An attempt to solve the 'company standard log library' problem with a very simple logging interface for go built around context values.

I have worked in a few large Go teams. Each one of them had a pkg or std github repo containing the team's standard for basic go tasks like logging, tracing, http (circuit breaking), sqs wrappers etc etc. As a rule, I always encourage these to be individual independent components, which could be open-source and reusable for many teams. The problem has always been how these standard components log without tightly coupling them to the logging library itself.

The popular logging frameworks for go cannot be described purely as an interface. e.g., Logrus uses *logrus.Fields

Including log information from the context. e.g. trace headers.

Assuming a middleware stack, include this as or in a middleware after the context value is created:

ctx = log.WithField(ctx, "trace", trace.FromContext(ctx))

Later calls to log.Info(ctx, "hello") will include the trace header

Libraries shouldn't log. But they need to. But they shouldn't.

Solution: Include a minimal logging interface in your library, which defaults to do nothing, but can be wired in to a real logger.

type SQSWorker struct {
	Logger interface{
		Info(context.Context, msg)
		WithFields(context.Context, map[string]interface{}) context.Context
	}
}

The logger interface is designed to be copied (not referenced) in libraries.