TracerProvider shutdown release resources (#3551)

* TracerProvider shutdown release resources

Signed-off-by: Ziqi Zhao <zhaoziqi9146@gmail.com>

* add changelog

Signed-off-by: Ziqi Zhao <zhaoziqi9146@gmail.com>

* Update CHANGELOG.md

Co-authored-by: David Ashpole <dashpole@google.com>

* prevent registered span processors after shutdown

Signed-off-by: Ziqi Zhao <zhaoziqi9146@gmail.com>

* Update CHANGELOG.md

Signed-off-by: Ziqi Zhao <zhaoziqi9146@gmail.com>
Co-authored-by: David Ashpole <dashpole@google.com>
Co-authored-by: Chester Cheung <cheung.zhy.csu@gmail.com>
Co-authored-by: Tyler Yahn <MrAlias@users.noreply.github.com>

CHANGELOG.md

@@ -24,6 +24,7 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 - Add `Producer` interface and `Reader.RegisterProducer(Producer)` to `go.opentelemetry.io/otel/sdk/metric` to enable external metric Producers. (#3524)
 - Add `NewMetricProducer` to `go.opentelemetry.io/otel/bridge/opencensus`, which can be used to pass OpenCensus metrics to an OpenTelemetry Reader. (#3541)
 - Global logger uses an atomic value instead of a mutex. (#3545)
+- The `Shutdown` method of the `"go.opentelemetry.io/otel/sdk/trace".TracerProvider` releases all computational resources when called the first time. (#3551)
 - `traceIDRatioSampler` (given by `TraceIDRatioBased(float64)`) now uses the rightmost bits for sampling decisions,
   fixing random sampling when using ID generators like `xray.IDGenerator`
   and increasing parity with other language implementations. (#3557)

sdk/trace/provider.go

@@ -76,6 +76,7 @@ type TracerProvider struct {
 	mu             sync.Mutex
 	namedTracer    map[instrumentation.Scope]*tracer
 	spanProcessors atomic.Value
+	isShutdown     bool
 
 	// These fields are not protected by the lock mu. They are assumed to be
 	// immutable after creation of the TracerProvider.
@@ -163,6 +164,9 @@ func (p *TracerProvider) Tracer(name string, opts ...trace.TracerOption) trace.T
 func (p *TracerProvider) RegisterSpanProcessor(sp SpanProcessor) {
 	p.mu.Lock()
 	defer p.mu.Unlock()
+	if p.isShutdown {
+		return
+	}
 	newSPS := spanProcessorStates{}
 	newSPS = append(newSPS, p.spanProcessors.Load().(spanProcessorStates)...)
 	newSPS = append(newSPS, newSpanProcessorState(sp))
@@ -173,6 +177,9 @@ func (p *TracerProvider) RegisterSpanProcessor(sp SpanProcessor) {
 func (p *TracerProvider) UnregisterSpanProcessor(sp SpanProcessor) {
 	p.mu.Lock()
 	defer p.mu.Unlock()
+	if p.isShutdown {
+		return
+	}
 	old := p.spanProcessors.Load().(spanProcessorStates)
 	if len(old) == 0 {
 		return
@@ -227,13 +234,18 @@ func (p *TracerProvider) ForceFlush(ctx context.Context) error {
 	return nil
 }
 
-// Shutdown shuts down the span processors in the order they were registered.
+// Shutdown shuts down TracerProvider. All registered span processors are shut down
+// in the order they were registered and any held computational resources are released.
 func (p *TracerProvider) Shutdown(ctx context.Context) error {
 	spss := p.spanProcessors.Load().(spanProcessorStates)
 	if len(spss) == 0 {
 		return nil
 	}
 
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	p.isShutdown = true
+
 	var retErr error
 	for _, sps := range spss {
 		select {
@@ -255,6 +267,7 @@ func (p *TracerProvider) Shutdown(ctx context.Context) error {
 			}
 		}
 	}
+	p.spanProcessors.Store(spanProcessorStates{})
 	return retErr
 }