| Abhay Kumar | a2ae599 | 2025-11-10 14:02:24 +0000 | [diff] [blame^] | 1 | // Copyright The OpenTelemetry Authors |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| 4 | package trace // import "go.opentelemetry.io/otel/sdk/trace" |
| 5 | |
| 6 | import ( |
| 7 | "context" |
| 8 | "fmt" |
| 9 | "sync" |
| 10 | "sync/atomic" |
| 11 | |
| 12 | "go.opentelemetry.io/otel" |
| 13 | "go.opentelemetry.io/otel/internal/global" |
| 14 | "go.opentelemetry.io/otel/sdk/instrumentation" |
| 15 | "go.opentelemetry.io/otel/sdk/resource" |
| 16 | "go.opentelemetry.io/otel/trace" |
| 17 | "go.opentelemetry.io/otel/trace/embedded" |
| 18 | "go.opentelemetry.io/otel/trace/noop" |
| 19 | ) |
| 20 | |
| 21 | const ( |
| 22 | defaultTracerName = "go.opentelemetry.io/otel/sdk/tracer" |
| 23 | ) |
| 24 | |
| 25 | // tracerProviderConfig. |
| 26 | type tracerProviderConfig struct { |
| 27 | // processors contains collection of SpanProcessors that are processing pipeline |
| 28 | // for spans in the trace signal. |
| 29 | // SpanProcessors registered with a TracerProvider and are called at the start |
| 30 | // and end of a Span's lifecycle, and are called in the order they are |
| 31 | // registered. |
| 32 | processors []SpanProcessor |
| 33 | |
| 34 | // sampler is the default sampler used when creating new spans. |
| 35 | sampler Sampler |
| 36 | |
| 37 | // idGenerator is used to generate all Span and Trace IDs when needed. |
| 38 | idGenerator IDGenerator |
| 39 | |
| 40 | // spanLimits defines the attribute, event, and link limits for spans. |
| 41 | spanLimits SpanLimits |
| 42 | |
| 43 | // resource contains attributes representing an entity that produces telemetry. |
| 44 | resource *resource.Resource |
| 45 | } |
| 46 | |
| 47 | // MarshalLog is the marshaling function used by the logging system to represent this Provider. |
| 48 | func (cfg tracerProviderConfig) MarshalLog() interface{} { |
| 49 | return struct { |
| 50 | SpanProcessors []SpanProcessor |
| 51 | SamplerType string |
| 52 | IDGeneratorType string |
| 53 | SpanLimits SpanLimits |
| 54 | Resource *resource.Resource |
| 55 | }{ |
| 56 | SpanProcessors: cfg.processors, |
| 57 | SamplerType: fmt.Sprintf("%T", cfg.sampler), |
| 58 | IDGeneratorType: fmt.Sprintf("%T", cfg.idGenerator), |
| 59 | SpanLimits: cfg.spanLimits, |
| 60 | Resource: cfg.resource, |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | // TracerProvider is an OpenTelemetry TracerProvider. It provides Tracers to |
| 65 | // instrumentation so it can trace operational flow through a system. |
| 66 | type TracerProvider struct { |
| 67 | embedded.TracerProvider |
| 68 | |
| 69 | mu sync.Mutex |
| 70 | namedTracer map[instrumentation.Scope]*tracer |
| 71 | spanProcessors atomic.Pointer[spanProcessorStates] |
| 72 | |
| 73 | isShutdown atomic.Bool |
| 74 | |
| 75 | // These fields are not protected by the lock mu. They are assumed to be |
| 76 | // immutable after creation of the TracerProvider. |
| 77 | sampler Sampler |
| 78 | idGenerator IDGenerator |
| 79 | spanLimits SpanLimits |
| 80 | resource *resource.Resource |
| 81 | } |
| 82 | |
| 83 | var _ trace.TracerProvider = &TracerProvider{} |
| 84 | |
| 85 | // NewTracerProvider returns a new and configured TracerProvider. |
| 86 | // |
| 87 | // By default the returned TracerProvider is configured with: |
| 88 | // - a ParentBased(AlwaysSample) Sampler |
| 89 | // - a random number IDGenerator |
| 90 | // - the resource.Default() Resource |
| 91 | // - the default SpanLimits. |
| 92 | // |
| 93 | // The passed opts are used to override these default values and configure the |
| 94 | // returned TracerProvider appropriately. |
| 95 | func NewTracerProvider(opts ...TracerProviderOption) *TracerProvider { |
| 96 | o := tracerProviderConfig{ |
| 97 | spanLimits: NewSpanLimits(), |
| 98 | } |
| 99 | o = applyTracerProviderEnvConfigs(o) |
| 100 | |
| 101 | for _, opt := range opts { |
| 102 | o = opt.apply(o) |
| 103 | } |
| 104 | |
| 105 | o = ensureValidTracerProviderConfig(o) |
| 106 | |
| 107 | tp := &TracerProvider{ |
| 108 | namedTracer: make(map[instrumentation.Scope]*tracer), |
| 109 | sampler: o.sampler, |
| 110 | idGenerator: o.idGenerator, |
| 111 | spanLimits: o.spanLimits, |
| 112 | resource: o.resource, |
| 113 | } |
| 114 | global.Info("TracerProvider created", "config", o) |
| 115 | |
| 116 | spss := make(spanProcessorStates, 0, len(o.processors)) |
| 117 | for _, sp := range o.processors { |
| 118 | spss = append(spss, newSpanProcessorState(sp)) |
| 119 | } |
| 120 | tp.spanProcessors.Store(&spss) |
| 121 | |
| 122 | return tp |
| 123 | } |
| 124 | |
| 125 | // Tracer returns a Tracer with the given name and options. If a Tracer for |
| 126 | // the given name and options does not exist it is created, otherwise the |
| 127 | // existing Tracer is returned. |
| 128 | // |
| 129 | // If name is empty, DefaultTracerName is used instead. |
| 130 | // |
| 131 | // This method is safe to be called concurrently. |
| 132 | func (p *TracerProvider) Tracer(name string, opts ...trace.TracerOption) trace.Tracer { |
| 133 | // This check happens before the mutex is acquired to avoid deadlocking if Tracer() is called from within Shutdown(). |
| 134 | if p.isShutdown.Load() { |
| 135 | return noop.NewTracerProvider().Tracer(name, opts...) |
| 136 | } |
| 137 | c := trace.NewTracerConfig(opts...) |
| 138 | if name == "" { |
| 139 | name = defaultTracerName |
| 140 | } |
| 141 | is := instrumentation.Scope{ |
| 142 | Name: name, |
| 143 | Version: c.InstrumentationVersion(), |
| 144 | SchemaURL: c.SchemaURL(), |
| 145 | Attributes: c.InstrumentationAttributes(), |
| 146 | } |
| 147 | |
| 148 | t, ok := func() (trace.Tracer, bool) { |
| 149 | p.mu.Lock() |
| 150 | defer p.mu.Unlock() |
| 151 | // Must check the flag after acquiring the mutex to avoid returning a valid tracer if Shutdown() ran |
| 152 | // after the first check above but before we acquired the mutex. |
| 153 | if p.isShutdown.Load() { |
| 154 | return noop.NewTracerProvider().Tracer(name, opts...), true |
| 155 | } |
| 156 | t, ok := p.namedTracer[is] |
| 157 | if !ok { |
| 158 | t = &tracer{ |
| 159 | provider: p, |
| 160 | instrumentationScope: is, |
| 161 | } |
| 162 | p.namedTracer[is] = t |
| 163 | } |
| 164 | return t, ok |
| 165 | }() |
| 166 | if !ok { |
| 167 | // This code is outside the mutex to not hold the lock while calling third party logging code: |
| 168 | // - That code may do slow things like I/O, which would prolong the duration the lock is held, |
| 169 | // slowing down all tracing consumers. |
| 170 | // - Logging code may be instrumented with tracing and deadlock because it could try |
| 171 | // acquiring the same non-reentrant mutex. |
| 172 | global.Info( |
| 173 | "Tracer created", |
| 174 | "name", |
| 175 | name, |
| 176 | "version", |
| 177 | is.Version, |
| 178 | "schemaURL", |
| 179 | is.SchemaURL, |
| 180 | "attributes", |
| 181 | is.Attributes, |
| 182 | ) |
| 183 | } |
| 184 | return t |
| 185 | } |
| 186 | |
| 187 | // RegisterSpanProcessor adds the given SpanProcessor to the list of SpanProcessors. |
| 188 | func (p *TracerProvider) RegisterSpanProcessor(sp SpanProcessor) { |
| 189 | // This check prevents calls during a shutdown. |
| 190 | if p.isShutdown.Load() { |
| 191 | return |
| 192 | } |
| 193 | p.mu.Lock() |
| 194 | defer p.mu.Unlock() |
| 195 | // This check prevents calls after a shutdown. |
| 196 | if p.isShutdown.Load() { |
| 197 | return |
| 198 | } |
| 199 | |
| 200 | current := p.getSpanProcessors() |
| 201 | newSPS := make(spanProcessorStates, 0, len(current)+1) |
| 202 | newSPS = append(newSPS, current...) |
| 203 | newSPS = append(newSPS, newSpanProcessorState(sp)) |
| 204 | p.spanProcessors.Store(&newSPS) |
| 205 | } |
| 206 | |
| 207 | // UnregisterSpanProcessor removes the given SpanProcessor from the list of SpanProcessors. |
| 208 | func (p *TracerProvider) UnregisterSpanProcessor(sp SpanProcessor) { |
| 209 | // This check prevents calls during a shutdown. |
| 210 | if p.isShutdown.Load() { |
| 211 | return |
| 212 | } |
| 213 | p.mu.Lock() |
| 214 | defer p.mu.Unlock() |
| 215 | // This check prevents calls after a shutdown. |
| 216 | if p.isShutdown.Load() { |
| 217 | return |
| 218 | } |
| 219 | old := p.getSpanProcessors() |
| 220 | if len(old) == 0 { |
| 221 | return |
| 222 | } |
| 223 | spss := make(spanProcessorStates, len(old)) |
| 224 | copy(spss, old) |
| 225 | |
| 226 | // stop the span processor if it is started and remove it from the list |
| 227 | var stopOnce *spanProcessorState |
| 228 | var idx int |
| 229 | for i, sps := range spss { |
| 230 | if sps.sp == sp { |
| 231 | stopOnce = sps |
| 232 | idx = i |
| 233 | } |
| 234 | } |
| 235 | if stopOnce != nil { |
| 236 | stopOnce.state.Do(func() { |
| 237 | if err := sp.Shutdown(context.Background()); err != nil { |
| 238 | otel.Handle(err) |
| 239 | } |
| 240 | }) |
| 241 | } |
| 242 | if len(spss) > 1 { |
| 243 | copy(spss[idx:], spss[idx+1:]) |
| 244 | } |
| 245 | spss[len(spss)-1] = nil |
| 246 | spss = spss[:len(spss)-1] |
| 247 | |
| 248 | p.spanProcessors.Store(&spss) |
| 249 | } |
| 250 | |
| 251 | // ForceFlush immediately exports all spans that have not yet been exported for |
| 252 | // all the registered span processors. |
| 253 | func (p *TracerProvider) ForceFlush(ctx context.Context) error { |
| 254 | spss := p.getSpanProcessors() |
| 255 | if len(spss) == 0 { |
| 256 | return nil |
| 257 | } |
| 258 | |
| 259 | for _, sps := range spss { |
| 260 | select { |
| 261 | case <-ctx.Done(): |
| 262 | return ctx.Err() |
| 263 | default: |
| 264 | } |
| 265 | |
| 266 | if err := sps.sp.ForceFlush(ctx); err != nil { |
| 267 | return err |
| 268 | } |
| 269 | } |
| 270 | return nil |
| 271 | } |
| 272 | |
| 273 | // Shutdown shuts down TracerProvider. All registered span processors are shut down |
| 274 | // in the order they were registered and any held computational resources are released. |
| 275 | // After Shutdown is called, all methods are no-ops. |
| 276 | func (p *TracerProvider) Shutdown(ctx context.Context) error { |
| 277 | // This check prevents deadlocks in case of recursive shutdown. |
| 278 | if p.isShutdown.Load() { |
| 279 | return nil |
| 280 | } |
| 281 | p.mu.Lock() |
| 282 | defer p.mu.Unlock() |
| 283 | // This check prevents calls after a shutdown has already been done concurrently. |
| 284 | if !p.isShutdown.CompareAndSwap(false, true) { // did toggle? |
| 285 | return nil |
| 286 | } |
| 287 | |
| 288 | var retErr error |
| 289 | for _, sps := range p.getSpanProcessors() { |
| 290 | select { |
| 291 | case <-ctx.Done(): |
| 292 | return ctx.Err() |
| 293 | default: |
| 294 | } |
| 295 | |
| 296 | var err error |
| 297 | sps.state.Do(func() { |
| 298 | err = sps.sp.Shutdown(ctx) |
| 299 | }) |
| 300 | if err != nil { |
| 301 | if retErr == nil { |
| 302 | retErr = err |
| 303 | } else { |
| 304 | // Poor man's list of errors |
| 305 | retErr = fmt.Errorf("%w; %w", retErr, err) |
| 306 | } |
| 307 | } |
| 308 | } |
| 309 | p.spanProcessors.Store(&spanProcessorStates{}) |
| 310 | return retErr |
| 311 | } |
| 312 | |
| 313 | func (p *TracerProvider) getSpanProcessors() spanProcessorStates { |
| 314 | return *(p.spanProcessors.Load()) |
| 315 | } |
| 316 | |
| 317 | // TracerProviderOption configures a TracerProvider. |
| 318 | type TracerProviderOption interface { |
| 319 | apply(tracerProviderConfig) tracerProviderConfig |
| 320 | } |
| 321 | |
| 322 | type traceProviderOptionFunc func(tracerProviderConfig) tracerProviderConfig |
| 323 | |
| 324 | func (fn traceProviderOptionFunc) apply(cfg tracerProviderConfig) tracerProviderConfig { |
| 325 | return fn(cfg) |
| 326 | } |
| 327 | |
| 328 | // WithSyncer registers the exporter with the TracerProvider using a |
| 329 | // SimpleSpanProcessor. |
| 330 | // |
| 331 | // This is not recommended for production use. The synchronous nature of the |
| 332 | // SimpleSpanProcessor that will wrap the exporter make it good for testing, |
| 333 | // debugging, or showing examples of other feature, but it will be slow and |
| 334 | // have a high computation resource usage overhead. The WithBatcher option is |
| 335 | // recommended for production use instead. |
| 336 | func WithSyncer(e SpanExporter) TracerProviderOption { |
| 337 | return WithSpanProcessor(NewSimpleSpanProcessor(e)) |
| 338 | } |
| 339 | |
| 340 | // WithBatcher registers the exporter with the TracerProvider using a |
| 341 | // BatchSpanProcessor configured with the passed opts. |
| 342 | func WithBatcher(e SpanExporter, opts ...BatchSpanProcessorOption) TracerProviderOption { |
| 343 | return WithSpanProcessor(NewBatchSpanProcessor(e, opts...)) |
| 344 | } |
| 345 | |
| 346 | // WithSpanProcessor registers the SpanProcessor with a TracerProvider. |
| 347 | func WithSpanProcessor(sp SpanProcessor) TracerProviderOption { |
| 348 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 349 | cfg.processors = append(cfg.processors, sp) |
| 350 | return cfg |
| 351 | }) |
| 352 | } |
| 353 | |
| 354 | // WithResource returns a TracerProviderOption that will configure the |
| 355 | // Resource r as a TracerProvider's Resource. The configured Resource is |
| 356 | // referenced by all the Tracers the TracerProvider creates. It represents the |
| 357 | // entity producing telemetry. |
| 358 | // |
| 359 | // If this option is not used, the TracerProvider will use the |
| 360 | // resource.Default() Resource by default. |
| 361 | func WithResource(r *resource.Resource) TracerProviderOption { |
| 362 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 363 | var err error |
| 364 | cfg.resource, err = resource.Merge(resource.Environment(), r) |
| 365 | if err != nil { |
| 366 | otel.Handle(err) |
| 367 | } |
| 368 | return cfg |
| 369 | }) |
| 370 | } |
| 371 | |
| 372 | // WithIDGenerator returns a TracerProviderOption that will configure the |
| 373 | // IDGenerator g as a TracerProvider's IDGenerator. The configured IDGenerator |
| 374 | // is used by the Tracers the TracerProvider creates to generate new Span and |
| 375 | // Trace IDs. |
| 376 | // |
| 377 | // If this option is not used, the TracerProvider will use a random number |
| 378 | // IDGenerator by default. |
| 379 | func WithIDGenerator(g IDGenerator) TracerProviderOption { |
| 380 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 381 | if g != nil { |
| 382 | cfg.idGenerator = g |
| 383 | } |
| 384 | return cfg |
| 385 | }) |
| 386 | } |
| 387 | |
| 388 | // WithSampler returns a TracerProviderOption that will configure the Sampler |
| 389 | // s as a TracerProvider's Sampler. The configured Sampler is used by the |
| 390 | // Tracers the TracerProvider creates to make their sampling decisions for the |
| 391 | // Spans they create. |
| 392 | // |
| 393 | // This option overrides the Sampler configured through the OTEL_TRACES_SAMPLER |
| 394 | // and OTEL_TRACES_SAMPLER_ARG environment variables. If this option is not used |
| 395 | // and the sampler is not configured through environment variables or the environment |
| 396 | // contains invalid/unsupported configuration, the TracerProvider will use a |
| 397 | // ParentBased(AlwaysSample) Sampler by default. |
| 398 | func WithSampler(s Sampler) TracerProviderOption { |
| 399 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 400 | if s != nil { |
| 401 | cfg.sampler = s |
| 402 | } |
| 403 | return cfg |
| 404 | }) |
| 405 | } |
| 406 | |
| 407 | // WithSpanLimits returns a TracerProviderOption that configures a |
| 408 | // TracerProvider to use the SpanLimits sl. These SpanLimits bound any Span |
| 409 | // created by a Tracer from the TracerProvider. |
| 410 | // |
| 411 | // If any field of sl is zero or negative it will be replaced with the default |
| 412 | // value for that field. |
| 413 | // |
| 414 | // If this or WithRawSpanLimits are not provided, the TracerProvider will use |
| 415 | // the limits defined by environment variables, or the defaults if unset. |
| 416 | // Refer to the NewSpanLimits documentation for information about this |
| 417 | // relationship. |
| 418 | // |
| 419 | // Deprecated: Use WithRawSpanLimits instead which allows setting unlimited |
| 420 | // and zero limits. This option will be kept until the next major version |
| 421 | // incremented release. |
| 422 | func WithSpanLimits(sl SpanLimits) TracerProviderOption { |
| 423 | if sl.AttributeValueLengthLimit <= 0 { |
| 424 | sl.AttributeValueLengthLimit = DefaultAttributeValueLengthLimit |
| 425 | } |
| 426 | if sl.AttributeCountLimit <= 0 { |
| 427 | sl.AttributeCountLimit = DefaultAttributeCountLimit |
| 428 | } |
| 429 | if sl.EventCountLimit <= 0 { |
| 430 | sl.EventCountLimit = DefaultEventCountLimit |
| 431 | } |
| 432 | if sl.AttributePerEventCountLimit <= 0 { |
| 433 | sl.AttributePerEventCountLimit = DefaultAttributePerEventCountLimit |
| 434 | } |
| 435 | if sl.LinkCountLimit <= 0 { |
| 436 | sl.LinkCountLimit = DefaultLinkCountLimit |
| 437 | } |
| 438 | if sl.AttributePerLinkCountLimit <= 0 { |
| 439 | sl.AttributePerLinkCountLimit = DefaultAttributePerLinkCountLimit |
| 440 | } |
| 441 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 442 | cfg.spanLimits = sl |
| 443 | return cfg |
| 444 | }) |
| 445 | } |
| 446 | |
| 447 | // WithRawSpanLimits returns a TracerProviderOption that configures a |
| 448 | // TracerProvider to use these limits. These limits bound any Span created by |
| 449 | // a Tracer from the TracerProvider. |
| 450 | // |
| 451 | // The limits will be used as-is. Zero or negative values will not be changed |
| 452 | // to the default value like WithSpanLimits does. Setting a limit to zero will |
| 453 | // effectively disable the related resource it limits and setting to a |
| 454 | // negative value will mean that resource is unlimited. Consequentially, this |
| 455 | // means that the zero-value SpanLimits will disable all span resources. |
| 456 | // Because of this, limits should be constructed using NewSpanLimits and |
| 457 | // updated accordingly. |
| 458 | // |
| 459 | // If this or WithSpanLimits are not provided, the TracerProvider will use the |
| 460 | // limits defined by environment variables, or the defaults if unset. Refer to |
| 461 | // the NewSpanLimits documentation for information about this relationship. |
| 462 | func WithRawSpanLimits(limits SpanLimits) TracerProviderOption { |
| 463 | return traceProviderOptionFunc(func(cfg tracerProviderConfig) tracerProviderConfig { |
| 464 | cfg.spanLimits = limits |
| 465 | return cfg |
| 466 | }) |
| 467 | } |
| 468 | |
| 469 | func applyTracerProviderEnvConfigs(cfg tracerProviderConfig) tracerProviderConfig { |
| 470 | for _, opt := range tracerProviderOptionsFromEnv() { |
| 471 | cfg = opt.apply(cfg) |
| 472 | } |
| 473 | |
| 474 | return cfg |
| 475 | } |
| 476 | |
| 477 | func tracerProviderOptionsFromEnv() []TracerProviderOption { |
| 478 | var opts []TracerProviderOption |
| 479 | |
| 480 | sampler, err := samplerFromEnv() |
| 481 | if err != nil { |
| 482 | otel.Handle(err) |
| 483 | } |
| 484 | |
| 485 | if sampler != nil { |
| 486 | opts = append(opts, WithSampler(sampler)) |
| 487 | } |
| 488 | |
| 489 | return opts |
| 490 | } |
| 491 | |
| 492 | // ensureValidTracerProviderConfig ensures that given TracerProviderConfig is valid. |
| 493 | func ensureValidTracerProviderConfig(cfg tracerProviderConfig) tracerProviderConfig { |
| 494 | if cfg.sampler == nil { |
| 495 | cfg.sampler = ParentBased(AlwaysSample()) |
| 496 | } |
| 497 | if cfg.idGenerator == nil { |
| 498 | cfg.idGenerator = defaultIDGenerator() |
| 499 | } |
| 500 | if cfg.resource == nil { |
| 501 | cfg.resource = resource.Default() |
| 502 | } |
| 503 | return cfg |
| 504 | } |