It all started when i was sitting with a colleague to resolve some application issue, when i noticed something interesting. He was merging the code and my eyes caught the attention of this class “org.apache.log4j.MDC”. This led to few discoveries on the way which follow here:
What is MDC?
MDC stands for Mapped Diagnostic Context. It helps you to distinguish inter-leaving logs from multiple sources. Let me explain in detail. When we have multiple user-requests coming in for a given servlet, each request of an user is serviced using a thread. This leaves multiple users logging to the same log file and the log
statements get inter-mixed. Now, to filter out logs of a particular user, we need to append the user-id to the log statements so that we can grep(search) them in the log file, to make some sense of it. An obvious way of logging, is to append the user-id in the log statements i.e. log.info(userId+” logged something “); A non-invasive way of logging is to use MDC. With MDC, you put the user-id in a context-map which is attached to the thread (of each user request) by the logger. MDC is thread-safe and uses a Map internally to store the context information.[Courtesy : Kalyan Dabburi]
How to use MDC?
a. Configure the information, which needs to be logged (user-id in this case) in the log4j.xml as part of ConversionPattern.
log4j.appender.consoleAppender.layout.ConversionPattern = %d %i - %m - %X{user-id}%nb. In your respective class, before you start processing the user request, place the actual user-id in the context(MDC).
MDC.put("user-id","SKRS786");c. Remove the context information from MDC at the end of the processing.
MDC.remove("user-id");Resources :
What is NDC ? Which one to use MDC or NDC?
NDC stands for Nested Diagnostic Context. It is a stack-based implementation of attaching context information. For all purposes, use MDC over NDC, as MDC is memory efficient. For a detailed comparison, click here.
Which logging framework to use? Log4J or SLF4J or logback?
For all new application development, use logback. logback is a run-time implementation of SLF4J. If you have an existing application with Log4J, it is still worth-while to switch to logback. For a detailed explanation, click here. To understand the evolution of logging in Java and JEE world, refer to this article by Micheal Andrews.
Reference: Effective Logging in Java/JEE from our JCG partner Srinivas Ovn at the Bemused blog.
I like the approach by adding context information into the MDC to trace real application behavior in production environment by log statements. To delegate the MDC to backend servers you should take a look into the TracEE-Framework ( http://tracee.io ). It wraps and delegates the MDC to (and from) the backend service as well. It’s cool to use your approach with the user-id in a microservice architecture for example.
Thank you for visiting and leaving behind useful info
I understand that your MDC concept . it is nice .
why not use ,this way for mult-user servlet ?
String userid = (String)session.getAttribute(“user-id”);
log.info(userid);
this statement give same output against ” MDC ” statement
Thank you for visiting the blog. The purpose is not to log userid once. If we have to log userid all across the user journey in the application along with a lot other operational info then your approach is untenable.