Fork me on GitHub

Shutdown of Forked JVM

Embedded shutdown

After the test-set has completed, the process executes java.lang.System.exit(0) which starts shutdown hooks. At this point the process may run next 30 seconds until all non daemon Threads die. After the period of time has elapsed, the process kills itself by java.lang.Runtime.halt(0). The timeout of 30 seconds can be customized by configuration parameter forkedProcessExitTimeoutInSeconds.

Pinging forked JVM

Since Surefire Plugin 2.20.1 ping is platform dependent and fallbacks to old mechanism if PID of Maven process or platform is not recognized, native commands fail in Java.

Simply the mechanism checks the Maven PID is still alive and it is not reused by OS in another application. If Maven process has died, the forked JVM is killed.

Implementation: The Maven PID is determined by:

  • Java 9 call ProcessHandle.current().pid() , or
  • resolving PID from /proc/self/stat on Linux and /proc/curproc/status on BSD, or
  • the JMX call ManagementFactory.getRuntimeMXBean().getName() .

On Unix like systems the process' uptime is determined by native command (/usr)/bin/ps -o etime= -p [PID] .

On Windows the start time is determined using wmic process where (ProcessId=[PID]) get CreationDate in the forked JVM.

Since Surefire Plugin 2.19 the old mechanism is significantly slower:

The master process sends NOOP command to a forked JVM every 10 seconds. Forked JVM is waiting for the command every 20 seconds (changed to 30 seconds since version 2.20.1, see SUREFIRE-1302). If the master process is killed (received SIGKILL signal) or shutdown (pressed CTRL+C, received SIGTERM signal), forked JVM is killed after timing out waiting period.

NOTICE: After version 3.0.0-M4 this feature has to be enabled by setting the configuration parameter enableProcessChecker to ping or all.

Shutdown of forked JVM by stopping the build

After the master process of the build is shutdown by sending SIGTERM signal or pressing CTRL+C, the master process immediately sends SHUTDOWN command to every forked JVM. By default (configuration parameter shutdown=testset) forked JVM would not pick up a new test which means that the current test may still continue to run. The SIGTERM signal triggers Java shutdownhook which executes java.lang.Process.destroy() in the forked JVM (not always reliable depending on VM and OS). The parameter shutdown can be configured with other two values exit and kill. Using exit forked JVM executes java.lang.System.exit(1) after the master process has received SIGTERM. Using kill the JVM executes java.lang.Runtime.halt(1), example:

<project>
  [...]
  <build>
    <plugins>
      <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-surefire-plugin</artifactId>
        <version>3.2.5</version>
        <configuration>
          <shutdown>kill</shutdown>
        </configuration>
      </plugin>
    </plugins>
  </build>
  [...]
</project>

NOTICE: After version 3.0.0-M4 this feature has to be enabled by setting the configuration parameter enableProcessChecker to ping or native or all.

Shutdown of forked JVM after certain timeout

Using parameter forkedProcessTimeoutInSeconds forked JVMs are killed separately after every individual process elapsed certain amount of time and the whole plugin fails with the error message:

There was a timeout in the fork

Crashed forked JVM caused listing the crashed test(s)

After the JVM exited abruptly, the console lists the message Crashed tests: with a list of crashed tests if the entire test-set has not been yet completed. This happens if a test exited, killed JVM or a segmentation fault crashed JVM. In such cases you may be interested in dump files generated in reports directory, see FAQ.