When no driver is started, and the log level is set to DEBUG for the package org.jppf.management and trying to connect via JMXDriverConnectionWrapper, the following are shown in the log:
These exceptions are harmless, as they are indeed caught and handled in the JPPF code, however they may cause some worries:
* the InterruptedException is logged as a warning. This is wrong, because it is expected when the connection attempts fail after a specified timeout. This should be logged at TRACE level
* the NullPointerException results from poor handling in the JPPFJMXConnector code, this must be fixed.
From [https://www.jppf.org/forums/index.php/topic,8057.0.html '''this forums thread''']:
The method [https://www.jppf.org/javadoc/6.0/org/jppf/management/JMXDriverConnectionWrapper.html#getAllJobIds() getAllJobIds()] still exists in class [https://www.jppf.org/javadoc/6.0/index.html?org/jppf/management/JMXDriverConnectionWrapper.html JMXDriverConnectionWrapper], but it was removed from the [https://www.jppf.org/javadoc/6.0/index.html?org/jppf/server/job/management/DriverJobManagementMBean.html DriverJobManagementMBean] interface.
Using this method on a connected JMX wrapper always raises an exception:
We should remove this method from JMXDriverConnectionWrapper as well and update the documentation to reflect that, and in particular state that to achieve the same goal the following should be used:
Following feature request JPPF-563, create a new sample in the samples pack which demonstrates how to start a driver, node and client programmatically, all embedded within the same JVM. The sample will show the following functionalities:
* embedded driver life cycle: create, start, stop
* embedded node life cycle: create, start, stop
* connecting a client and submitting a job
* programmatically creating the configuration for a driver, node and client
* using management and monitoring APIs for an embedded driver and node
We propose that the following features be either deprecated or dropped altogether:
'''1. .Net integration'''
This feature relies heavily on the [http://jni4net.com/ '''jni4net'''] framework, which hasn't seen a new version in 4 years. Following the switch to Java 8 (feature request JPPF-548), its .Net proxy generator is no longer fully working, as it doesn't handle new Java 8 constucts such as default methods in interfaces, It is currently not possible to build it with the current code, and I don't see any solution that can be mainained in the long term. I propose to drop this feature from JPPF 6.1 forward. We will still maintain it for prior versiosn.
'''2. Android integration'''
The switch to Java 8 requires a lot of changes to the Android port, including, but definitely not limited to, the min Android sdk version and build tools. I haven't evaluated the changes that need to be done to the code itself, and, given the lack of bandwith (I'm just 1 developer), I tend to think it should be dropped so we can focus on more modern features such as job streaming and big data. If anyone volunteers to take this feature on, I'll be happy to assist in any way, In any case, we'll keep maintaining it for versions up to 6.0.
'''3. Persistent data in NodeRunner'''
The [https://www.jppf.org/javadoc/6.1/index.html?org/jppf/node/NodeRunner.html '''NodeRunner'''] class provides 3 methods '''setPersistentData()''', '''getPersistentData()''' and '''removePersistentData()''', which were intended for tasks to be able to store, access and manage data on the nodes accross job executions. These methods are inherently dangerous, because they can cause the nodes to retain objects and classes from many different class loaders, resulting in class loader leaks and potential out-of-memory conditions. This feature isn't used anymore in the JPPF code, and I believe now's a good time to remove it.
'''4. Node security policy'''
I can no longer see the benefit of the [https://www.jppf.org/doc/6.0/index.php?title=Node_configuration#Security_policy '''security policy'''] in the node configuration. We haven't touched this code in years, the default node security policy in the node distribution is no longer close to being useful or even accurate, and this can be easily replaced with a standard SecurityManager and associated security policy file. This feature should definitely be removed.
Current the [https://www.jppf.org/javadoc/6.1/index.html?org/jppf/server/JPPFDriver.html '''JPPFDriver'''] is implemented as a singleton. This means there can be only one per JVM. The same is true for the nodes, whether local to the driver's JVM or remote.
We propose to change that and make it possible to have any number of drivers and/or nodes per JVM. They might have the possibility to share some common resources, for example the NIO thread pool.
Currently the [https://www.jppf.org/doc/6.1/index.php?title=Execution_Policy_Elements#Preference '''Preference'''] execution poliy is applied to each node individually and is identical to the [https://www.jppf.org/doc/6.1/index.php?title=Execution_Policy_Elements#OR '''OR'''] execution policy. To sum up: despite its name, it has nothing to do with a "preference".
We propose to make it live up to its name, which implies:
* it should define a real order of preference for a number of node execution policies, where a node that satisfies policy N in the list will have priority over a node that satisfies policy N + 1
* it should be applied globally to all the nodes available to the driver
* because of the previous point, it should be a separate attribute of the job SLA
* it should be applicable to the client-side as well, where it would define a driver preference rather than a node preference
* special care should be taken about perfomance, as the algorithm will be in O(nbNodes * nbJobs). Should we allow parallel (with regards to the nodes) evaluation of the policy for each job?