Examples with GridJobProcessor org.apache.ignite.internal.processors.job.GridJobProcessor used on opensource projects

Example 1 with GridJobProcessor

use of org.apache.ignite.internal.processors.job.GridJobProcessor in project ignite by apache.

the class IgniteKernal method start.

/**
     * @param cfg Configuration to use.
     * @param utilityCachePool Utility cache pool.
     * @param execSvc Executor service.
     * @param sysExecSvc System executor service.
     * @param stripedExecSvc Striped executor.
     * @param p2pExecSvc P2P executor service.
     * @param mgmtExecSvc Management executor service.
     * @param igfsExecSvc IGFS executor service.
     * @param dataStreamExecSvc data stream executor service.
     * @param restExecSvc Reset executor service.
     * @param affExecSvc Affinity executor service.
     * @param idxExecSvc Indexing executor service.
     * @param callbackExecSvc Callback executor service.
     * @param qryExecSvc Query executor service.
     * @param schemaExecSvc Schema executor service.
     * @param customExecSvcs Custom named executors.
     * @param errHnd Error handler to use for notification about startup problems.
     * @throws IgniteCheckedException Thrown in case of any errors.
     */
@SuppressWarnings({ "CatchGenericClass", "unchecked" })
public void start(final IgniteConfiguration cfg, ExecutorService utilityCachePool, final ExecutorService execSvc, final ExecutorService svcExecSvc, final ExecutorService sysExecSvc, final StripedExecutor stripedExecSvc, ExecutorService p2pExecSvc, ExecutorService mgmtExecSvc, ExecutorService igfsExecSvc, ExecutorService dataStreamExecSvc, ExecutorService restExecSvc, ExecutorService affExecSvc, @Nullable ExecutorService idxExecSvc, IgniteStripedThreadPoolExecutor callbackExecSvc, ExecutorService qryExecSvc, ExecutorService schemaExecSvc, Map<String, ? extends ExecutorService> customExecSvcs, GridAbsClosure errHnd) throws IgniteCheckedException {
    gw.compareAndSet(null, new GridKernalGatewayImpl(cfg.getIgniteInstanceName()));
    GridKernalGateway gw = this.gw.get();
    gw.writeLock();
    try {
        switch(gw.getState()) {
            case STARTED:
                {
                    U.warn(log, "Grid has already been started (ignored).");
                    return;
                }
            case STARTING:
                {
                    U.warn(log, "Grid is already in process of being started (ignored).");
                    return;
                }
            case STOPPING:
                {
                    throw new IgniteCheckedException("Grid is in process of being stopped");
                }
            case STOPPED:
                {
                    break;
                }
        }
        gw.setState(STARTING);
    } finally {
        gw.writeUnlock();
    }
    assert cfg != null;
    // Make sure we got proper configuration.
    validateCommon(cfg);
    igniteInstanceName = cfg.getIgniteInstanceName();
    this.cfg = cfg;
    log = (GridLoggerProxy) cfg.getGridLogger().getLogger(getClass().getName() + (igniteInstanceName != null ? '%' + igniteInstanceName : ""));
    RuntimeMXBean rtBean = ManagementFactory.getRuntimeMXBean();
    // Ack various information.
    ackAsciiLogo();
    ackConfigUrl();
    ackDaemon();
    ackOsInfo();
    ackLanguageRuntime();
    ackRemoteManagement();
    ackVmArguments(rtBean);
    ackClassPaths(rtBean);
    ackSystemProperties();
    ackEnvironmentVariables();
    ackMemoryConfiguration();
    ackCacheConfiguration();
    ackP2pConfiguration();
    ackRebalanceConfiguration();
    // Run background network diagnostics.
    GridDiagnostic.runBackgroundCheck(igniteInstanceName, execSvc, log);
    // Ack 3-rd party licenses location.
    if (log.isInfoEnabled() && cfg.getIgniteHome() != null)
        log.info("3-rd party licenses can be found at: " + cfg.getIgniteHome() + File.separatorChar + "libs" + File.separatorChar + "licenses");
    // with internally reserved names.
    for (String name : cfg.getUserAttributes().keySet()) if (name.startsWith(ATTR_PREFIX))
        throw new IgniteCheckedException("User attribute has illegal name: '" + name + "'. Note that all names " + "starting with '" + ATTR_PREFIX + "' are reserved for internal use.");
    // Ack local node user attributes.
    logNodeUserAttributes();
    // Ack configuration.
    ackSpis();
    List<PluginProvider> plugins = U.allPluginProviders();
    final boolean activeOnStart = cfg.isActiveOnStart();
    // Spin out SPIs & managers.
    try {
        ctx = new GridKernalContextImpl(log, this, cfg, gw, utilityCachePool, execSvc, svcExecSvc, sysExecSvc, stripedExecSvc, p2pExecSvc, mgmtExecSvc, igfsExecSvc, dataStreamExecSvc, restExecSvc, affExecSvc, idxExecSvc, callbackExecSvc, qryExecSvc, schemaExecSvc, customExecSvcs, plugins);
        cfg.getMarshaller().setContext(ctx.marshallerContext());
        ClusterProcessor clusterProc = new ClusterProcessor(ctx);
        startProcessor(clusterProc);
        U.onGridStart();
        // Start and configure resource processor first as it contains resources used
        // by all other managers and processors.
        GridResourceProcessor rsrcProc = new GridResourceProcessor(ctx);
        rsrcProc.setSpringContext(rsrcCtx);
        scheduler = new IgniteSchedulerImpl(ctx);
        startProcessor(rsrcProc);
        // Inject resources into lifecycle beans.
        if (!cfg.isDaemon() && cfg.getLifecycleBeans() != null) {
            for (LifecycleBean bean : cfg.getLifecycleBeans()) {
                if (bean != null)
                    rsrcProc.inject(bean);
            }
        }
        // Lifecycle notification.
        notifyLifecycleBeans(BEFORE_NODE_START);
        // Starts lifecycle aware components.
        U.startLifecycleAware(lifecycleAwares(cfg));
        addHelper(IGFS_HELPER.create(F.isEmpty(cfg.getFileSystemConfiguration())));
        addHelper(HADOOP_HELPER.createIfInClassPath(ctx, false));
        startProcessor(new IgnitePluginProcessor(ctx, cfg, plugins));
        startProcessor(new PoolProcessor(ctx));
        // Closure processor should be started before all others
        // (except for resource processor), as many components can depend on it.
        startProcessor(new GridClosureProcessor(ctx));
        // Start some other processors (order & place is important).
        startProcessor(new GridPortProcessor(ctx));
        startProcessor(new GridJobMetricsProcessor(ctx));
        // Timeout processor needs to be started before managers,
        // as managers may depend on it.
        startProcessor(new GridTimeoutProcessor(ctx));
        // Start security processors.
        startProcessor(createComponent(GridSecurityProcessor.class, ctx));
        // Start SPI managers.
        // NOTE: that order matters as there are dependencies between managers.
        startManager(new GridIoManager(ctx));
        startManager(new GridCheckpointManager(ctx));
        startManager(new GridEventStorageManager(ctx));
        startManager(new GridDeploymentManager(ctx));
        startManager(new GridLoadBalancerManager(ctx));
        startManager(new GridFailoverManager(ctx));
        startManager(new GridCollisionManager(ctx));
        startManager(new GridIndexingManager(ctx));
        ackSecurity();
        // Assign discovery manager to context before other processors start so they
        // are able to register custom event listener.
        GridManager discoMgr = new GridDiscoveryManager(ctx);
        ctx.add(discoMgr, false);
        // be able to start receiving messages once discovery completes.
        try {
            startProcessor(createComponent(DiscoveryNodeValidationProcessor.class, ctx));
            startProcessor(new GridAffinityProcessor(ctx));
            startProcessor(createComponent(GridSegmentationProcessor.class, ctx));
            startProcessor(createComponent(IgniteCacheObjectProcessor.class, ctx));
            startProcessor(new GridCacheProcessor(ctx));
            startProcessor(new GridClusterStateProcessor(ctx));
            startProcessor(new GridQueryProcessor(ctx));
            startProcessor(new SqlListenerProcessor(ctx));
            startProcessor(new GridServiceProcessor(ctx));
            startProcessor(new GridTaskSessionProcessor(ctx));
            startProcessor(new GridJobProcessor(ctx));
            startProcessor(new GridTaskProcessor(ctx));
            startProcessor((GridProcessor) SCHEDULE.createOptional(ctx));
            startProcessor(new GridRestProcessor(ctx));
            startProcessor(new DataStreamProcessor(ctx));
            startProcessor((GridProcessor) IGFS.create(ctx, F.isEmpty(cfg.getFileSystemConfiguration())));
            startProcessor(new GridContinuousProcessor(ctx));
            startProcessor(createHadoopComponent());
            startProcessor(new DataStructuresProcessor(ctx));
            startProcessor(createComponent(PlatformProcessor.class, ctx));
            startProcessor(new GridMarshallerMappingProcessor(ctx));
            // Start plugins.
            for (PluginProvider provider : ctx.plugins().allProviders()) {
                ctx.add(new GridPluginComponent(provider));
                provider.start(ctx.plugins().pluginContextForProvider(provider));
            }
            // Start platform plugins.
            if (ctx.config().getPlatformConfiguration() != null)
                startProcessor(new PlatformPluginProcessor(ctx));
            fillNodeAttributes(clusterProc.updateNotifierEnabled());
        } catch (Throwable e) {
            U.error(log, "Exception during start processors, node will be stopped and close connections", e);
            // Stop discovery spi to close tcp socket.
            ctx.discovery().stop(true);
            throw e;
        }
        gw.writeLock();
        try {
            gw.setState(STARTED);
            // Start discovery manager last to make sure that grid is fully initialized.
            startManager(discoMgr);
        } finally {
            gw.writeUnlock();
        }
        // Check whether physical RAM is not exceeded.
        checkPhysicalRam();
        // Suggest configuration optimizations.
        suggestOptimizations(cfg);
        // Suggest JVM optimizations.
        ctx.performance().addAll(JvmConfigurationSuggestions.getSuggestions());
        // Suggest Operation System optimizations.
        ctx.performance().addAll(OsConfigurationSuggestions.getSuggestions());
        // Notify discovery manager the first to make sure that topology is discovered.
        ctx.discovery().onKernalStart(activeOnStart);
        // Notify IO manager the second so further components can send and receive messages.
        ctx.io().onKernalStart(activeOnStart);
        // Start plugins.
        for (PluginProvider provider : ctx.plugins().allProviders()) provider.onIgniteStart();
        boolean recon = false;
        // Callbacks.
        for (GridComponent comp : ctx) {
            // Skip discovery manager.
            if (comp instanceof GridDiscoveryManager)
                continue;
            // Skip IO manager.
            if (comp instanceof GridIoManager)
                continue;
            if (comp instanceof GridPluginComponent)
                continue;
            if (!skipDaemon(comp)) {
                try {
                    comp.onKernalStart(activeOnStart);
                } catch (IgniteNeedReconnectException e) {
                    assert ctx.discovery().reconnectSupported();
                    if (log.isDebugEnabled())
                        log.debug("Failed to start node components on node start, will wait for reconnect: " + e);
                    recon = true;
                }
            }
        }
        if (recon)
            reconnectState.waitFirstReconnect();
        // Register MBeans.
        registerKernalMBean();
        registerLocalNodeMBean();
        registerExecutorMBeans(execSvc, sysExecSvc, p2pExecSvc, mgmtExecSvc, restExecSvc, qryExecSvc, schemaExecSvc);
        registerStripedExecutorMBean(stripedExecSvc);
        // Lifecycle bean notifications.
        notifyLifecycleBeans(AFTER_NODE_START);
    } catch (Throwable e) {
        IgniteSpiVersionCheckException verCheckErr = X.cause(e, IgniteSpiVersionCheckException.class);
        if (verCheckErr != null)
            U.error(log, verCheckErr.getMessage());
        else if (X.hasCause(e, InterruptedException.class, IgniteInterruptedCheckedException.class))
            U.warn(log, "Grid startup routine has been interrupted (will rollback).");
        else
            U.error(log, "Got exception while starting (will rollback startup routine).", e);
        errHnd.apply();
        stop(true);
        if (e instanceof Error)
            throw e;
        else if (e instanceof IgniteCheckedException)
            throw (IgniteCheckedException) e;
        else
            throw new IgniteCheckedException(e);
    }
    // Mark start timestamp.
    startTime = U.currentTimeMillis();
    String intervalStr = IgniteSystemProperties.getString(IGNITE_STARVATION_CHECK_INTERVAL);
    // Start starvation checker if enabled.
    boolean starveCheck = !isDaemon() && !"0".equals(intervalStr);
    if (starveCheck) {
        final long interval = F.isEmpty(intervalStr) ? PERIODIC_STARVATION_CHECK_FREQ : Long.parseLong(intervalStr);
        starveTask = ctx.timeout().schedule(new Runnable() {

            /** Last completed task count. */
            private long lastCompletedCntPub;

            /** Last completed task count. */
            private long lastCompletedCntSys;

            @Override
            public void run() {
                if (execSvc instanceof ThreadPoolExecutor) {
                    ThreadPoolExecutor exec = (ThreadPoolExecutor) execSvc;
                    lastCompletedCntPub = checkPoolStarvation(exec, lastCompletedCntPub, "public");
                }
                if (sysExecSvc instanceof ThreadPoolExecutor) {
                    ThreadPoolExecutor exec = (ThreadPoolExecutor) sysExecSvc;
                    lastCompletedCntSys = checkPoolStarvation(exec, lastCompletedCntSys, "system");
                }
                if (stripedExecSvc != null)
                    stripedExecSvc.checkStarvation();
            }

            /**
                 * @param exec Thread pool executor to check.
                 * @param lastCompletedCnt Last completed tasks count.
                 * @param pool Pool name for message.
                 * @return Current completed tasks count.
                 */
            private long checkPoolStarvation(ThreadPoolExecutor exec, long lastCompletedCnt, String pool) {
                long completedCnt = exec.getCompletedTaskCount();
                // at least one waiting request, then it is possible starvation.
                if (exec.getPoolSize() == exec.getActiveCount() && completedCnt == lastCompletedCnt && !exec.getQueue().isEmpty())
                    LT.warn(log, "Possible thread pool starvation detected (no task completed in last " + interval + "ms, is " + pool + " thread pool size large enough?)");
                return completedCnt;
            }
        }, interval, interval);
    }
    long metricsLogFreq = cfg.getMetricsLogFrequency();
    if (metricsLogFreq > 0) {
        metricsLogTask = ctx.timeout().schedule(new Runnable() {

            private final DecimalFormat dblFmt = new DecimalFormat("#.##");

            @Override
            public void run() {
                if (log.isInfoEnabled()) {
                    try {
                        ClusterMetrics m = cluster().localNode().metrics();
                        double cpuLoadPct = m.getCurrentCpuLoad() * 100;
                        double avgCpuLoadPct = m.getAverageCpuLoad() * 100;
                        double gcPct = m.getCurrentGcCpuLoad() * 100;
                        //Heap params
                        long heapUsed = m.getHeapMemoryUsed();
                        long heapMax = m.getHeapMemoryMaximum();
                        long heapUsedInMBytes = heapUsed / 1024 / 1024;
                        long heapCommInMBytes = m.getHeapMemoryCommitted() / 1024 / 1024;
                        double freeHeapPct = heapMax > 0 ? ((double) ((heapMax - heapUsed) * 100)) / heapMax : -1;
                        //Non heap params
                        long nonHeapUsed = m.getNonHeapMemoryUsed();
                        long nonHeapMax = m.getNonHeapMemoryMaximum();
                        long nonHeapUsedInMBytes = nonHeapUsed / 1024 / 1024;
                        long nonHeapCommInMBytes = m.getNonHeapMemoryCommitted() / 1024 / 1024;
                        double freeNonHeapPct = nonHeapMax > 0 ? ((double) ((nonHeapMax - nonHeapUsed) * 100)) / nonHeapMax : -1;
                        int hosts = 0;
                        int nodes = 0;
                        int cpus = 0;
                        try {
                            ClusterMetrics metrics = cluster().metrics();
                            Collection<ClusterNode> nodes0 = cluster().nodes();
                            hosts = U.neighborhood(nodes0).size();
                            nodes = metrics.getTotalNodes();
                            cpus = metrics.getTotalCpus();
                        } catch (IgniteException ignore) {
                        // No-op.
                        }
                        int pubPoolActiveThreads = 0;
                        int pubPoolIdleThreads = 0;
                        int pubPoolQSize = 0;
                        if (execSvc instanceof ThreadPoolExecutor) {
                            ThreadPoolExecutor exec = (ThreadPoolExecutor) execSvc;
                            int poolSize = exec.getPoolSize();
                            pubPoolActiveThreads = Math.min(poolSize, exec.getActiveCount());
                            pubPoolIdleThreads = poolSize - pubPoolActiveThreads;
                            pubPoolQSize = exec.getQueue().size();
                        }
                        int sysPoolActiveThreads = 0;
                        int sysPoolIdleThreads = 0;
                        int sysPoolQSize = 0;
                        if (sysExecSvc instanceof ThreadPoolExecutor) {
                            ThreadPoolExecutor exec = (ThreadPoolExecutor) sysExecSvc;
                            int poolSize = exec.getPoolSize();
                            sysPoolActiveThreads = Math.min(poolSize, exec.getActiveCount());
                            sysPoolIdleThreads = poolSize - sysPoolActiveThreads;
                            sysPoolQSize = exec.getQueue().size();
                        }
                        int loadedPages = 0;
                        Collection<MemoryPolicy> policies = ctx.cache().context().database().memoryPolicies();
                        if (!F.isEmpty(policies)) {
                            for (MemoryPolicy memPlc : policies) loadedPages += memPlc.pageMemory().loadedPages();
                        }
                        String id = U.id8(localNode().id());
                        String msg = NL + "Metrics for local node (to disable set 'metricsLogFrequency' to 0)" + NL + "    ^-- Node [id=" + id + ", name=" + name() + ", uptime=" + getUpTimeFormatted() + "]" + NL + "    ^-- H/N/C [hosts=" + hosts + ", nodes=" + nodes + ", CPUs=" + cpus + "]" + NL + "    ^-- CPU [cur=" + dblFmt.format(cpuLoadPct) + "%, avg=" + dblFmt.format(avgCpuLoadPct) + "%, GC=" + dblFmt.format(gcPct) + "%]" + NL + "    ^-- PageMemory [pages=" + loadedPages + "]" + NL + "    ^-- Heap [used=" + dblFmt.format(heapUsedInMBytes) + "MB, free=" + dblFmt.format(freeHeapPct) + "%, comm=" + dblFmt.format(heapCommInMBytes) + "MB]" + NL + "    ^-- Non heap [used=" + dblFmt.format(nonHeapUsedInMBytes) + "MB, free=" + dblFmt.format(freeNonHeapPct) + "%, comm=" + dblFmt.format(nonHeapCommInMBytes) + "MB]" + NL + "    ^-- Public thread pool [active=" + pubPoolActiveThreads + ", idle=" + pubPoolIdleThreads + ", qSize=" + pubPoolQSize + "]" + NL + "    ^-- System thread pool [active=" + sysPoolActiveThreads + ", idle=" + sysPoolIdleThreads + ", qSize=" + sysPoolQSize + "]" + NL + "    ^-- Outbound messages queue [size=" + m.getOutboundMessagesQueueSize() + "]";
                        log.info(msg);
                        ctx.cache().context().database().dumpStatistics(log);
                    } catch (IgniteClientDisconnectedException ignore) {
                    // No-op.
                    }
                }
            }
        }, metricsLogFreq, metricsLogFreq);
    }
    final long longOpDumpTimeout = IgniteSystemProperties.getLong(IgniteSystemProperties.IGNITE_LONG_OPERATIONS_DUMP_TIMEOUT, 60_000);
    if (longOpDumpTimeout > 0) {
        longOpDumpTask = ctx.timeout().schedule(new Runnable() {

            @Override
            public void run() {
                GridKernalContext ctx = IgniteKernal.this.ctx;
                if (ctx != null)
                    ctx.cache().context().exchange().dumpLongRunningOperations(longOpDumpTimeout);
            }
        }, longOpDumpTimeout, longOpDumpTimeout);
    }
    ctx.performance().add("Disable assertions (remove '-ea' from JVM options)", !U.assertionsEnabled());
    ctx.performance().logSuggestions(log, igniteInstanceName);
    U.quietAndInfo(log, "To start Console Management & Monitoring run ignitevisorcmd.{sh|bat}");
    ackStart(rtBean);
    if (!isDaemon())
        ctx.discovery().ackTopology(localNode().order());
}