[FELIX-6162] ConfigurationManager crashes on shutdown if PersistenceManager not yet available

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: configadmin-1.9.18
• Component/s: Configuration Admin
• Assignee: Carsten Ziegeler
• Reporter: Tobias Gunkel
• Created: 24/Jul/19 13:15
• Updated: 20/Jul/20 06:33
• Resolved: 26/Jul/19 05:38

Description

If ConfigurationManager shuts down, it tries to close the managedServiceFactoryTracker and managedServiceTracker.

ConfigurationManager.stop():

        // stop handling ManagedService[Factory] services
        managedServiceFactoryTracker.close();
        managedServiceTracker.close();

In integration tests both services might not available at this time and it crashes:

java.lang.NullPointerException
             at org.apache.felix.cm.impl.ConfigurationManager.stop(ConfigurationManager.java:222)
             at org.apache.felix.cm.impl.ConfigurationAdminStarter.deactivate(ConfigurationAdminStarter.java:95)
             at org.apache.felix.cm.impl.DependencyTracker.stop(DependencyTracker.java:112)
             at org.apache.felix.cm.impl.Activator.stop(Activator.java:160)
             at org.apache.felix.framework.util.SecureAction.stopActivator(SecureAction.java:720)
             at org.apache.felix.framework.Felix.stopBundle(Felix.java:2795)

++

             at org.apache.felix.framework.Felix.setActiveStartLevel(Felix.java:1557)
             at org.apache.felix.framework.FrameworkStartLevelImpl.run(FrameworkStartLevelImpl.java:308)
             at java.lang.Thread.run(Thread.java:748)

The reason for this is that managedServiceTracker and managedServiceFactoryTracker are assigned after service registration of ConfigurationAdmin in the start() method.
The integration test is (sometimes) directly run after the bundleContext.registerService(ConfigurationAdmin.class, ...) line. And directly after the integration test execution the Configuration.stop() method is called where it crashes as managedServiceTracker and managedServiceFactoryTracker are still null.

ConfigurationManager.start():

        configurationAdminRegistration = bundleContext.registerService(ConfigurationAdmin.class, caf,
                serviceProperties);

        // start handling ManagedService[Factory] services
        managedServiceTracker = new ManagedServiceTracker(this);
        managedServiceFactoryTracker = new ManagedServiceFactoryTracker(this);

Although this sounds like a rare race-condition, it is reproducible in our test environment. About 30% of the tests fail with the above exception.

A simple solution would be to just add null-checks (as with every other service that is accessed in the stop() method):
ConfigurationManager.stop():

        // stop handling ManagedService[Factory] services
        if (managedServiceFactoryTracker != null)
        {
            managedServiceFactoryTracker.close();
        }
        if (managedServiceTracker != null)
        {
            managedServiceTracker.close();
        }