Executes nested tasks in parallel with no guarantees of thread safety. Every task will run in its own thread, with the likelihood of concurrency problems scaling with the number of CPUs on the host system.

Warning: While the Apache Ant core is believed to be thread safe, no such guarantees are made about tasks, which are not tested for thread safety during Ant's test process. Third party tasks may or may not be thread safe, and some of Ant's core tasks, such as <javac> are definitely not re-entrant. This is because they use libraries that were never designed to be used in a multithreaded environment.

The primary use case for <parallel> is to run external programs such as an application server, and the JUnit or TestNG test suites at the same time. Anyone trying to run large Ant task sequences in parallel, such as javadoc and javac at the same time, is implicitly taking on the task of identifying and fixing all concurrency bugs the tasks that they run.

Accordingly, while this task has uses, it should be considered an advanced task which should be used in certain batch processing or testing situations, rather than an easy trick to speed up build times on a multicore CPU.


Attribute Description Required
threadCount Maximum numbers of thread to use. No
threadsPerProcessor Maximum number of threads to use per available processor (Java 1.4+) No; defers to threadCount
timeout Number of milliseconds before execution is terminated No
failonany If any of the nested tasks fails, execution of the task completes at that point without waiting for any other tasks to complete. No; default is false.
pollInterval Currently has no effect No; default is 1000

Parallel tasks have a number of uses in an Ant build file including:

Any valid Ant task may be embedded within a parallel task, including other parallel tasks, though there is no guarantee that the tasks will be thread safe in such an environment.

While the tasks within the parallel task are being run, the main thread will be blocked waiting for all the child threads to complete. If execution is terminated by a timeout or a nested task failure when the failonany flag is set, the parallel task will complete without waiting for other nested tasks to complete in other threads.

If any of the tasks within the <parallel> task fails and failonany is not set, the remaining tasks in other threads will continue to run until all threads have completed. In this situation, the parallel task will also fail.

The parallel task may be combined with the sequential task to define sequences of tasks to be executed on each thread within the parallel block.

The threadCount attribute can be used to place a maximum number of available threads for the execution. When not present all child tasks will be executed at once. When present then the maximum number of concurrently executing tasks will not exceed the number of threads specified. Furthermore, each task will be started in the order they are given. But no guarantee is made as to the speed of execution or the order of completion of the tasks, only that each will be started before the next.

If you are using Java 1.4 or later you can also use the threadsPerProcessor and the number of available threads will be the started multiple of the number of processors (there is no affinity to a particular processor, however). This will override the value in threadCount. If threadsPerProcessor is specified on any older JVM, then the value in threadCount will be used as is.

When using threadCount and threadsPerProcessor care should be taken to ensure that the build does not deadlock. This can be caused by tasks such as waitfor taking up all available threads before the tasks that would unlock the waitfor would occur. This is not a replacement for Java Language level thread semantics and is best used for "embarrassingly parallel" tasks.

Parameters specified as nested elements


The parallel task supports a <daemons> nested element. This is a list of tasks which are to be run in parallel daemon threads. The parallel task will not wait for these tasks to complete. Being daemon threads, however, they will not prevent Ant from completing, whereupon the threads are terminated. Failures in daemon threads which occur before the parallel task itself finishes will be reported and can cause parallel to throw an exception. Failures which occur after parallel has completed are not reported.

Daemon tasks can be used, for example, to start test servers which might not be easily terminated from Ant. By using <daemons> such servers do not halt the build.


This is a typical pattern for testing a server application. In one thread the server is started (the <wlrun> task). The other thread consists of a three tasks which are performed in sequence. The <sleep> task is used to give the server time to come up. Another task which is capable of validating that the server is available could be used in place of the <sleep> task. The <junit> test harness then runs, again in its own JVM. Once the tests are complete, the server is stopped (using <wlstop> in this example), allowing both threads to complete. The <parallel> task will also complete at this time and the build will then continue.

  <wlrun ... >
    <sleep seconds="30"/>
    <junit fork="true" forkmode="once" ... >

Here, two independent tasks run to achieve better resource utilization during the build. In this instance, some servlets are being compiled in one thread and a set of JSPs is being precompiled in another. Developers need to be careful that the two tasks are independent, both in terms of their dependencies and in terms of their potential interactions in Ant's external environment. Here we set fork=true for the <javac> task, so that it runs in a new process; if the <wljspc> task used the javac compiler in-VM (it may), concurrency problems may arise.

  <javac fork="true"...> <!-- compiler servlet code -->
  <wljspc ...> <!-- precompile JSPs -->

This example represents a typical need for use of the threadCount and threadsPerProcessor attributes. Spinning up all 40 of those tasks could cripple the system for memory and CPU time. By limiting the number of concurrent executions you can reduce contention for CPU, memory and disk IO, and so actually finish faster. This is also a good candidate for use of threadCount (and possibly threadsPerProcessor) because each task is independent (every new JVM is forked) and has no dependencies on the other tasks.

 <macrodef name="dbpurge">
    <attribute file="file"/>
      <java jar="utils/dbpurge.jar" fork="true" >
        <arg file="@{file}/>

<parallel threadCount="4">
  <dbpurge file="db/one"/>
  <dbpurge file="db/two"/>
  <dbpurge file="db/three"/>
  <dbpurge file="db/four"/>
  <dbpurge file="db/five"/>
  <dbpurge file="db/six"/>
  <dbpurge file="db/seven"/>
  <dbpurge file="db/eight"/>
  <!-- repeated about 40 times -->