Examples with FlinkException org.apache.flink.util.FlinkException used on opensource projects

Example 56 with FlinkException

use of org.apache.flink.util.FlinkException in project flink by apache.

the class StopWithSavepointTerminationHandlerImpl method terminateExceptionallyWithGlobalFailover.

/**
 * Handles the termination of the {@code StopWithSavepointTerminationHandler} exceptionally
 * after triggering a global job fail-over.
 *
 * @param unfinishedExecutionStates the unfinished states that caused the failure.
 * @param savepointPath the path to the successfully created savepoint.
 */
private void terminateExceptionallyWithGlobalFailover(Iterable<ExecutionState> unfinishedExecutionStates, String savepointPath) {
    String errorMessage = String.format("Inconsistent execution state after stopping with savepoint. At least one execution is still in one of the following states: %s. A global fail-over is triggered to recover the job %s.", StringUtils.join(unfinishedExecutionStates, ", "), jobId);
    FlinkException inconsistentFinalStateException = new FlinkException(errorMessage);
    log.warn("A savepoint was created at {} but the corresponding job {} didn't terminate successfully.", savepointPath, jobId, inconsistentFinalStateException);
    scheduler.handleGlobalFailure(inconsistentFinalStateException);
    result.completeExceptionally(inconsistentFinalStateException);
}

Example 57 with FlinkException

use of org.apache.flink.util.FlinkException in project flink by apache.

the class SlotSharingExecutionSlotAllocator method releaseSharedSlot.

private void releaseSharedSlot(ExecutionSlotSharingGroup executionSlotSharingGroup) {
    SharedSlot slot = sharedSlots.remove(executionSlotSharingGroup);
    Preconditions.checkNotNull(slot);
    Preconditions.checkState(slot.isEmpty(), "Trying to remove a shared slot with physical request id %s which has assigned logical slots", slot.getPhysicalSlotRequestId());
    slotProvider.cancelSlotRequest(slot.getPhysicalSlotRequestId(), new FlinkException("Slot is being returned from SlotSharingExecutionSlotAllocator."));
}

Example 58 with FlinkException

use of org.apache.flink.util.FlinkException in project flink by apache.

the class KubernetesResourceManagerDriver method onPodTerminated.

private void onPodTerminated(KubernetesPod pod) {
    final String podName = pod.getName();
    log.debug("TaskManager pod {} is terminated.", podName);
    // this is a safe net, in case onModified/onDeleted/onError is
    // received before onAdded
    final CompletableFuture<KubernetesWorkerNode> requestResourceFuture = requestResourceFutures.remove(podName);
    if (requestResourceFuture != null) {
        log.warn("Pod {} is terminated before being scheduled.", podName);
        requestResourceFuture.completeExceptionally(new FlinkException("Pod is terminated."));
    }
    getResourceEventHandler().onWorkerTerminated(new ResourceID(podName), pod.getTerminatedDiagnostics());
    stopPod(podName);
}

Example 59 with FlinkException

use of org.apache.flink.util.FlinkException in project flink by apache.

the class DefaultDeclarativeSlotPool method releaseIdleSlots.

@Override
public void releaseIdleSlots(long currentTimeMillis) {
    final Collection<AllocatedSlotPool.FreeSlotInfo> freeSlotsInformation = slotPool.getFreeSlotsInformation();
    ResourceCounter excessResources = fulfilledResourceRequirements.subtract(totalResourceRequirements);
    final Iterator<AllocatedSlotPool.FreeSlotInfo> freeSlotIterator = freeSlotsInformation.iterator();
    final Collection<AllocatedSlot> slotsToReturnToOwner = new ArrayList<>();
    while (!excessResources.isEmpty() && freeSlotIterator.hasNext()) {
        final AllocatedSlotPool.FreeSlotInfo idleSlot = freeSlotIterator.next();
        if (currentTimeMillis >= idleSlot.getFreeSince() + idleSlotTimeout.toMilliseconds()) {
            final ResourceProfile matchingProfile = getMatchingResourceProfile(idleSlot.getAllocationId());
            if (excessResources.containsResource(matchingProfile)) {
                excessResources = excessResources.subtract(matchingProfile, 1);
                final Optional<AllocatedSlot> removedSlot = slotPool.removeSlot(idleSlot.getAllocationId());
                final AllocatedSlot allocatedSlot = removedSlot.orElseThrow(() -> new IllegalStateException(String.format("Could not find slot for allocation id %s.", idleSlot.getAllocationId())));
                slotsToReturnToOwner.add(allocatedSlot);
            }
        }
    }
    releaseSlots(slotsToReturnToOwner, new FlinkException("Returning idle slots to their owners."));
    LOG.debug("Idle slots have been returned; new total acquired resources: {}", fulfilledResourceRequirements);
}

Example 60 with FlinkException

use of org.apache.flink.util.FlinkException in project flink by apache.

the class JobMaster method registerTaskManager.

@Override
public CompletableFuture<RegistrationResponse> registerTaskManager(final JobID jobId, final TaskManagerRegistrationInformation taskManagerRegistrationInformation, final Time timeout) {
    if (!jobGraph.getJobID().equals(jobId)) {
        log.debug("Rejecting TaskManager registration attempt because of wrong job id {}.", jobId);
        return CompletableFuture.completedFuture(new JMTMRegistrationRejection(String.format("The JobManager is not responsible for job %s. Maybe the TaskManager used outdated connection information.", jobId)));
    }
    final TaskManagerLocation taskManagerLocation;
    try {
        taskManagerLocation = resolveTaskManagerLocation(taskManagerRegistrationInformation.getUnresolvedTaskManagerLocation());
    } catch (FlinkException exception) {
        log.error("Could not accept TaskManager registration.", exception);
        return CompletableFuture.completedFuture(new RegistrationResponse.Failure(exception));
    }
    final ResourceID taskManagerId = taskManagerLocation.getResourceID();
    final UUID sessionId = taskManagerRegistrationInformation.getTaskManagerSession();
    final TaskManagerRegistration taskManagerRegistration = registeredTaskManagers.get(taskManagerId);
    if (taskManagerRegistration != null) {
        if (taskManagerRegistration.getSessionId().equals(sessionId)) {
            log.debug("Ignoring registration attempt of TaskManager {} with the same session id {}.", taskManagerId, sessionId);
            final RegistrationResponse response = new JMTMRegistrationSuccess(resourceId);
            return CompletableFuture.completedFuture(response);
        } else {
            disconnectTaskManager(taskManagerId, new FlinkException("A registered TaskManager re-registered with a new session id. This indicates a restart of the TaskManager. Closing the old connection."));
        }
    }
    return getRpcService().connect(taskManagerRegistrationInformation.getTaskManagerRpcAddress(), TaskExecutorGateway.class).handleAsync((TaskExecutorGateway taskExecutorGateway, Throwable throwable) -> {
        if (throwable != null) {
            return new RegistrationResponse.Failure(throwable);
        }
        slotPoolService.registerTaskManager(taskManagerId);
        registeredTaskManagers.put(taskManagerId, TaskManagerRegistration.create(taskManagerLocation, taskExecutorGateway, sessionId));
        // monitor the task manager as heartbeat target
        taskManagerHeartbeatManager.monitorTarget(taskManagerId, new TaskExecutorHeartbeatSender(taskExecutorGateway));
        return new JMTMRegistrationSuccess(resourceId);
    }, getMainThreadExecutor());
}