How to include CorrelationId in microservice architecture? - microservices

Question

I am creating a microservices architecture using ASP.NET Core web api. All the services are decoupled from each other, and may be deployed in different environments. Every service has its own logging. When requests flows through these services it could fail in any of the service, We need a way of tracing a series of events back to the source, even if it means traversing multiple services.
So to handle this issue, the service that originates the request creates a CorrelationId and pass it to the next service. The 2nd service pass it to 3rd service and so on. If exception occurs the corresponding service will log the exception message along with CorrelationId.
I wanted to know what would be a best place for the caller of the service to pass the correlationid?
Should the caller pass correlationid in HttpHeader or should it pass it as a part method parameter something like below
This is the service that is getting called
public class RequestDTO
{
public string CorrelationId {get;set;}
public string SomeOtherData {get;set;}
}
public Service2Controller:Controller
{
public Task<in> DoSomething(RequestDTO request)
{
// add the correlationid in current request Items collection
// So global exception handling can access it and log it
// along with the exception
HttpContext.Items.Add("CorrelationId", request.CorrelationId);
}
}
in the approach above if there is an exception before this method is invoked, the CorrelationId will not be available for global exception handler for logging.
Any suggestions? or alternate approach

Answer 1

The correlation identifier should not be something that you add yourself, some framework that sends the messages should do this. This way a developer can't forget it and it has a consistent behaviour all over the place.
Put a MessageId and CorrelationId in the headers of the message. On first message, both will be the same. On second message, CorrelationId is set to the MessageId of the previous one.
You could also set a unique ConversationId which never changes, to track all messages spawning from one originator. Useful when using pub/sub and/or calling back to the originator of a message.

Answer 2

Generally the best approach is to
let a library do it for you and
to follow standards where available
For example, see here and here how Application Insights does dependency & correlation tracking across different levels, using headers. And see this answer for a good description of it too.