Examples with CheckLightJobsOperation com.hazelcast.jet.impl.operation.CheckLightJobsOperation used on opensource projects

Example 1 with CheckLightJobsOperation

use of com.hazelcast.jet.impl.operation.CheckLightJobsOperation in project hazelcast by hazelcast.

the class JobExecutionService method checkExecutions.

/**
 * See also javadoc at {@link CheckLightJobsOperation}.
 */
private void checkExecutions() {
    try {
        long now = System.nanoTime();
        long uninitializedContextThreshold = now - UNINITIALIZED_CONTEXT_MAX_AGE_NS;
        Map<Address, List<Long>> executionsPerMember = new HashMap<>();
        for (ExecutionContext ctx : executionContexts.values()) {
            if (!ctx.isLightJob()) {
                continue;
            }
            Address coordinator = ctx.coordinator();
            if (coordinator != null) {
                // if coordinator is known, add execution to the list to check
                executionsPerMember.computeIfAbsent(coordinator, k -> new ArrayList<>()).add(ctx.executionId());
            } else {
                // if coordinator is not known, remove execution if it's not known for too long
                if (ctx.getCreatedOn() <= uninitializedContextThreshold) {
                    LoggingUtil.logFine(logger, "Terminating light job %s because it wasn't initialized during %d seconds", idToString(ctx.executionId()), NANOSECONDS.toSeconds(UNINITIALIZED_CONTEXT_MAX_AGE_NS));
                    terminateExecution0(ctx, TerminationMode.CANCEL_FORCEFUL, new CancellationException());
                }
            }
        }
        // submit the query to the coordinator
        for (Entry<Address, List<Long>> en : executionsPerMember.entrySet()) {
            long[] executionIds = en.getValue().stream().mapToLong(Long::longValue).toArray();
            Operation op = new CheckLightJobsOperation(executionIds);
            InvocationFuture<long[]> future = nodeEngine.getOperationService().createInvocationBuilder(JetServiceBackend.SERVICE_NAME, op, en.getKey()).invoke();
            future.whenComplete((r, t) -> {
                if (t instanceof TargetNotMemberException) {
                    // if the target isn't a member, then all executions are unknown
                    r = executionIds;
                } else if (t != null) {
                    logger.warning("Failed to check light job state with coordinator " + en.getKey() + ": " + t, t);
                    return;
                }
                assert r != null;
                for (long executionId : r) {
                    ExecutionContext execCtx = executionContexts.get(executionId);
                    if (execCtx != null) {
                        logger.fine("Terminating light job " + idToString(executionId) + " because the coordinator doesn't know it");
                        terminateExecution0(execCtx, TerminationMode.CANCEL_FORCEFUL, new CancellationException());
                    }
                }
            });
        }
        // clean up failedJobs
        failedJobs.values().removeIf(expiryTime -> expiryTime < now);
    } catch (Throwable e) {
        logger.severe("Failed to query live light executions: " + e, e);
    }
}