Examples with ClusterState org.apache.asterix.common.api.IClusterManagementWork.ClusterState used on opensource projects

Example 1 with ClusterState

use of org.apache.asterix.common.api.IClusterManagementWork.ClusterState in project asterixdb by apache.

the class GlobalRecoveryManager method startGlobalRecovery.

@Override
public void startGlobalRecovery(ICcApplicationContext appCtx) {
    // perform global recovery if state changed to active
    final ClusterState newState = ClusterStateManager.INSTANCE.getState();
    boolean needToRecover = !newState.equals(state) && (newState == ClusterState.ACTIVE);
    if (needToRecover) {
        setState(newState);
        ccServiceCtx.getControllerService().getExecutor().submit(() -> {
            LOGGER.info("Starting Global Recovery");
            MetadataTransactionContext mdTxnCtx = null;
            try {
                MetadataManager.INSTANCE.init();
                // Loop over datasets
                mdTxnCtx = MetadataManager.INSTANCE.beginTransaction();
                for (Dataverse dataverse : MetadataManager.INSTANCE.getDataverses(mdTxnCtx)) {
                    mdTxnCtx = recoverDataset(appCtx, mdTxnCtx, dataverse);
                }
                MetadataManager.INSTANCE.commitTransaction(mdTxnCtx);
            } catch (Exception e) {
                // This needs to be fixed <-- Needs to shutdown the system -->
                /*
                     * Note: Throwing this illegal state exception will terminate this thread
                     * and feeds listeners will not be notified.
                     */
                LOGGER.log(Level.SEVERE, "Global recovery was not completed successfully: ", e);
                if (mdTxnCtx != null) {
                    try {
                        MetadataManager.INSTANCE.abortTransaction(mdTxnCtx);
                    } catch (Exception e1) {
                        LOGGER.log(Level.SEVERE, "Exception in aborting", e1);
                        e1.addSuppressed(e);
                        throw new IllegalStateException(e1);
                    }
                }
            }
            ClusterStateManager.INSTANCE.setGlobalRecoveryCompleted(true);
            LOGGER.info("Global Recovery Completed");
        });
    }
}

Example 2 with ClusterState

use of org.apache.asterix.common.api.IClusterManagementWork.ClusterState in project asterixdb by apache.

the class StartConfig method execCommand.

@Override
protected void execCommand() throws Exception {
    InstallerDriver.initConfig(true);
    String asterixInstanceName = ((StartConfig) config).name;
    AsterixInstance instance = AsterixEventServiceUtil.validateAsterixInstanceExists(asterixInstanceName, State.INACTIVE);
    AsterixEventServiceUtil.createAsterixZip(instance);
    AsterixEventServiceClient client = AsterixEventService.getAsterixEventServiceClient(instance.getCluster());
    Patterns asterixBinaryTransferPattern = PatternCreator.INSTANCE.getAsterixBinaryTransferPattern(asterixInstanceName, instance.getCluster());
    client.submit(asterixBinaryTransferPattern);
    // Start the watcher
    ClusterStateWatcher stateWatcher = ServiceProvider.INSTANCE.getLookupService().startWatchingClusterState(asterixInstanceName);
    AsterixEventServiceUtil.createClusterProperties(instance.getCluster(), instance.getAsterixConfiguration());
    Patterns patterns = PatternCreator.INSTANCE.getStartAsterixPattern(asterixInstanceName, instance.getCluster(), false);
    client.submit(patterns);
    // Check the cluster state
    ClusterState clusterState = stateWatcher.waitForClusterStart();
    if (clusterState != ClusterState.ACTIVE) {
        throw new Exception("CC failed to start");
    }
    AsterixEventServiceUtil.deleteDirectory(InstallerDriver.getManagixHome() + File.separator + InstallerDriver.ASTERIX_DIR + File.separator + asterixInstanceName);
    AsterixRuntimeState runtimeState = VerificationUtil.getAsterixRuntimeState(instance);
    VerificationUtil.updateInstanceWithRuntimeDescription(instance, runtimeState, true);
    LOGGER.info(instance.getDescription(false));
    ServiceProvider.INSTANCE.getLookupService().updateAsterixInstance(instance);
}

Example 3 with ClusterState

use of org.apache.asterix.common.api.IClusterManagementWork.ClusterState in project asterixdb by apache.

the class CreateConfig method execCommand.

@Override
protected void execCommand() throws Exception {
    InstallerDriver.initConfig(true);
    ValidateCommand validateCommand = new ValidateCommand();
    boolean valid = validateCommand.validateCluster(((CreateConfig) config).clusterPath);
    if (!valid) {
        throw new Exception("Cannot create an Asterix instance.");
    }
    asterixInstanceName = ((CreateConfig) config).name;
    AsterixEventServiceUtil.validateAsterixInstanceNotExists(asterixInstanceName);
    CreateConfig createConfig = (CreateConfig) config;
    cluster = EventUtil.getCluster(createConfig.clusterPath);
    cluster.setInstanceName(asterixInstanceName);
    asterixConfiguration = InstallerUtil.getAsterixConfiguration(createConfig.asterixConfPath);
    AsterixInstance asterixInstance = AsterixEventServiceUtil.createAsterixInstance(asterixInstanceName, cluster, asterixConfiguration);
    AsterixEventServiceUtil.evaluateConflictWithOtherInstances(asterixInstance);
    AsterixEventServiceUtil.createAsterixZip(asterixInstance);
    AsterixEventServiceUtil.createClusterProperties(cluster, asterixConfiguration);
    AsterixEventServiceClient eventrixClient = AsterixEventService.getAsterixEventServiceClient(cluster, true, false);
    // Store the cluster initially in Zookeeper and start watching
    ServiceProvider.INSTANCE.getLookupService().writeAsterixInstance(asterixInstance);
    ClusterStateWatcher stateWatcher = ServiceProvider.INSTANCE.getLookupService().startWatchingClusterState(asterixInstanceName);
    Patterns asterixBinarytrasnferPattern = PatternCreator.INSTANCE.getAsterixBinaryTransferPattern(asterixInstanceName, cluster);
    eventrixClient.submit(asterixBinarytrasnferPattern);
    Patterns patterns = PatternCreator.INSTANCE.getStartAsterixPattern(asterixInstanceName, cluster, true);
    eventrixClient.submit(patterns);
    // Check the cluster state
    ClusterState clusterState = stateWatcher.waitForClusterStart();
    if (clusterState != ClusterState.ACTIVE) {
        throw new Exception("CC failed to start");
    }
    AsterixRuntimeState runtimeState = VerificationUtil.getAsterixRuntimeState(asterixInstance);
    VerificationUtil.updateInstanceWithRuntimeDescription(asterixInstance, runtimeState, true);
    ServiceProvider.INSTANCE.getLookupService().updateAsterixInstance(asterixInstance);
    AsterixEventServiceUtil.deleteDirectory(InstallerDriver.getManagixHome() + File.separator + InstallerDriver.ASTERIX_DIR + File.separator + asterixInstanceName);
    LOGGER.info(asterixInstance.getDescription(false));
}

Example 4 with ClusterState

use of org.apache.asterix.common.api.IClusterManagementWork.ClusterState in project asterixdb by apache.

the class AutoFaultToleranceStrategy method validateClusterState.

private void validateClusterState() throws HyracksDataException {
    clusterManager.refreshState();
    ClusterState newState = clusterManager.getState();
    // PENDING: all partitions are active but metadata node is not
    if (newState == ClusterState.PENDING) {
        requestMetadataNodeTakeover();
    } else if (newState == ClusterState.ACTIVE) {
        processPendingFailbackPlans();
    }
}

Example 5 with ClusterState

use of org.apache.asterix.common.api.IClusterManagementWork.ClusterState in project asterixdb by apache.

the class AutoFaultToleranceStrategy method processPendingFailbackPlans.

private synchronized void processPendingFailbackPlans() {
    ClusterState state = clusterManager.getState();
    /*
         * if the cluster state is not ACTIVE, then failbacks should not be processed
         * since some partitions are not active
         */
    if (state == ClusterState.ACTIVE) {
        while (!pendingProcessingFailbackPlans.isEmpty()) {
            //take the first pending failback plan
            NodeFailbackPlan plan = pendingProcessingFailbackPlans.pop();
            /*
                 * A plan at this stage will be in one of two states:
                 * 1. PREPARING -> the participants were selected but we haven't sent any request.
                 * 2. PENDING_ROLLBACK -> a participant failed before we send any requests
                 */
            if (plan.getState() == FailbackPlanState.PREPARING) {
                //set the partitions that will be failed back as inactive
                String failbackNode = plan.getNodeId();
                for (Integer partitionId : plan.getPartitionsToFailback()) {
                    //partition expected to be returned to the failing back node
                    clusterManager.updateClusterPartition(partitionId, failbackNode, false);
                }
                /*
                     * if the returning node is the original metadata node,
                     * then metadata node will change after the failback completes
                     */
                ICcApplicationContext appCtx = (ICcApplicationContext) serviceCtx.getApplicationContext();
                String originalMetadataNode = appCtx.getMetadataProperties().getMetadataNodeName();
                if (originalMetadataNode.equals(failbackNode)) {
                    plan.setNodeToReleaseMetadataManager(currentMetadataNode);
                    currentMetadataNode = "";
                    metadataNodeActive = false;
                    clusterManager.updateMetadataNode(currentMetadataNode, metadataNodeActive);
                }
                //force new jobs to wait
                clusterManager.setState(ClusterState.REBALANCING);
                handleFailbackRequests(plan, messageBroker);
                /*
                     * wait until the current plan is completed before processing the next plan.
                     * when the current one completes or is reverted, the cluster state will be
                     * ACTIVE again, and the next failback plan (if any) will be processed.
                     */
                break;
            } else if (plan.getState() == FailbackPlanState.PENDING_ROLLBACK) {
                //this plan failed before sending any requests -> nothing to rollback
                planId2FailbackPlanMap.remove(plan.getPlanId());
            }
        }
    }
}