Search in sources :

Example 1 with RESTART

use of com.hazelcast.jet.impl.TerminationMode.ActionAfterTerminate.RESTART in project hazelcast by hazelcast.

the class MasterJobContext method tryStartJob.

/**
 * Starts the execution of the job if it is not already completed,
 * cancelled or failed.
 * <p>
 * If the job is already cancelled, triggers the job completion procedure.
 * <p>
 * If the job quorum is not satisfied, reschedules the job restart.
 * <p>
 * If there was a membership change and the partition table is not completely
 * fixed yet, reschedules the job restart.
 */
void tryStartJob(Supplier<Long> executionIdSupplier) {
    mc.coordinationService().submitToCoordinatorThread(() -> {
        executionStartTime = System.currentTimeMillis();
        try {
            JobExecutionRecord jobExecRec = mc.jobExecutionRecord();
            jobExecRec.markExecuted();
            Tuple2<DAG, ClassLoader> dagAndClassloader = resolveDagAndCL(executionIdSupplier);
            if (dagAndClassloader == null) {
                return;
            }
            DAG dag = dagAndClassloader.f0();
            assert dag != null;
            ClassLoader classLoader = dagAndClassloader.f1();
            // must call this before rewriteDagWithSnapshotRestore()
            String dotRepresentation = dag.toDotString(defaultParallelism, defaultQueueSize);
            long snapshotId = jobExecRec.snapshotId();
            String snapshotName = mc.jobConfig().getInitialSnapshotName();
            String mapName = snapshotId >= 0 ? jobExecRec.successfulSnapshotDataMapName(mc.jobId()) : snapshotName != null ? EXPORTED_SNAPSHOTS_PREFIX + snapshotName : null;
            if (mapName != null) {
                rewriteDagWithSnapshotRestore(dag, snapshotId, mapName, snapshotName);
            } else {
                logger.info("Didn't find any snapshot to restore for " + mc.jobIdString());
            }
            MembersView membersView = Util.getMembersView(mc.nodeEngine());
            logger.info("Start executing " + mc.jobIdString() + ", execution graph in DOT format:\n" + dotRepresentation + "\nHINT: You can use graphviz or http://viz-js.com to visualize the printed graph.");
            logger.fine("Building execution plan for " + mc.jobIdString());
            Util.doWithClassLoader(classLoader, () -> mc.setExecutionPlanMap(createExecutionPlans(mc.nodeEngine(), membersView.getMembers(), dag, mc.jobId(), mc.executionId(), mc.jobConfig(), jobExecRec.ongoingSnapshotId(), false, mc.jobRecord().getSubject())));
            logger.fine("Built execution plans for " + mc.jobIdString());
            Set<MemberInfo> participants = mc.executionPlanMap().keySet();
            Version coordinatorVersion = mc.nodeEngine().getLocalMember().getVersion().asVersion();
            Function<ExecutionPlan, Operation> operationCtor = plan -> new InitExecutionOperation(mc.jobId(), mc.executionId(), membersView.getVersion(), coordinatorVersion, participants, mc.nodeEngine().getSerializationService().toData(plan), false);
            mc.invokeOnParticipants(operationCtor, this::onInitStepCompleted, null, false);
        } catch (Throwable e) {
            finalizeJob(e);
        }
    });
}
Also used : Address(com.hazelcast.cluster.Address) SUSPEND(com.hazelcast.jet.impl.TerminationMode.ActionAfterTerminate.SUSPEND) NOT_RUNNING(com.hazelcast.jet.core.JobStatus.NOT_RUNNING) GetLocalJobMetricsOperation(com.hazelcast.jet.impl.operation.GetLocalJobMetricsOperation) CompletableFuture.completedFuture(java.util.concurrent.CompletableFuture.completedFuture) NonCompletableFuture(com.hazelcast.jet.impl.util.NonCompletableFuture) ExceptionUtil.isTopologyException(com.hazelcast.jet.impl.util.ExceptionUtil.isTopologyException) JobTerminateRequestedException(com.hazelcast.jet.impl.exception.JobTerminateRequestedException) SourceProcessors.readMapP(com.hazelcast.jet.core.processor.SourceProcessors.readMapP) RESTART(com.hazelcast.jet.impl.TerminationMode.ActionAfterTerminate.RESTART) JetDelegatingClassLoader(com.hazelcast.jet.impl.deployment.JetDelegatingClassLoader) TerminatedWithSnapshotException(com.hazelcast.jet.impl.exception.TerminatedWithSnapshotException) Collectors.toMap(java.util.stream.Collectors.toMap) Functions.entryKey(com.hazelcast.function.Functions.entryKey) MemberInfo(com.hazelcast.internal.cluster.MemberInfo) Map(java.util.Map) STARTING(com.hazelcast.jet.core.JobStatus.STARTING) SUSPENDED(com.hazelcast.jet.core.JobStatus.SUSPENDED) DAG(com.hazelcast.jet.core.DAG) JobStatus(com.hazelcast.jet.core.JobStatus) ExceptionUtil(com.hazelcast.jet.impl.util.ExceptionUtil) JobMetrics(com.hazelcast.jet.core.metrics.JobMetrics) CancellationException(java.util.concurrent.CancellationException) CANCEL_GRACEFUL(com.hazelcast.jet.impl.TerminationMode.CANCEL_GRACEFUL) Collections.emptyList(java.util.Collections.emptyList) Collection(java.util.Collection) Set(java.util.Set) UUID(java.util.UUID) MILLISECONDS(java.util.concurrent.TimeUnit.MILLISECONDS) Collectors(java.util.stream.Collectors) CANCEL_FORCEFUL(com.hazelcast.jet.impl.TerminationMode.CANCEL_FORCEFUL) Objects(java.util.Objects) Util(com.hazelcast.jet.impl.util.Util) List(java.util.List) Util.idToString(com.hazelcast.jet.Util.idToString) ExecutionPlan(com.hazelcast.jet.impl.execution.init.ExecutionPlan) MetricNames(com.hazelcast.jet.core.metrics.MetricNames) Entry(java.util.Map.Entry) TopologyChangedException(com.hazelcast.jet.core.TopologyChangedException) COMPLETED(com.hazelcast.jet.core.JobStatus.COMPLETED) JetDisabledException(com.hazelcast.jet.impl.exception.JetDisabledException) LoggingUtil(com.hazelcast.jet.impl.util.LoggingUtil) ExecutionPlanBuilder.createExecutionPlans(com.hazelcast.jet.impl.execution.init.ExecutionPlanBuilder.createExecutionPlans) Collectors.partitioningBy(java.util.stream.Collectors.partitioningBy) TerminateExecutionOperation(com.hazelcast.jet.impl.operation.TerminateExecutionOperation) ExceptionUtil.isRestartableException(com.hazelcast.jet.impl.util.ExceptionUtil.isRestartableException) AtomicBoolean(java.util.concurrent.atomic.AtomicBoolean) LoggingUtil.logFinest(com.hazelcast.jet.impl.util.LoggingUtil.logFinest) Util.doWithClassLoader(com.hazelcast.jet.impl.util.Util.doWithClassLoader) HashMap(java.util.HashMap) CompletableFuture(java.util.concurrent.CompletableFuture) ExecutionService(com.hazelcast.spi.impl.executionservice.ExecutionService) StartExecutionOperation(com.hazelcast.jet.impl.operation.StartExecutionOperation) Function(java.util.function.Function) Supplier(java.util.function.Supplier) Util.formatJobDuration(com.hazelcast.jet.impl.util.Util.formatJobDuration) ActionAfterTerminate(com.hazelcast.jet.impl.TerminationMode.ActionAfterTerminate) ExecutionNotFoundException(com.hazelcast.jet.impl.exception.ExecutionNotFoundException) ArrayList(java.util.ArrayList) JetException(com.hazelcast.jet.JetException) HashSet(java.util.HashSet) InitExecutionOperation(com.hazelcast.jet.impl.operation.InitExecutionOperation) COORDINATOR(com.hazelcast.jet.impl.JobClassLoaderService.JobPhase.COORDINATOR) ILogger(com.hazelcast.logging.ILogger) SnapshotValidator.validateSnapshot(com.hazelcast.jet.impl.SnapshotValidator.validateSnapshot) ExceptionUtil.rethrow(com.hazelcast.jet.impl.util.ExceptionUtil.rethrow) Operation(com.hazelcast.spi.impl.operationservice.Operation) Util.entry(com.hazelcast.jet.Util.entry) ExceptionUtil.withTryCatch(com.hazelcast.jet.impl.util.ExceptionUtil.withTryCatch) BiConsumer(java.util.function.BiConsumer) MembersView(com.hazelcast.internal.cluster.impl.MembersView) LocalMemberResetException(com.hazelcast.core.LocalMemberResetException) RESTART_GRACEFUL(com.hazelcast.jet.impl.TerminationMode.RESTART_GRACEFUL) Edge(com.hazelcast.jet.core.Edge) Version(com.hazelcast.version.Version) EXPORTED_SNAPSHOTS_PREFIX(com.hazelcast.jet.impl.JobRepository.EXPORTED_SNAPSHOTS_PREFIX) Nonnull(javax.annotation.Nonnull) Tuple2(com.hazelcast.jet.datamodel.Tuple2) Nullable(javax.annotation.Nullable) Job(com.hazelcast.jet.Job) Measurement(com.hazelcast.jet.core.metrics.Measurement) SUSPENDED_EXPORTING_SNAPSHOT(com.hazelcast.jet.core.JobStatus.SUSPENDED_EXPORTING_SNAPSHOT) Util.toList(com.hazelcast.jet.impl.util.Util.toList) RawJobMetrics(com.hazelcast.jet.impl.metrics.RawJobMetrics) MetricTags(com.hazelcast.jet.core.metrics.MetricTags) NONE(com.hazelcast.jet.config.ProcessingGuarantee.NONE) Consumer(java.util.function.Consumer) Vertex(com.hazelcast.jet.core.Vertex) Tuple2.tuple2(com.hazelcast.jet.datamodel.Tuple2.tuple2) CustomClassLoadedObject.deserializeWithCustomClassLoader(com.hazelcast.jet.impl.execution.init.CustomClassLoadedObject.deserializeWithCustomClassLoader) ExceptionUtil.peel(com.hazelcast.jet.impl.util.ExceptionUtil.peel) FAILED(com.hazelcast.jet.core.JobStatus.FAILED) RUNNING(com.hazelcast.jet.core.JobStatus.RUNNING) Collections(java.util.Collections) IMap(com.hazelcast.map.IMap) Edge.between(com.hazelcast.jet.core.Edge.between) MembersView(com.hazelcast.internal.cluster.impl.MembersView) DAG(com.hazelcast.jet.core.DAG) Util.idToString(com.hazelcast.jet.Util.idToString) GetLocalJobMetricsOperation(com.hazelcast.jet.impl.operation.GetLocalJobMetricsOperation) TerminateExecutionOperation(com.hazelcast.jet.impl.operation.TerminateExecutionOperation) StartExecutionOperation(com.hazelcast.jet.impl.operation.StartExecutionOperation) InitExecutionOperation(com.hazelcast.jet.impl.operation.InitExecutionOperation) Operation(com.hazelcast.spi.impl.operationservice.Operation) ExecutionPlan(com.hazelcast.jet.impl.execution.init.ExecutionPlan) MemberInfo(com.hazelcast.internal.cluster.MemberInfo) Version(com.hazelcast.version.Version) InitExecutionOperation(com.hazelcast.jet.impl.operation.InitExecutionOperation) JetDelegatingClassLoader(com.hazelcast.jet.impl.deployment.JetDelegatingClassLoader) Util.doWithClassLoader(com.hazelcast.jet.impl.util.Util.doWithClassLoader) CustomClassLoadedObject.deserializeWithCustomClassLoader(com.hazelcast.jet.impl.execution.init.CustomClassLoadedObject.deserializeWithCustomClassLoader)

Aggregations

Address (com.hazelcast.cluster.Address)1 LocalMemberResetException (com.hazelcast.core.LocalMemberResetException)1 Functions.entryKey (com.hazelcast.function.Functions.entryKey)1 MemberInfo (com.hazelcast.internal.cluster.MemberInfo)1 MembersView (com.hazelcast.internal.cluster.impl.MembersView)1 JetException (com.hazelcast.jet.JetException)1 Job (com.hazelcast.jet.Job)1 Util.entry (com.hazelcast.jet.Util.entry)1 Util.idToString (com.hazelcast.jet.Util.idToString)1 NONE (com.hazelcast.jet.config.ProcessingGuarantee.NONE)1 DAG (com.hazelcast.jet.core.DAG)1 Edge (com.hazelcast.jet.core.Edge)1 Edge.between (com.hazelcast.jet.core.Edge.between)1 JobStatus (com.hazelcast.jet.core.JobStatus)1 COMPLETED (com.hazelcast.jet.core.JobStatus.COMPLETED)1 FAILED (com.hazelcast.jet.core.JobStatus.FAILED)1 NOT_RUNNING (com.hazelcast.jet.core.JobStatus.NOT_RUNNING)1 RUNNING (com.hazelcast.jet.core.JobStatus.RUNNING)1 STARTING (com.hazelcast.jet.core.JobStatus.STARTING)1 SUSPENDED (com.hazelcast.jet.core.JobStatus.SUSPENDED)1