FileDocCategorySizeDatePackage
NDC.javaAPI DocApache log4j 1.2.1513812Sat Aug 25 00:09:42 BST 2007org.apache.log4j

NDC

public class NDC extends Object
The NDC class implements nested diagnostic contexts as defined by Neil Harrison in the article "Patterns for Logging Diagnostic Messages" part of the book "Pattern Languages of Program Design 3" edited by Martin et al.

A Nested Diagnostic Context, or NDC in short, is an instrument to distinguish interleaved log output from different sources. Log output is typically interleaved when a server handles multiple clients near-simultaneously.

Interleaved log output can still be meaningful if each log entry from different contexts had a distinctive stamp. This is where NDCs come into play.

Note that NDCs are managed on a per thread basis. NDC operations such as {@link #push push}, {@link #pop}, {@link #clear}, {@link #getDepth} and {@link #setMaxDepth} affect the NDC of the current thread only. NDCs of other threads remain unaffected.

For example, a servlet can build a per client request NDC consisting the clients host name and other information contained in the the request. Cookies are another source of distinctive information. To build an NDC one uses the {@link #push push} operation. Simply put,

  • Contexts can be nested.

  • When entering a context, call NDC.push. As a side effect, if there is no nested diagnostic context for the current thread, this method will create it.

  • When leaving a context, call NDC.pop.

  • When exiting a thread make sure to call {@link #remove NDC.remove()}.

There is no penalty for forgetting to match each push operation with a corresponding pop, except the obvious mismatch between the real application context and the context set in the NDC.

If configured to do so, {@link PatternLayout} and {@link TTCCLayout} instances automatically retrieve the nested diagnostic context for the current thread without any user intervention. Hence, even if a servlet is serving multiple clients simultaneously, the logs emanating from the same code (belonging to the same category) can still be distinguished because each client request will have a different NDC tag.

Heavy duty systems should call the {@link #remove} method when leaving the run method of a thread. This ensures that the memory used by the thread can be freed by the Java garbage collector. There is a mechanism to lazily remove references to dead threads. In practice, this means that you can be a little sloppy and sometimes forget to call {@link #remove} before exiting a thread.

A thread may inherit the nested diagnostic context of another (possibly parent) thread using the {@link #inherit inherit} method. A thread may obtain a copy of its NDC with the {@link #cloneStack cloneStack} method and pass the reference to any other thread, in particular to a child.

author
Ceki Gülcü
since
0.7.0

Fields Summary
static Hashtable
ht
static int
pushCounter
static final int
REAP_THRESHOLD
Constructors Summary
private NDC()

  
  // No instances allowed.
Methods Summary
public static voidclear()
Clear any nested diagnostic information if any. This method is useful in cases where the same thread can be potentially used over and over in different unrelated contexts.

This method is equivalent to calling the {@link #setMaxDepth} method with a zero maxDepth argument.

since
0.8.4c

    Stack stack = getCurrentStack();    
    if(stack != null) 
      stack.setSize(0);
public static java.util.StackcloneStack()
Clone the diagnostic context for the current thread.

Internally a diagnostic context is represented as a stack. A given thread can supply the stack (i.e. diagnostic context) to a child thread so that the child can inherit the parent thread's diagnostic context.

The child thread uses the {@link #inherit inherit} method to inherit the parent's diagnostic context.

return
Stack A clone of the current thread's diagnostic context.

    Stack stack = getCurrentStack();
    if(stack == null)
      return null;
    else {
      return (Stack) stack.clone();
    }
public static java.lang.Stringget()
Never use this method directly, use the {@link org.apache.log4j.spi.LoggingEvent#getNDC} method instead.

    Stack s = getCurrentStack();
    if(s != null && !s.isEmpty()) 
      return ((DiagnosticContext) s.peek()).fullMessage;
    else
      return null;
private static java.util.StackgetCurrentStack()
Get NDC stack for current thread.

return
NDC stack for current thread.

      if (ht != null) {
          return (Stack) ht.get(Thread.currentThread());
      }
      return null;
public static intgetDepth()
Get the current nesting depth of this diagnostic context.

see
#setMaxDepth
since
0.7.5

    Stack stack = getCurrentStack();          
    if(stack == null)
      return 0;
    else
      return stack.size();
public static voidinherit(java.util.Stack stack)
Inherit the diagnostic context of another thread.

The parent thread can obtain a reference to its diagnostic context using the {@link #cloneStack} method. It should communicate this information to its child so that it may inherit the parent's diagnostic context.

The parent's diagnostic context is cloned before being inherited. In other words, once inherited, the two diagnostic contexts can be managed independently.

In java, a child thread cannot obtain a reference to its parent, unless it is directly handed the reference. Consequently, there is no client-transparent way of inheriting diagnostic contexts. Do you know any solution to this problem?

param
stack The diagnostic context of the parent thread.

    if(stack != null)
      ht.put(Thread.currentThread(), stack);
private static voidlazyRemove()

    if (ht == null) return;
     
    // The synchronization on ht is necessary to prevent JDK 1.2.x from
    // throwing ConcurrentModificationExceptions at us. This sucks BIG-TIME.
    // One solution is to write our own hashtable implementation.
    Vector v;
    
    synchronized(ht) {
      // Avoid calling clean-up too often.
      if(++pushCounter <= REAP_THRESHOLD) {
	return; // We release the lock ASAP.
      } else {
	pushCounter = 0; // OK let's do some work.
      }

      int misses = 0;
      v = new Vector(); 
      Enumeration enumeration = ht.keys();
      // We give up after 4 straigt missses. That is 4 consecutive
      // inspected threads in 'ht' that turn out to be alive.
      // The higher the proportion on dead threads in ht, the higher the
      // chances of removal.
      while(enumeration.hasMoreElements() && (misses <= 4)) {
	Thread t = (Thread) enumeration.nextElement();
	if(t.isAlive()) {
	  misses++;
	} else {
	  misses = 0;
	  v.addElement(t);
	}
      }
    } // synchronized

    int size = v.size();
    for(int i = 0; i < size; i++) {
      Thread t = (Thread) v.elementAt(i);
      LogLog.debug("Lazy NDC removal for thread [" + t.getName() + "] ("+ 
		   ht.size() + ").");
      ht.remove(t);
    }
public static java.lang.Stringpeek()
Looks at the last diagnostic context at the top of this NDC without removing it.

The returned value is the value that was pushed last. If no context is available, then the empty string "" is returned.

return
String The innermost diagnostic context.

    Stack stack = getCurrentStack();
    if(stack != null && !stack.isEmpty())
      return ((DiagnosticContext) stack.peek()).message;
    else
      return "";
public static java.lang.Stringpop()
Clients should call this method before leaving a diagnostic context.

The returned value is the value that was pushed last. If no context is available, then the empty string "" is returned.

return
String The innermost diagnostic context.

    Stack stack = getCurrentStack();
    if(stack != null && !stack.isEmpty()) 
      return ((DiagnosticContext) stack.pop()).message;
    else
      return "";
public static voidpush(java.lang.String message)
Push new diagnostic context information for the current thread.

The contents of the message parameter is determined solely by the client.

param
message The new diagnostic context information.

    Stack stack = getCurrentStack();
      
    if(stack == null) {
      DiagnosticContext dc = new DiagnosticContext(message, null);      
      stack = new Stack();
      Thread key = Thread.currentThread();
      ht.put(key, stack);
      stack.push(dc);
    } else if (stack.isEmpty()) {
      DiagnosticContext dc = new DiagnosticContext(message, null);            
      stack.push(dc);
    } else {
      DiagnosticContext parent = (DiagnosticContext) stack.peek();
      stack.push(new DiagnosticContext(message, parent));
    }
public static voidremove()
Remove the diagnostic context for this thread.

Each thread that created a diagnostic context by calling {@link #push} should call this method before exiting. Otherwise, the memory used by the thread cannot be reclaimed by the VM.

As this is such an important problem in heavy duty systems and because it is difficult to always guarantee that the remove method is called before exiting a thread, this method has been augmented to lazily remove references to dead threads. In practice, this means that you can be a little sloppy and occasionally forget to call {@link #remove} before exiting a thread. However, you must call remove sometime. If you never call it, then your application is sure to run out of memory.

    ht.remove(Thread.currentThread());
    
    // Lazily remove dead-thread references in ht.
    lazyRemove();
public static voidsetMaxDepth(int maxDepth)
Set maximum depth of this diagnostic context. If the current depth is smaller or equal to maxDepth, then no action is taken.

This method is a convenient alternative to multiple {@link #pop} calls. Moreover, it is often the case that at the end of complex call sequences, the depth of the NDC is unpredictable. The setMaxDepth method circumvents this problem.

For example, the combination 

void foo() {
  int depth = NDC.getDepth();

  ... complex sequence of calls

  NDC.setMaxDepth(depth);
}
ensures that between the entry and exit of foo the depth of the diagnostic stack is conserved.

see
#getDepth
since
0.7.5

    Stack stack = getCurrentStack();    
    if(stack != null && maxDepth < stack.size()) 
      stack.setSize(maxDepth);