any { |
| 154 | http.Error(w, "not allowed", http.StatusUnauthorized) |
| 155 | return |
| 156 | } |
| 157 | w.Header().Set("Content-Type", "text/html; charset=utf-8") |
| 158 | Render(w, req, sensitive) |
| 159 | } |
| 160 | |
| 161 | // Events responds with a page of events collected by EventLogs. |
| 162 | // The package initialization registers it in http.DefaultServeMux |
| 163 | // at /debug/events. |
| 164 | // |
| 165 | // It performs authorization by running AuthRequest. |
| 166 | func Events(w http.ResponseWriter, req *http.Request) { |
| 167 | any, sensitive := AuthRequest(req) |
| 168 | if !any { |
| 169 | http.Error(w, "not allowed", http.StatusUnauthorized) |
| 170 | return |
| 171 | } |
| 172 | w.Header().Set("Content-Type", "text/html; charset=utf-8") |
| 173 | RenderEvents(w, req, sensitive) |
| 174 | } |
| 175 | |
| 176 | // Render renders the HTML page typically served at /debug/requests. |
| 177 | // It does not do any auth checking. The request may be nil. |
| 178 | // |
| 179 | // Most users will use the Traces handler. |
| 180 | func Render(w io.Writer, req *http.Request, sensitive bool) { |
| 181 | data := &struct { |
| 182 | Families []string |
| 183 | ActiveTraceCount map[string]int |
| 184 | CompletedTraces map[string]*family |
| 185 | |
| 186 | // Set when a bucket has been selected. |
| 187 | Traces traceList |
| 188 | Family string |
| 189 | Bucket int |
| 190 | Expanded bool |
| 191 | Traced bool |
| 192 | Active bool |
| 193 | ShowSensitive bool // whether to show sensitive events |
| 194 | |
| 195 | Histogram template.HTML |
| 196 | HistogramWindow string // e.g. "last minute", "last hour", "all time" |
| 197 | |
| 198 | // If non-zero, the set of traces is a partial set, |
| 199 | // and this is the total number. |
| 200 | Total int |
| 201 | }{ |
| 202 | CompletedTraces: completedTraces, |
| 203 | } |
| 204 | |
| 205 | data.ShowSensitive = sensitive |
| 206 | if req != nil { |
| 207 | // Allow show_sensitive=0 to force hiding of sensitive data for testing. |
| 208 | // This only goes one way; you can't use show_sensitive=1 to see things. |
| 209 | if req.FormValue("show_sensitive") == "0" { |
| 210 | data.ShowSensitive = false |
| 211 | } |
| 212 | |
| 213 | if exp, err := strconv.ParseBool(req.FormValue("exp")); err == nil { |
| 214 | data.Expanded = exp |
| 215 | } |
| 216 | if exp, err := strconv.ParseBool(req.FormValue("rtraced")); err == nil { |
| 217 | data.Traced = exp |
| 218 | } |
| 219 | } |
| 220 | |
| 221 | completedMu.RLock() |
| 222 | data.Families = make([]string, 0, len(completedTraces)) |
| 223 | for fam := range completedTraces { |
| 224 | data.Families = append(data.Families, fam) |
| 225 | } |
| 226 | completedMu.RUnlock() |
| 227 | sort.Strings(data.Families) |
| 228 | |
| 229 | // We are careful here to minimize the time spent locking activeMu, |
| 230 | // since that lock is required every time an RPC starts and finishes. |
| 231 | data.ActiveTraceCount = make(map[string]int, len(data.Families)) |
| 232 | activeMu.RLock() |
| 233 | for fam, s := range activeTraces { |
| 234 | data.ActiveTraceCount[fam] = s.Len() |
| 235 | } |
| 236 | activeMu.RUnlock() |
| 237 | |
| 238 | var ok bool |
| 239 | data.Family, data.Bucket, ok = parseArgs(req) |
| 240 | switch { |
| 241 | case !ok: |
| 242 | // No-op |
| 243 | case data.Bucket == -1: |
| 244 | data.Active = true |
| 245 | n := data.ActiveTraceCount[data.Family] |
| 246 | data.Traces = getActiveTraces(data.Family) |
| 247 | if len(data.Traces) < n { |
| 248 | data.Total = n |
| 249 | } |
| 250 | case data.Bucket < bucketsPerFamily: |
| 251 | if b := lookupBucket(data.Family, data.Bucket); b != nil { |
| 252 | data.Traces = b.Copy(data.Traced) |
| 253 | } |
| 254 | default: |
| 255 | if f := getFamily(data.Family, false); f != nil { |
| 256 | var obs timeseries.Observable |
| 257 | f.LatencyMu.RLock() |
| 258 | switch o := data.Bucket - bucketsPerFamily; o { |
| 259 | case 0: |
| 260 | obs = f.Latency.Minute() |
| 261 | data.HistogramWindow = "last minute" |
| 262 | case 1: |
| 263 | obs = f.Latency.Hour() |
| 264 | data.HistogramWindow = "last hour" |
| 265 | case 2: |
| 266 | obs = f.Latency.Total() |
| 267 | data.HistogramWindow = "all time" |
| 268 | } |
| 269 | f.LatencyMu.RUnlock() |
| 270 | if obs != nil { |
| 271 | data.Histogram = obs.(*histogram).html() |
| 272 | } |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | if data.Traces != nil { |
| 277 | defer data.Traces.Free() |
| 278 | sort.Sort(data.Traces) |
| 279 | } |
| 280 | |
| 281 | completedMu.RLock() |
| 282 | defer completedMu.RUnlock() |
| 283 | if err := pageTmpl().ExecuteTemplate(w, "Page", data); err != nil { |
| 284 | log.Printf("net/trace: Failed executing template: %v", err) |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | func parseArgs(req *http.Request) (fam string, b int, ok bool) { |
| 289 | if req == nil { |
| 290 | return "", 0, false |
| 291 | } |
| 292 | fam, bStr := req.FormValue("fam"), req.FormValue("b") |
| 293 | if fam == "" || bStr == "" { |
| 294 | return "", 0, false |
| 295 | } |
| 296 | b, err := strconv.Atoi(bStr) |
| 297 | if err != nil || b < -1 { |
| 298 | return "", 0, false |
| 299 | } |
| 300 | |
| 301 | return fam, b, true |
| 302 | } |
| 303 | |
| 304 | func lookupBucket(fam string, b int) *traceBucket { |
| 305 | f := getFamily(fam, false) |
| 306 | if f == nil || b < 0 || b >= len(f.Buckets) { |
| 307 | return nil |
| 308 | } |
| 309 | return f.Buckets[b] |
| 310 | } |
| 311 | |
| 312 | type contextKeyT string |
| 313 | |
| 314 | var contextKey = contextKeyT("golang.org/x/net/trace.Trace") |
| 315 | |
| 316 | // Trace represents an active request. |
| 317 | type Trace interface { |
| 318 | // LazyLog adds x to the event log. It will be evaluated each time the |
| 319 | // /debug/requests page is rendered. Any memory referenced by x will be |
| 320 | // pinned until the trace is finished and later discarded. |
| 321 | LazyLog(x fmt.Stringer, sensitive bool) |
| 322 | |
| 323 | // LazyPrintf evaluates its arguments with fmt.Sprintf each time the |
| 324 | // /debug/requests page is rendered. Any memory referenced by a will be |
| 325 | // pinned until the trace is finished and later discarded. |
| 326 | LazyPrintf(format string, a ...interface{}) |
| 327 | |
| 328 | // SetError declares that this trace resulted in an error. |
| 329 | SetError() |
| 330 | |
| 331 | // SetRecycler sets a recycler for the trace. |
| 332 | // f will be called for each event passed to LazyLog at a time when |
| 333 | // it is no longer required, whether while the trace is still active |
| 334 | // and the event is discarded, or when a completed trace is discarded. |
| 335 | SetRecycler(f func(interface{})) |
| 336 | |
| 337 | // SetTraceInfo sets the trace info for the trace. |
| 338 | // This is currently unused. |
| 339 | SetTraceInfo(traceID, spanID uint64) |
| 340 | |
| 341 | // SetMaxEvents sets the maximum number of events that will be stored |
| 342 | // in the trace. This has no effect if any events have already been |
| 343 | // added to the trace. |
| 344 | SetMaxEvents(m int) |
| 345 | |
| 346 | // Finish declares that this trace is complete. |
| 347 | // The trace should not be used after calling this method. |
| 348 | Finish() |
| 349 | } |
| 350 | |
| 351 | type lazySprintf struct { |
| 352 | format string |
| 353 | a []interface{} |
| 354 | } |
| 355 | |
| 356 | func (l *lazySprintf) String() string { |
| 357 | return fmt.Sprintf(l.format, l.a...) |
| 358 | } |
| 359 | |
| 360 | // New returns a new Trace with the specified family and title. |
| 361 | func New(family, title string) Trace { |
| 362 | tr := newTrace() |
| 363 | tr.ref() |
| 364 | tr.Family, tr.Title = family, title |
| 365 | tr.Start = time.Now() |
| 366 | tr.maxEvents = maxEventsPerTrace |
| 367 | tr.events = tr.eventsBuf[:0] |
| 368 | |
| 369 | activeMu.RLock() |
| 370 | s := activeTraces[tr.Family] |
| 371 | activeMu.RUnlock() |
| 372 | if s == nil { |
| 373 | activeMu.Lock() |
| 374 | s = activeTraces[tr.Family] // check again |
| 375 | if s == nil { |
| 376 | s = new(traceSet) |
| 377 | activeTraces[tr.Family] = s |
| 378 | } |
| 379 | activeMu.Unlock() |
| 380 | } |
| 381 | s.Add(tr) |
| 382 | |
| 383 | // Trigger allocation of the completed trace structure for this family. |
| 384 | // This will cause the family to be present in the request page during |
| 385 | // the first trace of this family. We don't care about the return value, |
| 386 | // nor is there any need for this to run inline, so we execute it in its |
| 387 | // own goroutine, but only if the family isn't allocated yet. |
| 388 | completedMu.RLock() |
| 389 | if _, ok := completedTraces[tr.Family]; !ok { |
| 390 | go allocFamily(tr.Family) |
| 391 | } |
| 392 | completedMu.RUnlock() |
| 393 | |
| 394 | return tr |
| 395 | } |
| 396 | |
| 397 | func (tr *trace) Finish() { |
| 398 | elapsed := time.Now().Sub(tr.Start) |
| 399 | tr.mu.Lock() |
| 400 | tr.Elapsed = elapsed |
| 401 | tr.mu.Unlock() |
| 402 | |
| 403 | if DebugUseAfterFinish { |
| 404 | buf := make([]byte, 4<<10) // 4 KB should be enough |
| 405 | n := runtime.Stack(buf, false) |
| 406 | tr.finishStack = buf[:n] |
| 407 | } |
| 408 | |
| 409 | activeMu.RLock() |
| 410 | m := activeTraces[tr.Family] |
| 411 | activeMu.RUnlock() |
| 412 | m.Remove(tr) |
| 413 | |
| 414 | f := getFamily(tr.Family, true) |
| 415 | tr.mu.RLock() // protects tr fields in Cond.match calls |
| 416 | for _, b := range f.Buckets { |
| 417 | if b.Cond.match(tr) { |
| 418 | b.Add(tr) |
| 419 | } |
| 420 | } |
| 421 | tr.mu.RUnlock() |
| 422 | |
| 423 | // Add a sample of elapsed time as microseconds to the family's timeseries |
| 424 | h := new(histogram) |
| 425 | h.addMeasurement(elapsed.Nanoseconds() / 1e3) |
| 426 | f.LatencyMu.Lock() |
| 427 | f.Latency.Add(h) |
| 428 | f.LatencyMu.Unlock() |
| 429 | |
| 430 | tr.unref() // matches ref in New |
| 431 | } |
| 432 | |
| 433 | const ( |
| 434 | bucketsPerFamily = 9 |
| 435 | tracesPerBucket = 10 |
| 436 | maxActiveTraces = 20 // Maximum number of active traces to show. |
| 437 | maxEventsPerTrace = 10 |
| 438 | numHistogramBuckets = 38 |
| 439 | ) |
| 440 | |
| 441 | var ( |
| 442 | // The active traces. |
| 443 | activeMu sync.RWMutex |
| 444 | activeTraces = make(map[string]*traceSet) // family -> traces |
| 445 | |
| 446 | // Families of completed traces. |
| 447 | completedMu sync.RWMutex |
| 448 | completedTraces = make(map[string]*family) // family -> traces |
| 449 | ) |
| 450 | |
| 451 | type traceSet struct { |
| 452 | mu sync.RWMutex |
| 453 | m map[*trace]bool |
| 454 | |
| 455 | // We could avoid the entire map scan in FirstN by having a slice of all the traces |
| 456 | // ordered by start time, and an index into that from the trace struct, with a periodic |
| 457 | // repack of the slice after enough traces finish; we could also use a skip list or similar. |
| 458 | // However, that would shift some of the expense from /debug/requests time to RPC time, |
| 459 | // which is probably the wrong trade-off. |
| 460 | } |
| 461 | |
| 462 | func (ts *traceSet) Len() int { |
| 463 | ts.mu.RLock() |
| 464 | defer ts.mu.RUnlock() |
| 465 | return len(ts.m) |
| 466 | } |
| 467 | |
| 468 | func (ts *traceSet) Add(tr *trace) { |
| 469 | ts.mu.Lock() |
| 470 | if ts.m == nil { |
| 471 | ts.m = make(map[*trace]bool) |
| 472 | } |
| 473 | ts.m[tr] = true |
| 474 | ts.mu.Unlock() |
| 475 | } |
| 476 | |
| 477 | func (ts *traceSet) Remove(tr *trace) { |
| 478 | ts.mu.Lock() |
| 479 | delete(ts.m, tr) |
| 480 | ts.mu.Unlock() |
| 481 | } |
| 482 | |
| 483 | // FirstN returns the first n traces ordered by time. |
| 484 | func (ts *traceSet) FirstN(n int) traceList { |
| 485 | ts.mu.RLock() |
| 486 | defer ts.mu.RUnlock() |
| 487 | |
| 488 | if n > len(ts.m) { |
| 489 | n = len(ts.m) |
| 490 | } |
| 491 | trl := make(traceList, 0, n) |
| 492 | |
| 493 | // Fast path for when no selectivity is needed. |
| 494 | if n == len(ts.m) { |
| 495 | for tr := range ts.m { |
| 496 | tr.ref() |
| 497 | trl = append(trl, tr) |
| 498 | } |
| 499 | sort.Sort(trl) |
| 500 | return trl |
| 501 | } |
| 502 | |
| 503 | // Pick the oldest n traces. |
| 504 | // This is inefficient. See the comment in the traceSet struct. |
| 505 | for tr := range ts.m { |
| 506 | // Put the first n traces into trl in the order they occur. |
| 507 | // When we have n, sort trl, and thereafter maintain its order. |
| 508 | if len(trl) < n { |
| 509 | tr.ref() |
| 510 | trl = append(trl, tr) |
| 511 | if len(trl) == n { |
| 512 | // This is guaranteed to happen exactly once during this loop. |
| 513 | sort.Sort(trl) |
| 514 | } |
| 515 | continue |
| 516 | } |
| 517 | if tr.Start.After(trl[n-1].Start) { |
| 518 | continue |
| 519 | } |
| 520 | |
| 521 | // Find where to insert this one. |
| 522 | tr.ref() |
| 523 | i := sort.Search(n, func(i int) bool { return trl[i].Start.After(tr.Start) }) |
| 524 | trl[n-1].unref() |
| 525 | copy(trl[i+1:], trl[i:]) |
| 526 | trl[i] = tr |
| 527 | } |
| 528 | |
| 529 | return trl |
| 530 | } |
| 531 | |
| 532 | func getActiveTraces(fam string) traceList { |
| 533 | activeMu.RLock() |
| 534 | s := activeTraces[fam] |
| 535 | activeMu.RUnlock() |
| 536 | if s == nil { |
| 537 | return nil |
| 538 | } |
| 539 | return s.FirstN(maxActiveTraces) |
| 540 | } |
| 541 | |
| 542 | func getFamily(fam string, allocNew bool) *family { |
| 543 | completedMu.RLock() |
| 544 | f := completedTraces[fam] |
| 545 | completedMu.RUnlock() |
| 546 | if f == nil && allocNew { |
| 547 | f = allocFamily(fam) |
| 548 | } |
| 549 | return f |
| 550 | } |
| 551 | |
| 552 | func allocFamily(fam string) *family { |
| 553 | completedMu.Lock() |
| 554 | defer completedMu.Unlock() |
| 555 | f := completedTraces[fam] |
| 556 | if f == nil { |
| 557 | f = newFamily() |
| 558 | completedTraces[fam] = f |
| 559 | } |
| 560 | return f |
| 561 | } |
| 562 | |
| 563 | // family represents a set of trace buckets and associated latency information. |
| 564 | type family struct { |
| 565 | // traces may occur in multiple buckets. |
| 566 | Buckets [bucketsPerFamily]*traceBucket |
| 567 | |
| 568 | // latency time series |
| 569 | LatencyMu sync.RWMutex |
| 570 | Latency *timeseries.MinuteHourSeries |
| 571 | } |
| 572 | |
| 573 | func newFamily() *family { |
| 574 | return &family{ |
| 575 | Buckets: [bucketsPerFamily]*traceBucket{ |
| 576 | {Cond: minCond(0)}, |
| 577 | {Cond: minCond(50 * time.Millisecond)}, |
| 578 | {Cond: minCond(100 * time.Millisecond)}, |
| 579 | {Cond: minCond(200 * time.Millisecond)}, |
| 580 | {Cond: minCond(500 * time.Millisecond)}, |
| 581 | {Cond: minCond(1 * time.Second)}, |
| 582 | {Cond: minCond(10 * time.Second)}, |
| 583 | {Cond: minCond(100 * time.Second)}, |
| 584 | {Cond: errorCond{}}, |
| 585 | }, |
| 586 | Latency: timeseries.NewMinuteHourSeries(func() timeseries.Observable { return new(histogram) }), |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | // traceBucket represents a size-capped bucket of historic traces, |
| 591 | // along with a condition for a trace to belong to the bucket. |
| 592 | type traceBucket struct { |
| 593 | Cond cond |
| 594 | |
| 595 | // Ring buffer implementation of a fixed-size FIFO queue. |
| 596 | mu sync.RWMutex |
| 597 | buf [tracesPerBucket]*trace |
| 598 | start int // < tracesPerBucket |
| 599 | length int // <= tracesPerBucket |
| 600 | } |
| 601 | |
| 602 | func (b *traceBucket) Add(tr *trace) { |
| 603 | b.mu.Lock() |
| 604 | defer b.mu.Unlock() |
| 605 | |
| 606 | i := b.start + b.length |
| 607 | if i >= tracesPerBucket { |
| 608 | i -= tracesPerBucket |
| 609 | } |
| 610 | if b.length == tracesPerBucket { |
| 611 | // "Remove" an element from the bucket. |
| 612 | b.buf[i].unref() |
| 613 | b.start++ |
| 614 | if b.start == tracesPerBucket { |
| 615 | b.start = 0 |
| 616 | } |
| 617 | } |
| 618 | b.buf[i] = tr |
| 619 | if b.length < tracesPerBucket { |
| 620 | b.length++ |
| 621 | } |
| 622 | tr.ref() |
| 623 | } |
| 624 | |
| 625 | // Copy returns a copy of the traces in the bucket. |
| 626 | // If tracedOnly is true, only the traces with trace information will be returned. |
| 627 | // The logs will be ref'd before returning; the caller should call |
| 628 | // the Free method when it is done with them. |
| 629 | // TODO(dsymonds): keep track of traced requests in separate buckets. |
| 630 | func (b *traceBucket) Copy(tracedOnly bool) traceList { |
| 631 | b.mu.RLock() |
| 632 | defer b.mu.RUnlock() |
| 633 | |
| 634 | trl := make(traceList, 0, b.length) |
| 635 | for i, x := 0, b.start; i < b.length; i++ { |
| 636 | tr := b.buf[x] |
| 637 | if !tracedOnly || tr.spanID != 0 { |
| 638 | tr.ref() |
| 639 | trl = append(trl, tr) |
| 640 | } |
| 641 | x++ |
| 642 | if x == b.length { |
| 643 | x = 0 |
| 644 | } |
| 645 | } |
| 646 | return trl |
| 647 | } |
| 648 | |
| 649 | func (b *traceBucket) Empty() bool { |
| 650 | b.mu.RLock() |
| 651 | defer b.mu.RUnlock() |
| 652 | return b.length == 0 |
| 653 | } |
| 654 | |
| 655 | // cond represents a condition on a trace. |
| 656 | type cond interface { |
| 657 | match(t *trace) bool |
| 658 | String() string |
| 659 | } |
| 660 | |
| 661 | type minCond time.Duration |
| 662 | |
| 663 | func (m minCond) match(t *trace) bool { return t.Elapsed >= time.Duration(m) } |
| 664 | func (m minCond) String() string { return fmt.Sprintf("≥%gs", time.Duration(m).Seconds()) } |
| 665 | |
| 666 | type errorCond struct{} |
| 667 | |
| 668 | func (e errorCond) match(t *trace) bool { return t.IsError } |
| 669 | func (e errorCond) String() string { return "errors" } |
| 670 | |
| 671 | type traceList []*trace |
| 672 | |
| 673 | // Free calls unref on each element of the list. |
| 674 | func (trl traceList) Free() { |
| 675 | for _, t := range trl { |
| 676 | t.unref() |
| 677 | } |
| 678 | } |
| 679 | |
| 680 | // traceList may be sorted in reverse chronological order. |
| 681 | func (trl traceList) Len() int { return len(trl) } |
| 682 | func (trl traceList) Less(i, j int) bool { return trl[i].Start.After(trl[j].Start) } |
| 683 | func (trl traceList) Swap(i, j int) { trl[i], trl[j] = trl[j], trl[i] } |
| 684 | |
| 685 | // An event is a timestamped log entry in a trace. |
| 686 | type event struct { |
| 687 | When time.Time |
| 688 | Elapsed time.Duration // since previous event in trace |
| 689 | NewDay bool // whether this event is on a different day to the previous event |
| 690 | Recyclable bool // whether this event was passed via LazyLog |
| 691 | Sensitive bool // whether this event contains sensitive information |
| 692 | What interface{} // string or fmt.Stringer |
| 693 | } |
| 694 | |
| 695 | // WhenString returns a string representation of the elapsed time of the event. |
| 696 | // It will include the date if midnight was crossed. |
| 697 | func (e event) WhenString() string { |
| 698 | if e.NewDay { |
| 699 | return e.When.Format("2006/01/02 15:04:05.000000") |
| 700 | } |
| 701 | return e.When.Format("15:04:05.000000") |
| 702 | } |
| 703 | |
| 704 | // discarded represents a number of discarded events. |
| 705 | // It is stored as *discarded to make it easier to update in-place. |
| 706 | type discarded int |
| 707 | |
| 708 | func (d *discarded) String() string { |
| 709 | return fmt.Sprintf("(%d events discarded)", int(*d)) |
| 710 | } |
| 711 | |
| 712 | // trace represents an active or complete request, |
| 713 | // either sent or received by this program. |
| 714 | type trace struct { |
| 715 | // Family is the top-level grouping of traces to which this belongs. |
| 716 | Family string |
| 717 | |
| 718 | // Title is the title of this trace. |
| 719 | Title string |
| 720 | |
| 721 | // Start time of the this trace. |
| 722 | Start time.Time |
| 723 | |
| 724 | mu sync.RWMutex |
| 725 | events []event // Append-only sequence of events (modulo discards). |
| 726 | maxEvents int |
| 727 | recycler func(interface{}) |
| 728 | IsError bool // Whether this trace resulted in an error. |
| 729 | Elapsed time.Duration // Elapsed time for this trace, zero while active. |
| 730 | traceID uint64 // Trace information if non-zero. |
| 731 | spanID uint64 |
| 732 | |
| 733 | refs int32 // how many buckets this is in |
| 734 | disc discarded // scratch space to avoid allocation |
| 735 | |
| 736 | finishStack []byte // where finish was called, if DebugUseAfterFinish is set |
| 737 | |
| 738 | eventsBuf [4]event // preallocated buffer in case we only log a few events |
| 739 | } |
| 740 | |
| 741 | func (tr *trace) reset() { |
| 742 | // Clear all but the mutex. Mutexes may not be copied, even when unlocked. |
| 743 | tr.Family = "" |
| 744 | tr.Title = "" |
| 745 | tr.Start = time.Time{} |
| 746 | |
| 747 | tr.mu.Lock() |
| 748 | tr.Elapsed = 0 |
| 749 | tr.traceID = 0 |
| 750 | tr.spanID = 0 |
| 751 | tr.IsError = false |
| 752 | tr.maxEvents = 0 |
| 753 | tr.events = nil |
| 754 | tr.recycler = nil |
| 755 | tr.mu.Unlock() |
| 756 | |
| 757 | tr.refs = 0 |
| 758 | tr.disc = 0 |
| 759 | tr.finishStack = nil |
| 760 | for i := range tr.eventsBuf { |
| 761 | tr.eventsBuf[i] = event{} |
| 762 | } |
| 763 | } |
| 764 | |
| 765 | // delta returns the elapsed time since the last event or the trace start, |
| 766 | // and whether it spans midnight. |
| 767 | // L >= tr.mu |
| 768 | func (tr *trace) delta(t time.Time) (time.Duration, bool) { |
| 769 | if len(tr.events) == 0 { |
| 770 | return t.Sub(tr.Start), false |
| 771 | } |
| 772 | prev := tr.events[len(tr.events)-1].When |
| 773 | return t.Sub(prev), prev.Day() != t.Day() |
| 774 | } |
| 775 | |
| 776 | func (tr *trace) addEvent(x interface{}, recyclable, sensitive bool) { |
| 777 | if DebugUseAfterFinish && tr.finishStack != nil { |
| 778 | buf := make([]byte, 4<<10) // 4 KB should be enough |
| 779 | n := runtime.Stack(buf, false) |
| 780 | log.Printf("net/trace: trace used after finish:\nFinished at:\n%s\nUsed at:\n%s", tr.finishStack, buf[:n]) |
| 781 | } |
| 782 | |
| 783 | /* |
| 784 | NOTE TO DEBUGGERS |
| 785 | |
| 786 | If you are here because your program panicked in this code, |
| 787 | it is almost definitely the fault of code using this package, |
| 788 | and very unlikely to be the fault of this code. |
| 789 | |
| 790 | The most likely scenario is that some code elsewhere is using |
| 791 | a trace.Trace after its Finish method is called. |
| 792 | You can temporarily set the DebugUseAfterFinish var |
| 793 | to help discover where that is; do not leave that var set, |
| 794 | since it makes this package much less efficient. |
| 795 | */ |
| 796 | |
| 797 | e := event{When: time.Now(), What: x, Recyclable: recyclable, Sensitive: sensitive} |
| 798 | tr.mu.Lock() |
| 799 | e.Elapsed, e.NewDay = tr.delta(e.When) |
| 800 | if len(tr.events) < tr.maxEvents { |
| 801 | tr.events = append(tr.events, e) |
| 802 | } else { |
| 803 | // Discard the middle events. |
| 804 | di := int((tr.maxEvents - 1) / 2) |
| 805 | if d, ok := tr.events[di].What.(*discarded); ok { |
| 806 | (*d)++ |
| 807 | } else { |
| 808 | // disc starts at two to count for the event it is replacing, |
| 809 | // plus the next one that we are about to drop. |
| 810 | tr.disc = 2 |
| 811 | if tr.recycler != nil && tr.events[di].Recyclable { |
| 812 | go tr.recycler(tr.events[di].What) |
| 813 | } |
| 814 | tr.events[di].What = &tr.disc |
| 815 | } |
| 816 | // The timestamp of the discarded meta-event should be |
| 817 | // the time of the last event it is representing. |
| 818 | tr.events[di].When = tr.events[di+1].When |
| 819 | |
| 820 | if tr.recycler != nil && tr.events[di+1].Recyclable { |
| 821 | go tr.recycler(tr.events[di+1].What) |
| 822 | } |
| 823 | copy(tr.events[di+1:], tr.events[di+2:]) |
| 824 | tr.events[tr.maxEvents-1] = e |
| 825 | } |
| 826 | tr.mu.Unlock() |
| 827 | } |
| 828 | |
| 829 | func (tr *trace) LazyLog(x fmt.Stringer, sensitive bool) { |
| 830 | tr.addEvent(x, true, sensitive) |
| 831 | } |
| 832 | |
| 833 | func (tr *trace) LazyPrintf(format string, a ...interface{}) { |
| 834 | tr.addEvent(&lazySprintf{format, a}, false, false) |
| 835 | } |
| 836 | |
| 837 | func (tr *trace) SetError() { |
| 838 | tr.mu.Lock() |
| 839 | tr.IsError = true |
| 840 | tr.mu.Unlock() |
| 841 | } |
| 842 | |
| 843 | func (tr *trace) SetRecycler(f func(interface{})) { |
| 844 | tr.mu.Lock() |
| 845 | tr.recycler = f |
| 846 | tr.mu.Unlock() |
| 847 | } |
| 848 | |
| 849 | func (tr *trace) SetTraceInfo(traceID, spanID uint64) { |
| 850 | tr.mu.Lock() |
| 851 | tr.traceID, tr.spanID = traceID, spanID |
| 852 | tr.mu.Unlock() |
| 853 | } |
| 854 | |
| 855 | func (tr *trace) SetMaxEvents(m int) { |
| 856 | tr.mu.Lock() |
| 857 | // Always keep at least three events: first, discarded count, last. |
| 858 | if len(tr.events) == 0 && m > 3 { |
| 859 | tr.maxEvents = m |
| 860 | } |
| 861 | tr.mu.Unlock() |
| 862 | } |
| 863 | |
| 864 | func (tr *trace) ref() { |
| 865 | atomic.AddInt32(&tr.refs, 1) |
| 866 | } |
| 867 | |
| 868 | func (tr *trace) unref() { |
| 869 | if atomic.AddInt32(&tr.refs, -1) == 0 { |
| 870 | tr.mu.RLock() |
| 871 | if tr.recycler != nil { |
| 872 | // freeTrace clears tr, so we hold tr.recycler and tr.events here. |
| 873 | go func(f func(interface{}), es []event) { |
| 874 | for _, e := range es { |
| 875 | if e.Recyclable { |
| 876 | f(e.What) |
| 877 | } |
| 878 | } |
| 879 | }(tr.recycler, tr.events) |
| 880 | } |
| 881 | tr.mu.RUnlock() |
| 882 | |
| 883 | freeTrace(tr) |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | func (tr *trace) When() string { |
| 888 | return tr.Start.Format("2006/01/02 15:04:05.000000") |
| 889 | } |
| 890 | |
| 891 | func (tr *trace) ElapsedTime() string { |
| 892 | tr.mu.RLock() |
| 893 | t := tr.Elapsed |
| 894 | tr.mu.RUnlock() |
| 895 | |
| 896 | if t == 0 { |
| 897 | // Active trace. |
| 898 | t = time.Since(tr.Start) |
| 899 | } |
| 900 | return fmt.Sprintf("%.6f", t.Seconds()) |
| 901 | } |
| 902 | |
| 903 | func (tr *trace) Events() []event { |
| 904 | tr.mu.RLock() |
| 905 | defer tr.mu.RUnlock() |
| 906 | return tr.events |
| 907 | } |
| 908 | |
| 909 | var traceFreeList = make(chan *trace, 1000) // TODO(dsymonds): Use sync.Pool? |
| 910 | |
| 911 | // newTrace returns a trace ready to use. |
| 912 | func newTrace() *trace { |
| 913 | select { |
| 914 | case tr := <-traceFreeList: |
| 915 | return tr |
| 916 | default: |
| 917 | return new(trace) |
| 918 | } |
| 919 | } |
| 920 | |
| 921 | // freeTrace adds tr to traceFreeList if there's room. |
| 922 | // This is non-blocking. |
| 923 | func freeTrace(tr *trace) { |
| 924 | if DebugUseAfterFinish { |
| 925 | return // never reuse |
| 926 | } |
| 927 | tr.reset() |
| 928 | select { |
| 929 | case traceFreeList <- tr: |
| 930 | default: |
| 931 | } |
| 932 | } |
| 933 | |
| 934 | func elapsed(d time.Duration) string { |
| 935 | b := []byte(fmt.Sprintf("%.6f", d.Seconds())) |
| 936 | |
| 937 | // For subsecond durations, blank all zeros before decimal point, |
| 938 | // and all zeros between the decimal point and the first non-zero digit. |
| 939 | if d < time.Second { |
| 940 | dot := bytes.IndexByte(b, '.') |
| 941 | for i := 0; i < dot; i++ { |
| 942 | b[i] = ' ' |
| 943 | } |
| 944 | for i := dot + 1; i < len(b); i++ { |
| 945 | if b[i] == '0' { |
| 946 | b[i] = ' ' |
| 947 | } else { |
| 948 | break |
| 949 | } |
| 950 | } |
| 951 | } |
| 952 | |
| 953 | return string(b) |
| 954 | } |
| 955 | |
| 956 | var pageTmplCache *template.Template |
| 957 | var pageTmplOnce sync.Once |
| 958 | |
| 959 | func pageTmpl() *template.Template { |
| 960 | pageTmplOnce.Do(func() { |
| 961 | pageTmplCache = template.Must(template.New("Page").Funcs(template.FuncMap{ |
| 962 | "elapsed": elapsed, |
| 963 | "add": func(a, b int) int { return a + b }, |
| 964 | }).Parse(pageHTML)) |
| 965 | }) |
| 966 | return pageTmplCache |
| 967 | } |
| 968 | |
| 969 | const pageHTML = ` |
| 970 | {{template "Prolog" .}} |
| 971 | {{template "StatusTable" .}} |
| 972 | {{template "Epilog" .}} |
| 973 | |
| 974 | {{define "Prolog"}} |
| 975 | <html> |
| 976 | <head> |
| 977 | <title>/debug/requests</title> |
| 978 | <style type="text/css"> |
| 979 | body { |
| 980 | font-family: sans-serif; |
| 981 | } |
| 982 | table#tr-status td.family { |
| 983 | padding-right: 2em; |
| 984 | } |
| 985 | table#tr-status td.active { |
| 986 | padding-right: 1em; |
| 987 | } |
| 988 | table#tr-status td.latency-first { |
| 989 | padding-left: 1em; |
| 990 | } |
| 991 | table#tr-status td.empty { |
| 992 | color: #aaa; |
| 993 | } |
| 994 | table#reqs { |
| 995 | margin-top: 1em; |
| 996 | } |
| 997 | table#reqs tr.first { |
| 998 | {{if $.Expanded}}font-weight: bold;{{end}} |
| 999 | } |
| 1000 | table#reqs td { |
| 1001 | font-family: monospace; |
| 1002 | } |
| 1003 | table#reqs td.when { |
| 1004 | text-align: right; |
| 1005 | white-space: nowrap; |
| 1006 | } |
| 1007 | table#reqs td.elapsed { |
| 1008 | padding: 0 0.5em; |
| 1009 | text-align: right; |
| 1010 | white-space: pre; |
| 1011 | width: 10em; |
| 1012 | } |
| 1013 | address { |
| 1014 | font-size: smaller; |
| 1015 | margin-top: 5em; |
| 1016 | } |
| 1017 | </style> |
| 1018 | </head> |
| 1019 | <body> |
| 1020 | |
| 1021 | <h1>/debug/requests</h1> |
| 1022 | {{end}} {{/* end of Prolog */}} |
| 1023 | |
| 1024 | {{define "StatusTable"}} |
| 1025 | <table id="tr-status"> |
| 1026 | {{range $fam := .Families}} |
| 1027 | <tr> |
| 1028 | <td class="family">{{$fam}}</td> |
| 1029 | |
| 1030 | {{$n := index $.ActiveTraceCount $fam}} |
| 1031 | <td class="active {{if not $n}}empty{{end}}"> |
| 1032 | {{if $n}}<a href="?fam={{$fam}}&b=-1{{if $.Expanded}}&exp=1{{end}}">{{end}} |
| 1033 | [{{$n}} active] |
| 1034 | {{if $n}}</a>{{end}} |
| 1035 | </td> |
| 1036 | |
| 1037 | {{$f := index $.CompletedTraces $fam}} |
| 1038 | {{range $i, $b := $f.Buckets}} |
| 1039 | {{$empty := $b.Empty}} |
| 1040 | <td {{if $empty}}class="empty"{{end}}> |
| 1041 | {{if not $empty}}<a href="?fam={{$fam}}&b={{$i}}{{if $.Expanded}}&exp=1{{end}}">{{end}} |
| 1042 | [{{.Cond}}] |
| 1043 | {{if not $empty}}</a>{{end}} |
| 1044 | </td> |
| 1045 | {{end}} |
| 1046 | |
| 1047 | {{$nb := len $f.Buckets}} |
| 1048 | <td class="latency-first"> |
| 1049 | <a href="?fam={{$fam}}&b={{$nb}}">[minute]</a> |
| 1050 | </td> |
| 1051 | <td> |
| 1052 | <a href="?fam={{$fam}}&b={{add $nb 1}}">[hour]</a> |
| 1053 | </td> |
| 1054 | <td> |
| 1055 | <a href="?fam={{$fam}}&b={{add $nb 2}}">[total]</a> |
| 1056 | </td> |
| 1057 | |
| 1058 | </tr> |
| 1059 | {{end}} |
| 1060 | </table> |
| 1061 | {{end}} {{/* end of StatusTable */}} |
| 1062 | |
| 1063 | {{define "Epilog"}} |
| 1064 | {{if $.Traces}} |
| 1065 | <hr /> |
| 1066 | <h3>Family: {{$.Family}}</h3> |
| 1067 | |
| 1068 | {{if or $.Expanded $.Traced}} |
| 1069 | <a href="?fam={{$.Family}}&b={{$.Bucket}}">[Normal/Summary]</a> |
| 1070 | {{else}} |
| 1071 | [Normal/Summary] |
| 1072 | {{end}} |
| 1073 | |
| 1074 | {{if or (not $.Expanded) $.Traced}} |
| 1075 | <a href="?fam={{$.Family}}&b={{$.Bucket}}&exp=1">[Normal/Expanded]</a> |
| 1076 | {{else}} |
| 1077 | [Normal/Expanded] |
| 1078 | {{end}} |
| 1079 | |
| 1080 | {{if not $.Active}} |
| 1081 | {{if or $.Expanded (not $.Traced)}} |
| 1082 | <a href="?fam={{$.Family}}&b={{$.Bucket}}&rtraced=1">[Traced/Summary]</a> |
| 1083 | {{else}} |
| 1084 | [Traced/Summary] |
| 1085 | {{end}} |
| 1086 | {{if or (not $.Expanded) (not $.Traced)}} |
| 1087 | <a href="?fam={{$.Family}}&b={{$.Bucket}}&exp=1&rtraced=1">[Traced/Expanded]</a> |
| 1088 | {{else}} |
| 1089 | [Traced/Expanded] |
| 1090 | {{end}} |
| 1091 | {{end}} |
| 1092 | |
| 1093 | {{if $.Total}} |
| 1094 | <p><em>Showing <b>{{len $.Traces}}</b> of <b>{{$.Total}}</b> traces.</em></p> |
| 1095 | {{end}} |
| 1096 | |
| 1097 | <table id="reqs"> |
| 1098 | <caption> |
| 1099 | {{if $.Active}}Active{{else}}Completed{{end}} Requests |
| 1100 | </caption> |
| 1101 | <tr><th>When</th><th>Elapsed (s)</th></tr> |
| 1102 | {{range $tr := $.Traces}} |
| 1103 | <tr class="first"> |
| 1104 | <td class="when">{{$tr.When}}</td> |
| 1105 | <td class="elapsed">{{$tr.ElapsedTime}}</td> |
| 1106 | <td>{{$tr.Title}}</td> |
| 1107 | {{/* TODO: include traceID/spanID */}} |
| 1108 | </tr> |
| 1109 | {{if $.Expanded}} |
| 1110 | {{range $tr.Events}} |
| 1111 | <tr> |
| 1112 | <td class="when">{{.WhenString}}</td> |
| 1113 | <td class="elapsed">{{elapsed .Elapsed}}</td> |
| 1114 | <td>{{if or $.ShowSensitive (not .Sensitive)}}... {{.What}}{{else}}<em>[redacted]</em>{{end}}</td> |
| 1115 | </tr> |
| 1116 | {{end}} |
| 1117 | {{end}} |
| 1118 | {{end}} |
| 1119 | </table> |
| 1120 | {{end}} {{/* if $.Traces */}} |
| 1121 | |
| 1122 | {{if $.Histogram}} |
| 1123 | <h4>Latency (µs) of {{$.Family}} over {{$.HistogramWindow}}</h4> |
| 1124 | {{$.Histogram}} |
| 1125 | {{end}} {{/* if $.Histogram */}} |
| 1126 | |
| 1127 | </body> |
| 1128 | </html> |
| 1129 | {{end}} {{/* end of Epilog */}} |
| 1130 | ` |