To improve the use of opentracing, we should do some minor changes:

• Tags should only be used with data that can be used to filter between operations (per example, on analyze operation, we can add a tag with the output of IsResolved of the plan). Right now the query is set on a tag, move it to a span.Log()

• On analyze and analyze_once add a log with the plan of the node that is being processed.

• Be able to use the default jaeger environment variables instead of custom configuration. To do that, there is a tracer constructor on jaeger config package.

• Remove spans from Merge and Update aggregation methods, they are really verbose and say almost nothing, and add the span to Eval.

• Right now, we cannot follow the entire trace of a query because it is splitted into several parts with no relations between them; parse, analyze and the execution of the query itself. We must be able to trace at query level.

• The tracer name must be configurable, not hardcoded to go-mysql-server (this can be done using JAEGER_SERVICE_NAME)