Search in sources :

Example 81 with NodeSource

use of org.ow2.proactive.resourcemanager.nodesource.NodeSource in project scheduling by ow2-proactive.

the class NodeSource method initActivity.

/**
 * Initialization of node source. Creates and activates a pinger to monitor nodes.
 *
 * @param body active object body
 */
public void initActivity(Body body) {
    this.stub = (NodeSource) PAActiveObject.getStubOnThis();
    this.infrastructureManager.setNodeSource(this);
    // Infrastructure has been configured and linked to the node source, so we can now persist the runtime
    // variables of the infrastructure for the first time (they have been initialized during the creation of the
    // infrastructure, in its configuration.
    this.infrastructureManager.persistInfrastructureVariables();
    this.activePolicy.setNodeSource((NodeSource) PAActiveObject.getStubOnThis());
    // Set permissions again according to the activated node source policy
    // node source admin permission
    // it's the PrincipalPermission of the user who created the node source
    this.adminPermission = new PrincipalPermission(this.administrator.getName(), this.administrator.getSubject().getPrincipals(UserNamePrincipal.class));
    // creating node source provider permission
    // could be one of the following: PrincipalPermission (NS creator) or PrincipalPermission (NS creator groups)
    // or PrincipalPermission (anyone)
    this.providerPermission = new PrincipalPermission(this.administrator.getName(), this.activePolicy.getProviderAccessType().getIdentityPrincipals(this.administrator));
    this.nodeUserAccessType = this.activePolicy.getUserAccessType();
    Thread.currentThread().setName("Node Source \"" + this.name + "\"");
}
Also used : PrincipalPermission(org.ow2.proactive.permissions.PrincipalPermission)

Example 82 with NodeSource

use of org.ow2.proactive.resourcemanager.nodesource.NodeSource in project scheduling by ow2-proactive.

the class NodeSource method detectedPingedDownNode.

/**
 * Marks node as down. Remove it from node source node set. It remains in rmcore nodes list until
 * user decides to remove them or node source is shutdown.
 * @see NodeSource#detectedPingedDownNode(String, String)
 */
public void detectedPingedDownNode(String nodeName, String nodeUrl) {
    if (toShutdown) {
        logger.warn("[" + name + "] detectedPingedDownNode request discarded because node source is shutting down");
        return;
    }
    Node downNode = nodes.remove(nodeUrl);
    if (downNode != null) {
        logger.warn("[" + name + "] Detected down node: " + nodeUrl);
        downNodes.put(nodeUrl, downNode);
        try {
            RMCore.topologyManager.removeNode(downNode);
            infrastructureManager.internalNotifyDownNode(nodeName, nodeUrl, downNode);
        } catch (RMException e) {
            logger.error("Error while removing down node: " + nodeUrl, e);
        }
    } else if (removedNodes.containsKey(nodeUrl)) {
        logger.info("[" + name + "] Detected down node has already been removed: " + nodeUrl);
    // node has been removed, ping detection is ignored
    } else {
        logger.warn("[" + name + "] Detected down node: " + nodeUrl);
        // almost no information about the node apart from its name and url
        try {
            infrastructureManager.internalNotifyDownNode(nodeName, nodeUrl, null);
        } catch (RMException e) {
            logger.error("New empty node " + nodeUrl + " could not be created to handle down node", e);
        }
    }
    rmcore.setDownNode(nodeUrl);
}
Also used : RMNode(org.ow2.proactive.resourcemanager.rmnode.RMNode) RMDeployingNode(org.ow2.proactive.resourcemanager.rmnode.RMDeployingNode) Node(org.objectweb.proactive.core.node.Node) AbstractRMNode(org.ow2.proactive.resourcemanager.rmnode.AbstractRMNode) RMException(org.ow2.proactive.resourcemanager.exception.RMException)

Example 83 with NodeSource

use of org.ow2.proactive.resourcemanager.nodesource.NodeSource in project scheduling by ow2-proactive.

the class AutoUpdateInfrastructure method startNodeImpl.

/**
 * Internal node acquisition method
 * <p>
 * Starts a PA runtime on remote host using a custom script, register it
 * manually in the nodesource.
 *
 * @param hostTracker The host on which one the node will be started
 * @param nbNodes number of nodes to deploy
 * @param depNodeURLs list of deploying or lost nodes urls created
 * @throws org.ow2.proactive.resourcemanager.exception.RMException
 *             acquisition failed
 */
protected void startNodeImpl(HostTracker hostTracker, int nbNodes, final List<String> depNodeURLs) throws RMException {
    final String nodeName = this.nodeSource.getName() + "-" + ProActiveCounter.getUniqID();
    String credentials = "";
    try {
        credentials = new String(nodeSource.getAdministrator().getCredentials().getBase64());
    } catch (KeyException e) {
        logger.error("Invalid credentials");
        return;
    }
    Properties localProperties = new Properties();
    localProperties.put(NODE_NAME, nodeName);
    localProperties.put(HOST_NAME, hostTracker.getResolvedAddress().getHostName());
    localProperties.put(NODESOURCE_CREDENTIALS, credentials);
    localProperties.put(NODESOURCE_NAME, nodeSource.getName());
    localProperties.put(NB_NODES, nbNodes);
    String filledCommand = replaceProperties(command, localProperties);
    filledCommand = replaceProperties(filledCommand, System.getProperties());
    final List<String> createdNodeNames = RMNodeStarter.getWorkersNodeNames(nodeName, nbNodes);
    depNodeURLs.addAll(addMultipleDeployingNodes(createdNodeNames, filledCommand, "Deploying node on host " + hostTracker.getResolvedAddress(), this.nodeTimeOut));
    addTimeouts(depNodeURLs);
    Process p;
    try {
        logger.debug("Deploying node: " + nodeName);
        logger.debug("Launching the command: " + filledCommand);
        p = Runtime.getRuntime().exec(new String[] { "bash", "-c", filledCommand });
    } catch (IOException e1) {
        multipleDeclareDeployingNodeLost(depNodeURLs, "Cannot run command: " + filledCommand + " - \n The following exception occurred: " + getStackTraceAsString(e1));
        throw new RMException("Cannot run command: " + filledCommand, e1);
    }
    String lf = System.lineSeparator();
    int circuitBreakerThreshold = 5;
    while (!anyTimedOut(depNodeURLs) && circuitBreakerThreshold > 0) {
        try {
            int exitCode = p.exitValue();
            if (exitCode != 0) {
                logger.error("Child process at " + hostTracker.getResolvedAddress().getHostName() + " exited abnormally (" + exitCode + ").");
            } else {
                logger.error("Launching node script has exited normally whereas it shouldn't.");
            }
            String pOutPut = Utils.extractProcessOutput(p);
            String pErrPut = Utils.extractProcessErrput(p);
            final String description = "Script failed to launch a node on host " + hostTracker.getResolvedAddress().getHostName() + lf + "   >Error code: " + exitCode + lf + "   >Errput: " + pErrPut + "   >Output: " + pOutPut;
            logger.error(description);
            if (super.checkNodeIsAcquiredAndDo(nodeName, null, new Runnable() {

                public void run() {
                    multipleDeclareDeployingNodeLost(depNodeURLs, description);
                }
            })) {
                return;
            } else {
                // there isn't any race regarding node registration
                throw new RMException("A node " + nodeName + " is not expected anymore because of an error.");
            }
        } catch (IllegalThreadStateException e) {
            logger.trace("IllegalThreadStateException while waiting for " + nodeName + " registration");
        }
        if (super.checkNodeIsAcquiredAndDo(nodeName, null, null)) {
            // registration is ok, we destroy the process
            logger.debug("Destroying the process: " + p);
            try {
                ProcessTree.get().get(p).kill();
            } catch (InterruptedException e) {
                Thread.currentThread().interrupt();
            }
            return;
        }
        try {
            Thread.sleep(1000);
        } catch (Exception e) {
            circuitBreakerThreshold--;
            logger.trace("An exception occurred while monitoring a child process", e);
        }
    }
    // if we exit because of a timeout
    if (anyTimedOut(depNodeURLs)) {
        // we remove it
        removeTimeouts(depNodeURLs);
        // we destroy the process
        p.destroy();
        throw new RMException("Deploying Node " + nodeName + " not expected any more");
    }
    if (circuitBreakerThreshold <= 0) {
        logger.error("Circuit breaker threshold reached while monitoring a child process.");
        throw new RMException("Several exceptions occurred while monitoring a child process.");
    }
}
Also used : Throwables.getStackTraceAsString(com.google.common.base.Throwables.getStackTraceAsString) IOException(java.io.IOException) Properties(java.util.Properties) KeyException(java.security.KeyException) RMException(org.ow2.proactive.resourcemanager.exception.RMException) KeyException(java.security.KeyException) IOException(java.io.IOException) RMException(org.ow2.proactive.resourcemanager.exception.RMException)

Example 84 with NodeSource

use of org.ow2.proactive.resourcemanager.nodesource.NodeSource in project scheduling by ow2-proactive.

the class InfrastructureManager method emitNodeEvent.

// **********************************************************************************************//
// *********************** Package private accessors & Helpers
// **********************************//
// **********************************************************************************************//
/**
 * To emit a node event and register it in the database
 */
private void emitNodeEvent(final RMNodeEvent event) {
    NodeSource nsStub = this.nodeSource.getStub();
    nsStub.internalEmitDeployingNodeEvent(event);
}
Also used : NodeSource(org.ow2.proactive.resourcemanager.nodesource.NodeSource)

Aggregations

NodeSource (org.ow2.proactive.resourcemanager.nodesource.NodeSource)32 RMNode (org.ow2.proactive.resourcemanager.rmnode.RMNode)19 Test (org.junit.Test)17 BooleanWrapper (org.objectweb.proactive.core.util.wrapper.BooleanWrapper)17 Node (org.objectweb.proactive.core.node.Node)12 RMDeployingNode (org.ow2.proactive.resourcemanager.rmnode.RMDeployingNode)12 RMException (org.ow2.proactive.resourcemanager.exception.RMException)11 Client (org.ow2.proactive.resourcemanager.authentication.Client)10 RMNodeSourceEvent (org.ow2.proactive.resourcemanager.common.event.RMNodeSourceEvent)9 AddingNodesException (org.ow2.proactive.resourcemanager.exception.AddingNodesException)9 URISyntaxException (java.net.URISyntaxException)7 ArrayList (java.util.ArrayList)7 NSAdminPermission (org.ow2.proactive.permissions.NSAdminPermission)7 RMCoreAllPermission (org.ow2.proactive.permissions.RMCoreAllPermission)7 NodeSourceData (org.ow2.proactive.resourcemanager.db.NodeSourceData)7 ScriptException (org.ow2.proactive.scripting.ScriptException)7 HashSet (java.util.HashSet)6 ActiveObjectCreationException (org.objectweb.proactive.ActiveObjectCreationException)6 NodeException (org.objectweb.proactive.core.node.NodeException)6 IOException (java.io.IOException)5