Introduction

Log4j introduced the concept of the Mapped Diagnostic Context or MDC. It has been documented and discussed in numerous places including Log4j MDC: What and Why and Log4j and the Mapped Diagnostic Context. In addition, Log4j 1.x provides support for a Nested Diagnostic Context or NDC. It too has been documented and discussed in various places such as Log4j NDC. SLF4J/Logback followed with its own implementation of the MDC, which is documented very well at Mapped Diagnostic Context.

Log4j 2 continues with the idea of the MDC and the NDC but merges them into a single Thread Context. The Thread Context Map is the equivalent of the MDC and the Thread Context Stack is the equivalent of the NDC. Although these are frequently used for purposes other than diagnosing problems, they are still frequently referred to as the MDC and NDC in Log4j 2 since they are already well known by those acronyms.

Fish Tagging

Most real-world systems have to deal with multiple clients simultaneously. In a typical multithreaded implementation of such a system, different threads will handle different clients. Logging is especially well suited to trace and debug complex distributed applications. A common approach to differentiate the logging output of one client from another is to instantiate a new separate logger for each client. This promotes the proliferation of loggers and increases the management overhead of logging.

A lighter technique is to uniquely stamp each log request initiated from the same client interaction. Neil Harrison described this method in the book "Patterns for Logging Diagnostic Messages," in Pattern Languages of Program Design 3, edited by R. Martin, D. Riehle, and F. Buschmann (Addison-Wesley, 1997). Just as a fish can be tagged and have its movement tracked, stamping log events with a common tag or set of data elements allows the complete flow of a transaction or a request to be tracked. We call this Fish Tagging.

Log4j provides two mechanisms for performing Fish Tagging; the Thread Context Map and the Thread Context Stack. The Thread Context Map allows any number of items to be added and be identified using key/value pairs. The Thread Context Stack allows one or more items to be pushed on the Stack and then be identified by their order in the Stack or by the data itself. Since key/value pairs are more flexible, the Thread Context Map is recommended when data items may be added during the processing of the request or when there are more than one or two items.

To uniquely stamp each request using the Thread Context Stack, the user pushes contextual information on to the Stack.

ThreadContext.push(UUID.randomUUID().toString()); // Add the fishtag;

logger.debug("Message 1");
.
.
.
logger.debug("Message 2");
.
.
ThreadContext.pop();

The alternative to the Thread Context Stack is the Thread Context Map. In this case, attributes associated with the request being processed are adding at the beginning and removed at the end as follows:

ThreadContext.put("id", UUID.randomUUID().toString()); // Add the fishtag;
ThreadContext.put("ipAddress", request.getRemoteAddr());
ThreadContext.put("loginId", session.getAttribute("loginId"));
ThreadContext.put("hostName", request.getServerName());
.
logger.debug("Message 1");
.
.
logger.debug("Message 2");
.
.
ThreadContext.clear();

CloseableThreadContext

When placing items on the stack or map, it's necessary to remove then again when appropriate. To assist with this, the CloseableThreadContext implements the AutoCloseable interface. This allows items to be pushed to the stack or put in the map, and removed when the close() method is called - or automatically as part of a try-with-resources. For example, to temporarily push something on to the stack and then remove it:

// Add to the ThreadContext stack for this try block only;
try (final CloseableThreadContext.Instance ctc = CloseableThreadContext.push(UUID.randomUUID().toString())) {

    logger.debug("Message 1");
.
.
    logger.debug("Message 2");
.
.
}

Or, to temporarily put something in the map:

// Add to the ThreadContext map for this try block only;
try (final CloseableThreadContext.Instance ctc = CloseableThreadContext.put("id", UUID.randomUUID().toString())
                                                                .put("loginId", session.getAttribute("loginId"))) {

    logger.debug("Message 1");
.
.
    logger.debug("Message 2");
.
.
}

for( final Session session : sessions ) {
    try (final CloseableThreadContext.Instance ctc = CloseableThreadContext.put("loginId", session.getAttribute("loginId"))) {
        logger.debug("Starting background thread for user");
        final Map<String, String> values = ThreadContext.getImmutableContext();
        final List<String> messages = ThreadContext.getImmutableStack().asList();
        executor.submit(new Runnable() {
        public void run() {
            try (final CloseableThreadContext.Instance ctc = CloseableThreadContext.putAll(values).pushAll(messages)) {
                logger.debug("Processing for user started");
                .
                logger.debug("Processing for user completed");
            }
        });
    }
}

Implementation details

The Stack and the Map are managed per thread and are based on ThreadLocal by default. The Map can be configured to use an InheritableThreadLocal by setting system property log4j2.isThreadContextMapInheritable to true. When configured this way, the contents of the Map will be passed to child threads. However, as discussed in the Executors class and in other cases where thread pooling is utilized, the ThreadContext may not always be automatically passed to worker threads. In those cases the pooling mechanism should provide a means for doing so. The getContext() and cloneStack() methods can be used to obtain copies of the Map and Stack respectively.

Note that all methods of the ThreadContext class are static.

Including the ThreadContext when writing logs

The PatternLayout provides mechanisms to print the contents of the ThreadContext Map and Stack.

• Use %X by itself to include the full contents of the Map.
• Use %X{key} to include the specified key.
• Use %x to include the full contents of the Stack.

Custom context data injectors for non thread-local context data

With the ThreadContext logging statements can be tagged so log entries that were related in some way can be linked via these tags. The limitation is that this only works for logging done on the same application thread (or child threads when configured).

Some applications have a thread model that delegates work to other threads, and in such models, tagging attributes that are put into a thread-local map in one thread are not visible in the other threads and logging done in the other threads will not show these attributes.

Log4j 2.7 adds a flexible mechanism to tag logging statements with context data coming from other sources than the ThreadContext. See the manual page on extending Log4j for details.