ow2-proactive / scheduling

Multi-platform Scheduling and Workflows Engine

ProActive Workflows & Scheduling

You can download binaries and access a trial platform for free at:

http://activeeon.com/register/web-download

Quick start

To run ProActive Scheduler (server part):

$> ./bin/proactive-server

This will start all components, including 4 local nodes and Web portals. URLs to access them will be displayed in the server's output.

From here you can submit a job to the Scheduler, using for instance the XML files in samples/proactive-examples.

To run the command line client:

$> ./bin/proactive-client

To start a node:

$> ./bin/proactive-node -r SCHEDULER_URL

Building from sources

You can build a distribution that contains all binaries with ./gradlew build (linux/mac) or gradlew build (windows).

This will produce the following archives:

build/distributions/
├── scheduling-XXXXX-SNAPSHOT.tar
└── scheduling-XXXXX-SNAPSHOT.zip

Documentation

http://doc.activeeon.com

Enjoy ProActive Workflows & Scheduling!

scheduling's Issues

Insfrastructure Manger - node verifying scripts

Original issue created by Germain SIGETY on 22, Oct 2007 at 13:46 PM – SCHEDULING-36

---

The Infrastruture Manager doesn’t memorize results for nodes who have answered “no” to the verifying scripts. Each node have a hash map to save results of verifying scripts. After a verification by a script, the result is put in this hashmap. If the node haven’t succed the verifying test (by the script), on a second test, the Hashmap is still empty.

disconnecting reconnecting from the Scheduler Eclipse plugin doesn't work

Original issue created by Fabien Viale on 22, Oct 2007 at 18:58 PM – SCHEDULING-33

---

After we click on disconnect on the Scheduler interface, we can’t reconnect to it (window remains on “disconnected” state)

scheduler's public API's are not marked with @PublicAPI annotation

Original issue created by Mario Leyton on 24, Oct 2007 at 10:24 AM – SCHEDULING-42

---

Use the @PublicAPI annotation to clearly mark classes that are ment for user APIs.

Client behavior on job selection is incoherent

Original issue created by Clement Mathieu on 17, May 2008 at 16:20 PM – SCHEDULING-27

---

I have 2 jobs running. The first one is selected and I choose to display its tasks. Then I select the second job. tasks view remains enabled. Then I select the first one again. The view is now the job output.

This behavior is incoherent. If the user selected a view, this view should not be changed when a new job is selected.

wrap Scheduler and Ressource Manager into component

Original issue created by Cédric Dalmasso on 19, Dec 2007 at 16:45 PM – SCHEDULING-114

---

wrap Scheduler and Ressource Manager into component to allow the use of them by other component in GCM application.

The main idea is to expose server interfaces corresponding to the Scheduler and RM public API.

Task view brocken in the Scheduler GUI

Original issue created by Salageanu Emil on 28, Apr 2008 at 15:54 PM – SCHEDULING-4

---

The tasks tab is not functional. No refresh is done inside this tab.

org.objectweb.proactive.extensions.scheduler.policy.PolicyInterface should be refactored

Original issue created by Fabien Viale on 10, Jan 2008 at 11:41 AM – SCHEDULING-10

---

The PolicyInterface should be in the PublicAPI because it’s meant to be user-defined,

It contains references to JobDescriptor and EligibleJobDescriptor classes which contain themselves references to many Internal Scheduler Classes (InternalJob, InternalTask, …).

Theses classes should be refactored for a more transparent usage.

Security mechanism decouples active object creation from instantiation

Original issue created by Mario Leyton on 06, Nov 2007 at 09:39 AM – SCHEDULING-18

---

In Proactive the thread that instantiates an active object is not the same as the thread that will serve requests.

Since the scheduler security mechanism is based on threads, UserSchedulerInterface references can not be created inside the active object’s constructor. Instead, programmers have to pass a SchedulerAuthenticationInterface to the constructor, and then implement the initActivity method where the UserSchedulerInterface reference can be created.

The result of this workaround is that scheduler connection exceptions are not detected during the construction of the active object, and can not be communicated from inside the initActivity method, because this method does not throw exceptions.

Scheduler client JVM crashes when submitting incorrect XML file

Original issue created by Fabrice Huet on 08, Feb 2008 at 13:10 PM – SCHEDULING-24

---

When submitting an incorrect XML file to the scheduler (e.g having 2 different tasks with the same name), an error is reported but the scheduler crashes.

ERROR:“Xmen-Task” is used as an ID value more than once. at line 98, column 1
1 mistakes.
*\\* glibc detected \* /home/fhuet/NOSAVE/Test_Giovanni/Scheduler/Scheduler: double free or corruption (!prev): 0×0000000000534ca0 \* === Backtrace: === /lib64/libc.so.6(link: 0x3447070412) /home/fhuet/NOSAVE/Test_Giovanni/Scheduler/Scheduler(realpath+0x3c2)(link: 0x4018ea) /lib64/libc.so.6(__libc_start_main+0xf4)(link: 0x344701dab4) /home/fhuet/NOSAVE/Test_Giovanni/Scheduler/Scheduler(strcat+0x42)(link: 0x40157a) === Memory map: ==

job@name should be job@id

Original issue created by Clement Mathieu on 17, May 2008 at 15:42 PM – SCHEDULING-35

---

A common pratice is to use “id” for id attributes.

An XML editor will warn you but most of users edit XML file with notepad…

Resource manager - No lookup for a node after the node went down

Original issue created by Salageanu Emil on 20, May 2008 at 10:49 AM – SCHEDULING-50

---

I monitor a node X on a machine N. I see the node in the resource manager client. I reboot the machine M and restart the same node.
The resource manager will never perform a new lookup for this node so it will always be “down” unless removed and added again by hand.

ExecutableTask and ExecutableJavaTask have misleading names

Original issue created by Mario Leyton on 24, Oct 2007 at 10:21 AM – SCHEDULING-43

---

I suggest renaming ExecutableTask → Executable, and ExecutableJavaTask → ExecutableJava since they are not really Tasks (do not extend the Task class)

a reifed method, getResult(), could return null

Original issue created by Clement Mathieu on 16, Oct 2007 at 17:29 PM – SCHEDULING-31

---

In the Scheduler.java, the method getResult() could return null. As it’s a refied method called asynchronously (with
a future), the behavior is not correct.

InfrsatructureManager should become ResourceManager

Original issue created by Christian Delbé on 05, Dec 2007 at 17:01 PM – SCHEDULING-41

---

No description

Bad handling of broken job XML file: Exception thrown on stdout

Original issue created by Clement Mathieu on 17, May 2008 at 15:29 PM – SCHEDULING-26

---

If a job does not define at least one previous result, then an exception is thrown on stdout and the job is rejected. Users MUST be warned that their XML file is not valid (with a clear explanation if possible)

---

schubby$ Job : job_denoise
Job description = Run Junit Tests with a random JDK
name = junit
desc = Plop
final = false
reRun = 1
org.objectweb.proactive.extensions.scheduler.common.exception.SchedulerException: You must specify at least on precious result in your job !
at org.objectweb.proactive.extensions.scheduler.job.InternalJobFactory.createJob(InternalJobFactory.java:160)
at org.objectweb.proactive.extensions.scheduler.job.InternalJobFactory.createJob(InternalJobFactory.java:79)
at org.objectweb.proactive.extensions.scheduler.core.SchedulerFrontend.submit(SchedulerFrontend.java:232)
at sun.reflect.GeneratedMethodAccessor296.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
at java.lang.reflect.Method.invoke(Method.java:597)
at org.objectweb.proactive.core.mop.MethodCall.execute(MethodCall.java:388)
at org.objectweb.proactive.core.body.request.RequestImpl.serveInternal(RequestImpl.java:237)
at org.objectweb.proactive.core.body.request.RequestImpl.serve(RequestImpl.java:181)
at org.objectweb.proactive.core.body.BodyImpl$ActiveLocalBodyStrategy.serveInternal(BodyImpl.java:496)
at org.objectweb.proactive.core.body.BodyImpl$ActiveLocalBodyStrategy.serve(BodyImpl.java:465)
at org.objectweb.proactive.core.body.AbstractBody.serve(AbstractBody.java:938)
at org.objectweb.proactive.Service.blockingServeOldest(Service.java:194)
at org.objectweb.proactive.Service.blockingServeOldest(Service.java:169)
at org.objectweb.proactive.Service.fifoServing(Service.java:145)
at org.objectweb.proactive.core.body.ActiveBody$FIFORunActive.runActivity(ActiveBody.java:289)
at org.objectweb.proactive.core.body.ActiveBody.run(ActiveBody.java:174)
at java.lang.Thread.run(Thread.java:619)

Selection scripts: "selected" variable

Original issue created by Clement Mathieu on 17, May 2008 at 16:26 PM – SCHEDULING-3

---

About selection script the documentation says:
“selection script The selection script is always executed before the task itself on any candidate node: the execution of a selection script must set the boolean variable selected, that indicates if the candidate node is suitable for the execution of the associated task.”

But the example included within the scheduler source code, does not declare or set a such variable.

Anyway the documentation should include some selection scripts examples and templates. I’m not able to understand how to write a selection script in Java, JS or python.

Resource Manager Product interface freezes when adding a non-vaid node

Original issue created by Salageanu Emil on 28, May 2008 at 13:26 PM – SCHEDULING-49

---

On the RM Product interface: I add a valid node “Node” to the rm. The “Node” is located on a win xp machine (and has been started with “startNode.bat” on that machine).
I remove the node from the rm interface (which kills the node on the remote machine).
then, on the rm interface I add again the node “Node” Normally, the rm should tell me the node is not reachable, instead, the RM Interface freezes for a long time (several minutes) before reaching a stable state again. this problem seems to come from the rm itself (and not from the interface) because if I kill the interface and start it again, it will still take severeal minutes to conect to the rm.

Client RCP displays a stacktrace on console when user authentication fail

Original issue created by Clement Mathieu on 11, Jun 2008 at 16:45 PM – SCHEDULING-5

---

A pop-up is displayed AND a stacktrace is printed on the console.

The stacktrace should NOT be printed.

accessing the Task name in user code

Original issue created by Fabien Viale on 22, Oct 2007 at 13:33 PM – SCHEDULING-22

---

it would be nice to have access to the task name when we write a user-defined JavaTask.

For example, for logging purpose

Selection scripts: checkExecutable should be built-in

Original issue created by Clement Mathieu on 17, May 2008 at 16:29 PM – SCHEDULING-16

---

Checking for available file or executable is such a common use case, that checkExecutable should be built-in.

You should also ship template scripts to check the version of a tool (of course it has to be tuned for each tool)

The method JobResult.addTaskResult(String taskName, TaskResult taskResult); should not be part of the user accesible API.

Original issue created by Mario Leyton on 24, Oct 2007 at 10:42 AM – SCHEDULING-25

---

The method public void addTaskResult(String taskName, TaskResult taskResult); should not be part of the user accesible API. I suggest moving this object into the implementation class.

Exceptions from a non-final task are returned to the user inside a TaskResult object

Original issue created by Mario Leyton on 24, Oct 2007 at 10:31 AM – SCHEDULING-23

---

If an exception takes place inside a non-final task, and the workflow is configured to stop when an exception takes place. Then this exception is placed into a TaskResult object inside the JobResult.

The problem is that the programmer is not expectingt TaskResult from a non-final task. Instead it should be possible to do something like:

boolean JobResult.hasUnexpectedExceptions();
Collection JobResult.getUnexpectedExceptions();

nameclash between a scheduler Job and proactive.Job

Original issue created by Mario Leyton on 24, Oct 2007 at 10:44 AM – SCHEDULING-11

---

I suggest renaming the class Job to JobScheduler or something, since Job has a name clash with org.objectweb.proactive.Job; which can be confusing for ProActive beginners. Furthermore, the eclipse automatic import tends to import org.objectweb.proactive.job.

Published API

Original issue created by Clement Mathieu on 28, Nov 2007 at 14:06 PM – SCHEDULING-40

---

For the 3.9 release only classes tagged with the @PublicAPI will be exported to the user (documented).

You have to:
- Add missing annotations (or remove)
- check the API is consistent and well designed
- check that each method/class is documented

Javadoc can be generated with ./build javadoc (html files are in ../docs/api) or will be available soon in hudson https://galpage-exp.inria.fr:8181/hudson/job/ProActive_Documentation/lastSuccessfulBuild/artifact/trunk/docs/api/

Fix for system env vars does not work with RCP client.

Original issue created by Christian Delbé on 23, Jun 2008 at 18:51 PM – SCHEDULING-46

---

It seems that the path exported by the RCP client is empty : as a consequence, the native jobs does not work when submitted from the RCP client.

Scheduler documentation improvements

Original issue created by Clement Mathieu on 17, May 2008 at 17:04 PM – SCHEDULING-9

---

Actually it’s really hard to find the information that you need (example: Users don’t care about communicator.sh or ressource manager) Documentation must be task focused. Users know what they want to do and have to learn how to do it.

Currently both user and admin documentation are mixed in chapter 27.

The first step is to have two separate manuals. One for users, one for administrators.

User manual should be fully rewritten. Information is spread all over the chapter with multiple indirections. When I want to write a job descriptor, I don’t want to read API documentation. You can add three sub chapter. Introduction to job and tasks, job descriptors, job API. You must add standalone self contained examples for each common use case. NEVER split an XML file or source code in several boxes. It’s unreadable. Explain quickly how it works, give an example then comment it. MATLAB for scheduler documentation is far better than scheduler documentation.

f you have extra time, then write specification for power users. But most of the time they will send an email on the mailing list for advanced uses.

To help you to understand what a good documentation is, you should try to use an unknown application. Lets say hadoop or glusterf. Both have a really good documentation.

Some advises:
- Users are not fluent with the technologies you use: ""The ProActive scheduler supports portable scripts execution through the JSR 223 Java Scripting capabilities; scripts can be written in any language supported by the underlying Java Runtime Environment". You have to provide a pointer on languages supported by JSR223 or enumerate them.
- Explain why a feature is available with an use case: “pre-script The pre-script is always executed on the node that has been selected by the resource manager before the execution of the task itself.” Ok but I can’t see how pre-scripts can help me…
- Don’t assume that someone will read the documentation sequentially. Documentations are boring and users are lazy: “Create a java task is not much more difficult” So to create a java task I have to learn how to create a native task.

ResultDescripor becomes ResultPreview

Original issue created by Christian Delbé on 19, Nov 2007 at 11:43 AM – SCHEDULING-37

---

And some examples can be commited…

Scheduler client crash

Original issue created by Clement Mathieu on 17, May 2008 at 16:11 PM – SCHEDULING-1

---

After running some job, i tried to get the output of my job. Then the client crash.

Stacktrace and job descriptor in attachement

Job XM Descriptor should be typesafe with variable checks

Original issue created by Clement Mathieu on 17, May 2008 at 15:52 PM – SCHEDULING-17

---

Scheduler and GCM deployment mainly use the same code for variable handling. But they don’t share the code base.
Since the Scheduler code was forked, we did a lot of improvements. Our descriptor are now typesafe and variable definition is checked.

For example if an user define the WORK variable, but use ${WORk} somewhere else a clear error message is printed and the execution stops.
We should share the same library to resolve variables or, at least, you have to merge our changes into the scheduler.

The JavaDoc for UserSchedulerInterface.getResult(JobId jobId) doest not specify de behaviour if a job is not available

Original issue created by Mario Leyton on 24, Oct 2007 at 10:46 AM – SCHEDULING-29

---

The return value case is not specified if the result is not available (ie you have to say null is returned if the job is not yet available).

Exclusive access to a compute node

Original issue created by Clement Mathieu on 17, May 2008 at 16:38 PM – SCHEDULING-6

---

I have bunch of identical jobs to schedule. But my job use a runtime that can be started once and only once per machine. How to write a selection script for my use case ? With JSR223, I’m on top of a JVM so I cannot check for already existing processes.

How do you manage native parallel applications ? My application use as many processors as declared by the operating system and use a lot of memory. Can my job asks for a dedicated machine ?

[RFE] Multiple job selection in client GUI

Original issue created by Clement Mathieu on 17, May 2008 at 15:54 PM – SCHEDULING-14

---

It is not possible to select multiple jobs in the client scheduler. It is really painful to remove all our finished job.

Job removed from Scheduler after getting the job result. Add a timeout for a job.

Original issue created by Salageanu Emil on 27, May 2008 at 10:52 AM – SCHEDULING-15

---

I have an application that creates a job and submits it to the scheduler. I can see ob the scheduler interface the job running. When the job finishes my application needs to get the result in order to perform some post-treatment. the scheduler will removed the finished job from the database because the result have been retreived by the application. In my case, I would really want to see my job in the Finished – Jobs widget. Would it be possible to add a timeout variable for a job wich would define how long this job will remain in the scheduler after it was finished and its result retreived (i.e. timeout = 1 week) ? For not changing the current behaviour of the scheduler, the default timeout could be 0.

The interface ExecutableTask should have its return value typed with generics

Original issue created by Mario Leyton on 24, Oct 2007 at 10:52 AM – SCHEDULING-20

---

public interface ExecutableTask extends Serializable {
public Object execute(TaskResult… results) throws Throwable;
}

public interface ExecutableTask extends Serializable { public R execute(TaskResult… results) throws Throwable;

}

Move logforwarder into scheduler

Original issue created by Christian Delbé on 05, Dec 2007 at 17:20 PM – SCHEDULING-39

---

No description

Job Descriptor Schema not referenced

Original issue created by Fabien Viale on 23, Oct 2007 at 14:49 PM – SCHEDULING-28

---

The Job descriptor schema is not referenced by current job examples descriptors.

By extension, the descriptors are not validated against the schema when a job is submitted.

Finally, the schema seems non-consistent with the current descriptors format.

The Schema location should follow the same principle as ProActive :
1) descriptors reference an online version of the schema
2) JobFactory validates the given descriptor against an internal Schema (in src/Extra/o.o.p/scheduler/common/schemas for example).

Scheduler jobs can't have more than 1000 tasks !

Original issue created by Fabien Viale on 11, Jun 2008 at 17:07 PM – SCHEDULING-34

---

self explanating

Last submited job not known for a user

Original issue created by Salageanu Emil on 28, Apr 2008 at 15:57 PM – SCHEDULING-7

---

I submit a job to the scheduler without using the GUI (I connect to the scheduler and submit a job in my program). In the Scheduler GUI, in the Users tab, i don’t see the last submited job for my user.

Canceled state is misleading

Original issue created by Clement Mathieu on 17, May 2008 at 15:47 PM – SCHEDULING-44

---

When a job cannot be started due to invalid job descriptor (wrong path for instance), the job is put in the finished queue in ‘canceled’ state.

According to me, canceled means that the job have been canceled by the user (or administrator). My job is clearly in an error state since it has not been able to start.

Example:
(link: junit@00:37:44) java.io.IOException: Cannot run program “${WORK_DIR}/junitBatch.sh”: java.io.IOException: error=2, No such file or directory

JobLauncher.sh takes password as parameter

Original issue created by Clement Mathieu on 17, May 2008 at 15:39 PM – SCHEDULING-30

---

jobLauncher.sh takes username and password as parameter. passwords must NEVER be put on a command line since it is a world readable information.

password must be asked with echo turned off (google keywords: bash password stty)

Complete/Improve Java documentation for your part.

Original issue created by Jean-Luc Scheefer on 02, Jul 2008 at 15:18 PM – SCHEDULING-47

---

Think about completing or improving the java documentation in the scheduler.ext package. that includes :
- masterWorker
- matlab
- scilab

Eclipse can help you to find the errors/problems in the documentation.
See Windows>preference>java>compiler>javadoc >> check everything and enjoy ;).

Precious result incorrectly handled

Original issue created by Clement Mathieu on 17, May 2008 at 15:32 PM – SCHEDULING-32

---

A job with no precious result is rejected by the scheduler but it is possible to write a such document (with validation turned on) and documentation does not mention a such limitation.

“precious result is the way to define that a result of a task is important or not. For example, in a job result, you could have to retrieve only some task results that are important for you. By setting the precious result to true you’ll be able to retrieve easily these results. (default is false)”

Scheduler or (XSD and documentation) must be fixed

Resource Manager can't handle more than 500 nodes

Original issue created by Elton Mathias on 13, Jun 2008 at 13:38 PM – SCHEDULING-12

---

I think it’s a pretty critical issue the fact that the Resource manager can’t handle more than 500 nodes. At the same time I believe it’s not simple to change it for the release 4.0, because it’s something on the PA core and that may require some work … The error:

```

• Mapping VirtualNode Workers with Node: rmi://grid179.lal.in2p3.fr:1099/Workers318294075 done
  ```

(link: RMCORE) New node added, node ID is : rmi://grid95.lal.in2p3.fr:1099/Workers2105358747, node Source : Default
(link: Default) new node available : rmi://grid179.lal.in2p3.fr:1099/Workers318294075
*\\* Mapping VirtualNode Workers with Node: rmi://grid163.lal.in2p3.fr:1099/Workers980096541 done
(link: RMCORE) New node added, node ID is : rmi://grid179.lal.in2p3.fr:1099/Workers318294075, node Source : Default
Lookup of rmi://grid146.lal.in2p3.fr:1099/PA_JVM1529979507 succeed
(link: Default) new node available : rmi://grid163.lal.in2p3.fr:1099/Workers980096541
\* Mapping VirtualNode Workers with Node: rmi://grid78.lal.in2p3.fr:1099/Workers175105317 done
Lookup of rmi://grid254.lal.in2p3.fr:1099/PA_JVM283084253 succeed
\* Mapping VirtualNode Workers with Node: rmi://grid230.lal.in2p3.fr:1099/Workers579128335 done
Unable to create nodes on rmi://grid155.lal.in2p3.fr:1099/PA_JVM687477054
org.objectweb.proactive.core.ProActiveException: java.rmi.ConnectIOException: Exception creating connection to: 134.158.73.51; nested exception is:
java.net.SocketException: Too many open files
at org.objectweb.proactive.core.remoteobject.RemoteObjectAdapter.getURI(RemoteObjectAdapter.java:618)
at org.objectweb.proactive.core.remoteobject.RemoteObjectAdapter.(RemoteObjectAdapter.java:149)
at org.objectweb.proactive.core.remoteobject.rmi.RmiRemoteObjectFactory.lookup(RmiRemoteObjectFactory.java:211)
at org.objectweb.proactive.core.remoteobject.RemoteObjectHelper.lookup(RemoteObjectHelper.java:178)
at org.objectweb.proactive.core.runtime.RuntimeFactory.getRuntime(RuntimeFactory.java:195)
at org.objectweb.proactive.core.descriptor.data.VirtualNodeImpl.createNodes(VirtualNodeImpl.java:1035)
at org.objectweb.proactive.core.descriptor.data.VirtualNodeImpl.handleNotification(VirtualNodeImpl.java:1007)
at com.sun.jmx.interceptor.DefaultMBeanServerInterceptor$ListenerWrapper.handleNotification(DefaultMBeanServerInterceptor.java:1732)
at javax.management.NotificationBroadcasterSupport.handleNotification(NotificationBroadcasterSupport.java:257)
at javax.management.NotificationBroadcasterSupport$SendNotifJob.run(NotificationBroadcasterSupport.java:322)
at javax.management.NotificationBroadcasterSupport$1.execute(NotificationBroadcasterSupport.java:307)
at javax.management.NotificationBroadcasterSupport.sendNotification(NotificationBroadcasterSupport.java:229)
at org.objectweb.proactive.core.jmx.mbean.ProActiveRuntimeWrapper.sendNotification(ProActiveRuntimeWrapper.java:133)
at org.objectweb.proactive.core.runtime.ProActiveRuntimeImpl.register(ProActiveRuntimeImpl.java:605)
at sun.reflect.GeneratedMethodAccessor26.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
at java.lang.reflect.Method.invoke(Method.java:597)
at org.objectweb.proactive.core.mop.MethodCall.execute(MethodCall.java:384)
at org.objectweb.proactive.core.remoteobject.RemoteObjectImpl.receiveMessage(RemoteObjectImpl.java:99)
at org.objectweb.proactive.core.remoteobject.InternalRemoteRemoteObjectImpl.receiveMessage(InternalRemoteRemoteObjectImpl.java:144)
at org.objectweb.proactive.core.remoteobject.rmi.RmiRemoteObjectImpl.receiveMessage(RmiRemoteObjectImpl.java:70)
at sun.reflect.GeneratedMethodAccessor6.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
at java.lang.reflect.Method.invoke(Method.java:597)
at sun.rmi.server.UnicastServerRef.dispatch(UnicastServerRef.java:305)
at sun.rmi.transport.Transport$1.run(Transport.java:159)
at java.security.AccessController.doPrivileged(Native Method)
at sun.rmi.transport.Transport.serviceCall(Transport.java:155)
at sun.rmi.transport.tcp.TCPTransport.handleMessages(TCPTransport.java:535)
at sun.rmi.transport.tcp.TCPTransport$ConnectionHandler.run0(TCPTransport.java:790)
at sun.rmi.transport.tcp.TCPTransport$ConnectionHandler.run(TCPTransport.java:649)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:885)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:907)
at java.lang.Thread.run(Thread.java:619)
Caused by: java.rmi.ConnectIOException: Exception creating connection to: 134.158.73.51; nested exception is:
java.net.SocketException: Too many open files
at sun.rmi.transport.tcp.TCPEndpoint.newSocket(TCPEndpoint.java:614)
at sun.rmi.transport.tcp.TCPChannel.createConnection(TCPChannel.java:198)
at sun.rmi.transport.tcp.TCPChannel.newConnection(TCPChannel.java:184)
at sun.rmi.server.UnicastRef.invoke(UnicastRef.java:110)
at org.objectweb.proactive.core.remoteobject.rmi.RmiRemoteObjectImpl_Stub.receiveMessage(Unknown Source)
at org.objectweb.proactive.core.remoteobject.RemoteObjectAdapter.getURI(RemoteObjectAdapter.java:614)
… 33 more
Caused by: java.net.SocketException: Too many open files
at java.net.Socket.createImpl(Socket.java:387)
at java.net.Socket.(Socket.java:361)
at java.net.Socket.(Socket.java:179)
at sun.rmi.transport.proxy.RMIDirectSocketFactory.createSocket(RMIDirectSocketFactory.java:22)
at sun.rmi.transport.proxy.RMIMasterSocketFactory.createSocket(RMIMasterSocketFactory.java:128)
at sun.rmi.transport.tcp.TCPEndpoint.newSocket(TCPEndpoint.java:595)
… 38 more

Documentation: Scheduler documentation should be standalone

Original issue created by Clement Mathieu on 17, May 2008 at 16:44 PM – SCHEDULING-19

---

Scheduler should have its own manual. The PPS scheduler is not ProActive.

Scheduler and ResourceManager should migrate from Extra to Extension

Original issue created by Christian Delbé on 05, Dec 2007 at 17:02 PM – SCHEDULING-38

---

No description

Themethod TaskResult .getTaskId() is pointless, since the TaskId is only used internally in the scheduler

Original issue created by Mario Leyton on 24, Oct 2007 at 10:27 AM – SCHEDULING-2

---

What is the point of being able to do TaskResult.getTaskId(), if the taskId is only used internally in the framework, and the user never knows it?

Perhaps it would be nice to separete the public interface of a TaskResutl from the implementation TaskResultImpl

Real time job output

Original issue created by Clement Mathieu on 17, May 2008 at 16:10 PM – SCHEDULING-21

---

It is not possible to get the output of a job in real time (heavy buffering ?). I had to wait the end of my job to get the output. When you run a long job for the first time, real time output is mandatory. You don’t want to wait 2h to see that something gone wrong after 10 secondes

Automatic scatter of results mechanism

Original issue created by Fabien Viale on 22, Oct 2007 at 19:07 PM – SCHEDULING-8

---

It would be nice to have a mechanism that can scatter an array of results of one task into each child input.

The scatter/split task will be given the number of children as an extra input, and will return an array of objects.

The scatter mechanism will give to each child its corresponding input.

The size of the array will be checked at runtime to see if it matchs the number of children, otherwise an Exception will be thrown to each individual child.

The EvenListener mechanism is hardcoded around method names

Original issue created by Mario Leyton on 25, Oct 2007 at 18:43 PM – SCHEDULING-13

---

The EvenListener mechanism is hardcoded around method names. To be able to listen on an event, the programmer must:
1. Decide which events to listen to
2. Implement the corresponding methods
3. Check the mapping between method name → event enumeration
4. Add event listener instance for each enumeration.

The problem with this approach is that it is error prone, manually converting from method names to enumerations. And, that if the method names change, then java compiler will not detect this errors.

I suggest implementing one interface per event or group of events, and then annotating the methods for which you are really interested. Thus eliminating step 3, 4.
The drawback would be that a class is hardcoded (ie will always listen for the annotated methods).

Implementing a mechanism to cache Matlab/Scilab engine access inside Scheduler Nodes

Original issue created by Fabien Viale on 13, Dec 2007 at 13:18 PM – SCHEDULING-45

---

When launching Matlab or Scilab jobs in the scheduler, the current implementation starts a new JVM and a new Matlab engine for each new Matlab or Scilab task.

It would be nice to reduce this huge overhead by keeping a connection to the Scilab/Matlab engine between tasks execution.

In this approach, a leasing system or something equivalent is also needed to avoid using resources for ever.