When to use 'nested diagnostic context' (NDC)?

Question

Playing with log4net, I have seen the possibility to use a per-thread stack of context labels called the NDC.

The labels pushed on this stack are displayed in a Pat

Answer 1

In a server application such as ASP.NET.

For example, you can push information about the current request on to the NDC.

Answer 2

These feature come in handy when you have lots of logs to go through. When would you have lots of logs? Diagnosing weird bug on a production system with interleaving outputs. Having more contexts gives you way to filter the output or not outputting unneeded logs.

Another case nested contexts could be useful is if a method or some feature is called several times in different contexts and you need a way to distinguish between them.

Answer 3

Want an example?

Take the following Web API written using ASP.NET MVC4:

// GET api/HypervResource
public string Get()
{
    logger.Debug("Start of service test");
    System.Threading.Thread.Sleep(5000); // simulate work
    logger.Debug("End of service test");
    return "HypervResource controller running, use POST to send JSON encoded RPCs";
}

When server concurrent HTTP Requests are made, the logging can get interleaved. E.g.

2013-06-27 13:28:11,967 [10] DEBUG HypervResource.WmiCalls [(null)] - Start of service test
2013-06-27 13:28:12,976 [12] DEBUG HypervResource.WmiCalls [(null)] - Start of service test
2013-06-27 13:28:14,116 [13] DEBUG HypervResource.WmiCalls [(null)] - Start of service test
2013-06-27 13:28:16,971 [10] DEBUG HypervResource.WmiCalls [(null)] - End of service test
2013-06-27 13:28:17,979 [12] DEBUG HypervResource.WmiCalls [(null)] - End of service test
2013-06-27 13:28:19,119 [13] DEBUG HypervResource.WmiCalls [(null)] - End of service test

In this simple example, you could use the thread id to distinguish requests, but that can get tricky as the log file grows in complexity.

A better alternative is to provide unique identifiers that group together log messages for the same request. We can update the code as to the following:

// GET api/HypervResource
public string Get()
{
    using(log4net.NDC.Push(Guid.NewGuid().ToString()))
    {
        logger.Debug("Start of service test");
        System.Threading.Thread.Sleep(5000); // simulate work
        logger.Debug("End of service test");
        return "HypervResource controller running, use POST to send JSON encoded RPCs";
    }
}

This produces a log that you can grep to see the issues associated with a specific request. E.g.

2013-06-27 14:04:31,431 [11] DEBUG HypervResource.WmiCalls [525943cb-226a-43c2-8bd5-03c258d58a79] - Start of service test
2013-06-27 14:04:32,322 [12] DEBUG HypervResource.WmiCalls [5a8941ee-6e26-4c1d-a1dc-b4d9b776630d] - Start of service test
2013-06-27 14:04:34,450 [13] DEBUG HypervResource.WmiCalls [ff2246f1-04bc-4451-9e40-6aa1efb94073] - Start of service test
2013-06-27 14:04:36,434 [11] DEBUG HypervResource.WmiCalls [525943cb-226a-43c2-8bd5-03c258d58a79] - End of service test
2013-06-27 14:04:37,325 [12] DEBUG HypervResource.WmiCalls [5a8941ee-6e26-4c1d-a1dc-b4d9b776630d] - End of service test
2013-06-27 14:04:39,453 [13] DEBUG HypervResource.WmiCalls [ff2246f1-04bc-4451-9e40-6aa1efb94073] - End of service test

Answer 4

NDC.Push has been deprecated. The preferred way now (ThreadContext.Stacks["NDC"]) is this:

var disposable = ThreadContext.Stacks["NDC"].Push("context");
try
{
  Log.Info("begin"); // optional, but nice
  ...
}
finally
{
  Log.Info("end"); // optional, but nice
  disposable.Dispose();
}

Remember to check your conversion pattern so that it includes %property{NDC}:

<layout type="log4net.Layout.PatternLayout">
  <conversionPattern
    value="%date [%2thread] %-5level [%property{NDC}] - %.10240message%newline" />
</layout>