What causes "failed to close HTTP server: %w" in jaegertracing/jaeger?

Inspect the wrapped inner error: if it indicates the server is already closed, ensure Close is called exactly once (guard with sync.Once)

jaegertracing/jaeger · warning

failed to close HTTP server: %w

Error message

failed to close HTTP server: %w

What it means

Server.Close shuts down the HTTP server, stops the gRPC server, and waits for background work, collecting any errors into a joined error. If httpServer.Close() returns non-nil — e.g. the underlying listener is already closed or in a bad state — it is wrapped as 'failed to close HTTP server' and combined with any other shutdown errors.

Source

Thrown at cmd/jaeger/internal/extension/jaegerquery/internal/server.go:512

	})
	return nil
}

func (s *Server) HTTPAddr() string {
	return s.httpConn.Addr().String()
}

func (s *Server) GRPCAddr() string {
	return s.grpcConn.Addr().String()
}

// Close stops HTTP, GRPC servers and closes the port listener.
func (s *Server) Close() error {
	var errs []error

	s.telset.Logger.Info("Closing HTTP server")
	if err := s.httpServer.Close(); err != nil {
		errs = append(errs, fmt.Errorf("failed to close HTTP server: %w", err))
	}

	s.telset.Logger.Info("Stopping gRPC server")
	s.grpcServer.Stop()

	s.bgFinished.Wait()

	s.telset.Logger.Info("Server stopped")
	return errors.Join(errs...)
}

View on GitHub (pinned to 806f444784)

Solutions

Inspect the wrapped inner error: if it indicates the server is already closed, ensure Close is called exactly once (guard with sync.Once)
Check for hijacked/streaming connections (e.g. WebSocket upgrades through the proxy) and close/complete them before shutdown
Log and treat the remaining errors as non-fatal during shutdown — the joined error still reports all sub-system failures

Example fix

var closeOnce sync.Once
// before: Close may be called twice and fail on the second call
close(s.httpServer)
// after
closeOnce.Do(func() { _ = server.Close() })

Defensive patterns

Strategy: try-catch

Try / catch

if err := server.Close(); err != nil {
	// inspect joined sub-errors individually
	for _, e := range errors.Unwrap(err).([]error) {
		log.Warn("shutdown sub-error", zap.Error(e))
	}
}

Prevention

Call Close exactly once per server instance (sync.Once guard) to avoid double-close errors
Drain or complete hijacked/streaming connections (WebSockets) before Close
Treat shutdown errors as warnings — log them but do not block process exit

When it happens

Trigger: Calling Server.Close() (extension shutdown / SIGTERM path) when http.Server.Close() fails, which practically only happens if the server was already closed, its listener was force-closed, or an in-flight hijacked connection causes an error return from the stdlib.

Common situations: Double-invocation of Close (e.g. graceful-shutdown hook plus extension teardown both calling it); a listener hijacked by a streaming/upgrade connection resisting the abrupt close; process shutdown racing with the health-check or OTLP proxy connections.

Related errors

AI-assisted analysis of jaegertracing/jaeger@806f444784 (2026-09-01). Data as JSON: /api/errors/5c6be4e3f94a0a4f. Report an issue: GitHub.