Properly shutdown the server

[byo]

When starting the shutdown connection use a new
context with specific timeout since the one used
in the server has already been canceled.

[nymd]

If both shutdowns fail the second send blocks forever and defer close(ec) never runs.

Cleanest fix is collect both errors and send once, which keeps the deal with the consumer as "at most one value, then close":

Suggested change

	<-s.Context.Done()
	err := serv.Shutdown(s.Context)
	shutdownCtx, cancel := context.WithTimeout(context.Background(), config.Server.ShutdownTimeout)
	defer cancel()
	err := serv.Shutdown(shutdownCtx)
	if err != nil {
	log.Warnw("failed shutdown", "err", err)
	log.Warnw("failed to shutdown", "err", err)
	ec <- err
	}
	err = metricsServ.Shutdown(shutdownCtx)
	if err != nil {
	log.Warnw("failed to shutdown metrics server", "err", err)
	ec <- err
	}
	<-s.Context.Done()
	shutdownCtx, cancel := context.WithTimeout(context.Background(), config.Server.ShutdownTimeout)
	defer cancel()
	var errs []error
	if err := serv.Shutdown(shutdownCtx); err != nil {
	log.Warnw("failed to shutdown finder server", "err", err)
	errs = append(errs, fmt.Errorf("finder server: %w", err))
	}
	if err := metricsServ.Shutdown(shutdownCtx); err != nil {
	log.Warnw("failed to shutdown metrics server", "err", err)
	errs = append(errs, fmt.Errorf("metrics server: %w", err))
	}
	if err := errors.Join(errs...); err != nil {
	ec <- err
	}

Needs errors added to the import block.

[byo]

Thanks, I analyzed the code further and found out that there could be more issues with the approach.

Updated the code to properly handle a case where there are multiple errors sent through the channel with extra unit tests to ensure out application / signal interface layer handles such case properly.

[nymd]

Elegant fix to do it consumer-side. Thanks.