Documentation
¶
Overview ¶
Package traceutil adapts framingo's explicit model.Tracer handle to the plain OpenTelemetry interfaces that third-party instrumentation expects.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func Fail ¶ added in v0.6.13
Fail records err on the span, marks it failed, ends it, and hands err back - so an error return stays one line:
return nil, traceutil.Fail(span, errors.Wrap(err))
The success path ends the span itself with a plain span.End(). Every return must do one or the other: a path that does neither leaks an unended span that never exports.
func LiveProvider ¶
func LiveProvider(t model.Tracer) trace.TracerProvider
LiveProvider adapts a model.Tracer into a trace.TracerProvider that re-resolves the real provider on every span instead of holding one.
Instrumentation libraries almost all take a TracerProvider and read provider.Tracer(...) once, at construction. Handing them model.Tracer's provider directly would pin their spans to whichever provider happened to be live at wiring time - and a tracing-service restart retires that one, which then drops every span it is given without erroring. Use this wherever a library wants a TracerProvider:
tracing.WithTracerProvider(traceutil.LiveProvider(t))
Start resolves through model.Tracer on each call, which costs a lock and a map lookup in the SDK's tracer cache. Immaterial at database-statement rates; worth measuring before putting it on a per-request path.
func Start ¶ added in v0.6.13
func Start(ctx context.Context, scope, name string, opts ...trace.SpanStartOption) (context.Context, trace.Span)
Start begins a child span from the provider already bound to ctx's span - the handle-free path for business code. The middleware's request span carries the live provider, so no service holds a tracer; a context without a recording span (tracing off, unit tests) yields a noop span.
Types ¶
This section is empty.
Source Files