Config validation is broken

[pveentjer]

There is no good working validation of configuration when members join. See the following example:

public class Main {

    public static void main(String[] args){
        Config config1 = new Config();
        config1.setProperty(GroupProperties.PROP_PARTITION_COUNT,"100");
        Config config2 = new Config();

        HazelcastInstance hz1 = Hazelcast.newHazelcastInstance(config1);
        HazelcastInstance hz2 = Hazelcast.newHazelcastInstance(config2);

        IMap map = hz1.getMap("map");
        for(int k=0;k<1000;k++){
            map.put(UUID.randomUUID().toString(),"10");
        }
        System.out.println("done");
    }
}

The following logging:

java.lang.ArrayIndexOutOfBoundsException: 100
    at com.hazelcast.partition.impl.InternalPartitionServiceImpl.allocateReplicas(InternalPartitionServiceImpl.java:568)
    at com.hazelcast.partition.impl.InternalPartitionServiceImpl.finalizeOrRollbackMigration(InternalPartitionServiceImpl.java:562)
    at com.hazelcast.partition.impl.InternalPartitionServiceImpl.processPartitionRuntimeState(InternalPartitionServiceImpl.java:545)
    at com.hazelcast.partition.impl.PartitionStateOperation.run(PartitionStateOperation.java:51)
    at com.hazelcast.spi.impl.BasicOperationService$OperationHandler.handle(BasicOperationService.java:672)
    at com.hazelcast.spi.impl.BasicOperationService$OperationHandler.access$400(BasicOperationService.java:648)
    at com.hazelcast.spi.impl.BasicOperationService$OperationPacketHandler.handle(BasicOperationService.java:622)
    at com.hazelcast.spi.impl.BasicOperationService$OperationPacketHandler.handle(BasicOperationService.java:590)
    at com.hazelcast.spi.impl.BasicOperationService$OperationPacketHandler.access$1400(BasicOperationService.java:577)
    at com.hazelcast.spi.impl.BasicOperationService$BasicDispatcherImpl.dispatch(BasicOperationService.java:532)
    at com.hazelcast.spi.impl.BasicOperationScheduler$OperationThread.process(BasicOperationScheduler.java:439)
    at com.hazelcast.spi.impl.BasicOperationScheduler$OperationThread.processPriorityMessages(BasicOperationScheduler.java:452)
    at com.hazelcast.spi.impl.BasicOperationScheduler$OperationThread.doRun(BasicOperationScheduler.java:432)
    at com.hazelcast.spi.impl.BasicOperationScheduler$OperationThread.run(BasicOperationScheduler.java:408)

But this exception is internal catched and logged; so the application keeps running.

[pveentjer]

The Config.isCompatible is not called apart from in a unit test. So currently there is no validation.

[pveentjer]

Apparently it is broken since Hazelcast 3.0.

In 2.x the whole config was send to the remote machine. In 3.x the ConfigCheck is used and the Config.isCompatible isn't used at all. The ConfigCheck doesn't cover enough fields to make it useful.

[pveentjer]

There is another bug.

If the cluster configuration doesn't match, instead of letting the joining member crash, it will create its own cluster. This is a violation of the fail fast principle. So we need to get this fixed as well.