/**
* Notify RM to create a new node.
*
* @param parentID The ID of the machine the node belongs to
* @param name the name of the node
* @param number the number of the node (rank)
* @return the id of the new node
*/
public String createNode(String parentID, String name, int number) {
String id = generateID();
ElementAttributeManager mgr = new ElementAttributeManager();
AttributeManager attrMgr = new AttributeManager();
attrMgr.addAttribute(
NodeAttributes.getStateAttributeDefinition().create(NodeAttributes.State.UP));
try {
attrMgr.addAttribute(
NodeAttributes.getNumberAttributeDefinition().create(new Integer(number)));
} catch (IllegalValueException e) {
/*
* This exception is not possible, since number is always valid.
*/
RMCorePlugin.log(e);
assert false;
}
attrMgr.addAttribute(ElementAttributes.getNameAttributeDefinition().create(name));
mgr.setAttributeManager(new RangeSet(id), attrMgr);
fireRuntimeNewNodeEvent(eventFactory.newRuntimeNewNodeEvent(parentID, mgr));
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: new node #{1}",
getResourceManager().getConfiguration().getName(),
id); //$NON-NLS-1$
return id;
}
/**
* Create a single process.
*
* @param jobId the parent job that the process belongs to
* @param the index (job rank) of the new process
* @since 2.0
*/
public void createProcess(String jobId, int index) {
ElementAttributeManager mgr = new ElementAttributeManager();
AttributeManager attrMgr = new AttributeManager();
attrMgr.addAttribute(
ProcessAttributes.getStateAttributeDefinition().create(ProcessAttributes.State.STARTING));
mgr.setAttributeManager(new RangeSet(index), attrMgr);
fireRuntimeNewProcessEvent(eventFactory.newRuntimeNewProcessEvent(jobId, mgr));
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: new process #{1}",
getResourceManager().getConfiguration().getName(),
Integer.toString(index)); // $NON-NLS-1$
}
/**
* Notify RM to create a new queue.
*
* @param name the name of the queue
* @return the id of the new queue
*/
public String createQueue(String name) {
String id = generateID();
ElementAttributeManager mgr = new ElementAttributeManager();
AttributeManager attrMgr = new AttributeManager();
attrMgr.addAttribute(ElementAttributes.getNameAttributeDefinition().create(name));
mgr.setAttributeManager(new RangeSet(id), attrMgr);
IPResourceManager rm =
(IPResourceManager) getResourceManager().getAdapter(IPResourceManager.class);
fireRuntimeNewQueueEvent(eventFactory.newRuntimeNewQueueEvent(rm.getID(), mgr));
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: new queue #{1}",
getResourceManager().getConfiguration().getName(),
id); //$NON-NLS-1$
return id;
}
/**
* Change attributes of a node
*
* @param nodeID
* @param changedAttrMgr
*/
public void changeNode(String nodeID, AttributeManager changedAttrMgr) {
AttributeManager attrMgr = new AttributeManager();
attrMgr.addAttributes(changedAttrMgr.getAttributes());
ElementAttributeManager elementAttrs = new ElementAttributeManager();
elementAttrs.setAttributeManager(new RangeSet(nodeID), attrMgr);
IRuntimeNodeChangeEvent event = eventFactory.newRuntimeNodeChangeEvent(elementAttrs);
fireRuntimeNodeChangeEvent(event);
for (IAttribute<?, ?, ?> attr : changedAttrMgr.getAttributes()) {
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}, node #{1}: {2}={3}",
getResourceManager().getConfiguration().getName(),
nodeID,
attr.getDefinition().getId(),
attr.getValueAsString()); // $NON-NLS-1$
}
}
/**
* Change attributes of a process
*
* @param jobId id of processes' parent job
* @param processJobRanks
* @param changedAttrMgr
* @since 2.0
*/
public void changeProcesses(
String jobId, BitSet processJobRanks, AttributeManager changedAttrMgr) {
AttributeManager attrMgr = new AttributeManager();
attrMgr.addAttributes(changedAttrMgr.getAttributes());
ElementAttributeManager elementAttrs = new ElementAttributeManager();
elementAttrs.setAttributeManager(new RangeSet(processJobRanks), attrMgr);
IRuntimeProcessChangeEvent event =
eventFactory.newRuntimeProcessChangeEvent(jobId, elementAttrs);
fireRuntimeProcessChangeEvent(event);
for (IAttribute<?, ?, ?> attr : changedAttrMgr.getAttributes()) {
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}, processes #{1}: {2}={3}",
getResourceManager().getConfiguration().getName(),
processJobRanks,
attr.getDefinition().getId(),
attr.getValueAsString()); // $NON-NLS-1$
}
}
/*
* (non-Javadoc)
*
* @see
* org.eclipse.ptp.rtsystem.IRuntimeSystem#startup(org.eclipse.core.runtime
* .IProgressMonitor)
*/
public void startup(IProgressMonitor monitor) throws CoreException {
SubMonitor subMon = SubMonitor.convert(monitor, 100);
synchronized (this) {
startupMonitor = subMon;
}
initialize();
subMon.subTask(Messages.AbstractToolRuntimeSystem_1);
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: startup",
getResourceManager().getConfiguration().getName()); // $NON-NLS-1$
try {
remoteServices =
PTPRemoteCorePlugin.getDefault()
.getRemoteServices(
getResourceManager().getControlConfiguration().getRemoteServicesId(),
subMon.newChild(10));
if (remoteServices == null) {
throw new CoreException(
new Status(
IStatus.ERROR,
RMCorePlugin.PLUGIN_ID,
Messages.AbstractToolRuntimeSystem_Exception_NoRemoteServices));
}
IRemoteConnectionManager connectionManager = remoteServices.getConnectionManager();
Assert.isNotNull(connectionManager);
subMon.worked(10);
subMon.subTask(Messages.AbstractToolRuntimeSystem_2);
connection =
connectionManager.getConnection(
getResourceManager().getControlConfiguration().getConnectionName());
if (connection == null) {
throw new CoreException(
new Status(
IStatus.ERROR,
RMCorePlugin.PLUGIN_ID,
Messages.AbstractToolRuntimeSystem_Exception_NoConnection));
}
if (!connection.isOpen()) {
try {
connection.open(subMon.newChild(50));
} catch (RemoteConnectionException e) {
throw new CoreException(
new Status(IStatus.ERROR, RMCorePlugin.PLUGIN_ID, e.getMessage()));
}
}
if (subMon.isCanceled()) {
connection.close();
return;
}
try {
doStartup(subMon.newChild(40));
} catch (CoreException e) {
connection.close();
throw e;
}
if (subMon.isCanceled()) {
connection.close();
return;
}
/*
* Wait for discover job to complete so we can check it's status and
* throw an exception if it fails.
*/
Job discoverJob = createDiscoverJob();
if (discoverJob != null) {
/*
* Disable error messages being displayed on job's progress
* monitor. These are handled by the progress service starting
* the RM.
*/
discoverJob.setProperty(
IProgressConstants.NO_IMMEDIATE_ERROR_PROMPT_PROPERTY, Boolean.TRUE);
discoverJob.schedule();
try {
discoverJob.join();
} catch (InterruptedException e) {
// Just check result
}
IStatus status = discoverJob.getResult();
if (!status.isOK()) {
throw new CoreException(status);
}
}
if (jobQueueThread == null) {
jobQueueThread =
new Thread(
new JobRunner(), Messages.AbstractToolRuntimeSystem_JobQueueManagerThreadTitle);
jobQueueThread.start();
}
/*
* Register for connection events
*/
connection.addConnectionChangeListener(fConnectionChangeHandler);
fireRuntimeRunningStateEvent(eventFactory.newRuntimeRunningStateEvent());
} finally {
synchronized (this) {
startupMonitor = null;
}
if (monitor != null) {
monitor.done();
}
}
}
/*
* (non-Javadoc)
*
* @see org.eclipse.ptp.rtsystem.IRuntimeSystem#shutdown()
*/
public void shutdown() throws CoreException {
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: shutdown",
getResourceManager().getConfiguration().getName()); // $NON-NLS-1$
/*
* Remove listener to avoid re-entry
*/
if (connection != null) {
connection.removeConnectionChangeListener(fConnectionChangeHandler);
}
try {
stopEvents();
} catch (CoreException e) {
// Ignore exception and shutdown anyway
RMCorePlugin.log(e);
}
try {
doShutdown();
} finally {
/*
* Stop jobs that might be in the pending queue. Also stop the
* thread that dispatches pending jobs.
*/
if (jobQueueThread != null) {
jobQueueThread.interrupt();
for (Job job : pendingJobQueue) {
job.cancel();
}
}
/*
* Stop jobs that are running or that already finished.
*/
Iterator<Job> iterator = jobs.values().iterator();
while (iterator.hasNext()) {
Job job = iterator.next();
job.cancel();
iterator.remove();
}
synchronized (this) {
if (startupMonitor != null) {
startupMonitor.setCanceled(true);
}
}
/*
* Close the the connection.
*/
if (connection != null) {
connection.close();
}
jobQueueThread = null;
fireRuntimeShutdownStateEvent(eventFactory.newRuntimeShutdownStateEvent());
}
}
/**
* Notify RM to create a new job.
*
* @param parentID parent element ID
* @param jobID the ID of the job
* @param attrMgr attributes from the job thread
* @return job id of the newly created job
*/
public String createJob(String parentID, AttributeManager attrMgr) throws CoreException {
ElementAttributeManager mgr = new ElementAttributeManager();
AttributeManager jobAttrMgr = new AttributeManager();
/*
* Add generated attributes.
*/
String jobID = generateID().toString();
jobAttrMgr.addAttribute(JobAttributes.getJobIdAttributeDefinition().create(jobID));
jobAttrMgr.addAttribute(
JobAttributes.getStatusAttributeDefinition()
.create(MPIJobAttributes.Status.NORMAL.toString()));
jobAttrMgr.addAttribute(
JobAttributes.getUserIdAttributeDefinition().create(System.getenv("USER"))); // $NON-NLS-1$
jobAttrMgr.addAttribute(
ElementAttributes.getNameAttributeDefinition().create(generateJobName()));
/*
* Get mandatory launch attributes.
*/
String subId = getAttributeValue(JobAttributes.getSubIdAttributeDefinition(), attrMgr);
String execName =
getAttributeValue(JobAttributes.getExecutableNameAttributeDefinition(), attrMgr);
String execPath =
getAttributeValue(JobAttributes.getExecutablePathAttributeDefinition(), attrMgr);
String workDir =
getAttributeValue(JobAttributes.getWorkingDirectoryAttributeDefinition(), attrMgr);
Integer numProcs =
getAttributeValue(JobAttributes.getNumberOfProcessesAttributeDefinition(), attrMgr);
List<? extends String> progArgs =
getAttributeValue(JobAttributes.getProgramArgumentsAttributeDefinition(), attrMgr);
/*
* Copy these relevant attributes to IPJob.
*/
jobAttrMgr.addAttribute(JobAttributes.getSubIdAttributeDefinition().create(subId));
jobAttrMgr.addAttribute(JobAttributes.getExecutableNameAttributeDefinition().create(execName));
jobAttrMgr.addAttribute(JobAttributes.getExecutablePathAttributeDefinition().create(execPath));
jobAttrMgr.addAttribute(JobAttributes.getWorkingDirectoryAttributeDefinition().create(workDir));
try {
jobAttrMgr.addAttribute(
JobAttributes.getNumberOfProcessesAttributeDefinition().create(numProcs));
} catch (IllegalValueException e) {
RMCorePlugin.log(e);
}
jobAttrMgr.addAttribute(
JobAttributes.getProgramArgumentsAttributeDefinition()
.create(progArgs.toArray(new String[0])));
/*
* Copy optional attributes
*/
BooleanAttribute debugAttr =
attrMgr.getAttribute(JobAttributes.getDebugFlagAttributeDefinition());
if (debugAttr != null) {
jobAttrMgr.addAttribute(
JobAttributes.getDebugFlagAttributeDefinition().create(debugAttr.getValue()));
}
/*
* Notify RM.
*/
mgr.setAttributeManager(new RangeSet(jobID), jobAttrMgr);
fireRuntimeNewJobEvent(eventFactory.newRuntimeNewJobEvent(parentID, mgr));
DebugUtil.trace(
DebugUtil.RTS_TRACING,
"RTS {0}: new job #{1}",
getResourceManager().getConfiguration().getName(),
jobID); //$NON-NLS-1$
return jobID;
}
