Examples with CruiseControlUserTaskStatus io.strimzi.operator.cluster.operator.resource.cruisecontrol.CruiseControlUserTaskStatus used on opensource projects

Example 1 with CruiseControlUserTaskStatus

use of io.strimzi.operator.cluster.operator.resource.cruisecontrol.CruiseControlUserTaskStatus in project strimzi by strimzi.

the class KafkaRebalanceAssemblyOperator method onRebalancing.

/**
 * This method handles the transition from {@code Rebalancing} state.
 * It starts a periodic timer in order to check the status of the ongoing rebalance processing on Cruise Control side.
 * In order to do that, it calls the related Cruise Control REST API about asking the user task status.
 * When the rebalance is finished, the next state is {@code Ready}.
 * If the user sets the strimzi.io/rebalance annotation to 'stop', it calls the Cruise Control REST API for stopping the ongoing task
 * and then transitions to the {@code Stopped} state.
 * If the user sets any other values for the strimzi.io/rebalance annotation, it is just ignored and the user task checks continue.
 * This method holds the lock until the rebalance is finished, the ongoing task is stopped or any exception is raised.
 *
 * @param reconciliation Reconciliation information
 * @param host Cruise Control service to which sending the REST API requests
 * @param apiClient Cruise Control REST API client instance
 * @param kafkaRebalance Current {@code KafkaRebalance} resource
 * @param rebalanceAnnotation The current value for the strimzi.io/rebalance annotation
 * @return a Future with the next {@code MapAndStatus<ConfigMap, KafkaRebalanceStatus>} including the state
 */
private Future<MapAndStatus<ConfigMap, KafkaRebalanceStatus>> onRebalancing(Reconciliation reconciliation, String host, CruiseControlApi apiClient, KafkaRebalance kafkaRebalance, KafkaRebalanceAnnotation rebalanceAnnotation) {
    Promise<MapAndStatus<ConfigMap, KafkaRebalanceStatus>> p = Promise.promise();
    if (rebalanceAnnotation == KafkaRebalanceAnnotation.none) {
        LOGGER.infoCr(reconciliation, "Starting Cruise Control rebalance user task status timer");
        String sessionId = kafkaRebalance.getStatus().getSessionId();
        AtomicInteger ccApiErrorCount = new AtomicInteger();
        vertx.setPeriodic(REBALANCE_POLLING_TIMER_MS, t -> {
            // Check that we have not already failed to contact the API beyond the allowed number of times.
            if (ccApiErrorCount.get() >= MAX_API_RETRIES) {
                vertx.cancelTimer(t);
                p.fail(new CruiseControlRestException("Unable to reach Cruise Control API after " + MAX_API_RETRIES + " attempts"));
            }
            kafkaRebalanceOperator.getAsync(kafkaRebalance.getMetadata().getNamespace(), kafkaRebalance.getMetadata().getName()).onSuccess(currentKafkaRebalance -> {
                // Checking that the resource was not deleted between periodic polls
                if (currentKafkaRebalance != null) {
                    // Safety check as timer might be called again (from a delayed timer firing)
                    if (state(currentKafkaRebalance) == KafkaRebalanceState.Rebalancing) {
                        if (rebalanceAnnotation(reconciliation, currentKafkaRebalance) == KafkaRebalanceAnnotation.stop) {
                            LOGGER.debugCr(reconciliation, "Stopping current Cruise Control rebalance user task");
                            vertx.cancelTimer(t);
                            apiClient.stopExecution(host, CruiseControl.REST_API_PORT).onSuccess(r -> p.complete(buildRebalanceStatus(null, KafkaRebalanceState.Stopped, validate(reconciliation, kafkaRebalance)))).onFailure(e -> {
                                LOGGER.errorCr(reconciliation, "Cruise Control stopping execution failed", e.getCause());
                                p.fail(e.getCause());
                            });
                        } else {
                            LOGGER.infoCr(reconciliation, "Getting Cruise Control rebalance user task status");
                            apiClient.getUserTaskStatus(host, CruiseControl.REST_API_PORT, sessionId).onSuccess(cruiseControlResponse -> {
                                JsonObject taskStatusJson = cruiseControlResponse.getJson();
                                CruiseControlUserTaskStatus taskStatus = CruiseControlUserTaskStatus.lookup(taskStatusJson.getString("Status"));
                                switch(taskStatus) {
                                    case COMPLETED:
                                        vertx.cancelTimer(t);
                                        LOGGER.infoCr(reconciliation, "Rebalance ({}) is now complete", sessionId);
                                        p.complete(buildRebalanceStatus(kafkaRebalance, null, KafkaRebalanceState.Ready, taskStatusJson, validate(reconciliation, kafkaRebalance)));
                                        break;
                                    case COMPLETED_WITH_ERROR:
                                        // TODO: There doesn't seem to be a way to retrieve the actual error message from the user tasks endpoint?
                                        // We may need to propose an upstream PR for this.
                                        // TODO: Once we can get the error details we need to add an error field to the Rebalance Status to hold
                                        // details of any issues while rebalancing.
                                        LOGGER.errorCr(reconciliation, "Rebalance ({}) optimization proposal has failed to complete", sessionId);
                                        vertx.cancelTimer(t);
                                        p.complete(buildRebalanceStatus(sessionId, KafkaRebalanceState.NotReady, validate(reconciliation, kafkaRebalance)));
                                        break;
                                    case // Rebalance is still in progress
                                    IN_EXECUTION:
                                        // the proposal is complete but the optimisation proposal summary will be missing.
                                        if (currentKafkaRebalance.getStatus().getOptimizationResult() == null || currentKafkaRebalance.getStatus().getOptimizationResult().isEmpty()) {
                                            LOGGER.infoCr(reconciliation, "Rebalance ({}) optimization proposal is now ready and has been added to the status", sessionId);
                                            // Cancel the timer so that the status is returned and updated.
                                            vertx.cancelTimer(t);
                                            p.complete(buildRebalanceStatus(kafkaRebalance, sessionId, KafkaRebalanceState.Rebalancing, taskStatusJson, validate(reconciliation, kafkaRebalance)));
                                        }
                                        ccApiErrorCount.set(0);
                                        // We can then update the status at this point.
                                        break;
                                    case // Rebalance proposal is still being calculated
                                    ACTIVE:
                                        // If a rebalance(dryrun=false) was called and the proposal is still being prepared then the task
                                        // will be in an ACTIVE state. When the proposal is ready it will shift to IN_EXECUTION and we will
                                        // check that the optimisation proposal is added to the status on the next reconcile.
                                        LOGGER.infoCr(reconciliation, "Rebalance ({}) optimization proposal is still being prepared", sessionId);
                                        ccApiErrorCount.set(0);
                                        break;
                                    default:
                                        LOGGER.errorCr(reconciliation, "Unexpected state {}", taskStatus);
                                        vertx.cancelTimer(t);
                                        p.fail("Unexpected state " + taskStatus);
                                        break;
                                }
                            }).onFailure(e -> {
                                LOGGER.errorCr(reconciliation, "Cruise Control getting rebalance task status failed", e.getCause());
                                // To make sure this error is not just a temporary problem with the network we retry several times.
                                // If the number of errors pass the MAX_API_ERRORS limit then the period method will fail the promise.
                                ccApiErrorCount.getAndIncrement();
                            });
                        }
                    } else {
                        p.complete(new MapAndStatus<>(null, currentKafkaRebalance.getStatus()));
                    }
                } else {
                    LOGGER.debugCr(reconciliation, "Rebalance resource was deleted, stopping the request time");
                    vertx.cancelTimer(t);
                    p.complete();
                }
            }).onFailure(e -> {
                LOGGER.errorCr(reconciliation, "Cruise Control getting rebalance resource failed", e.getCause());
                vertx.cancelTimer(t);
                p.fail(e.getCause());
            });
        });
    } else {
        p.complete(new MapAndStatus<>(null, kafkaRebalance.getStatus()));
    }
    return p.future();
}

Example 2 with CruiseControlUserTaskStatus

use of io.strimzi.operator.cluster.operator.resource.cruisecontrol.CruiseControlUserTaskStatus in project strimzi-kafka-operator by strimzi.

the class KafkaRebalanceAssemblyOperator method onRebalancing.

/**
 * This method handles the transition from {@code Rebalancing} state.
 * It starts a periodic timer in order to check the status of the ongoing rebalance processing on Cruise Control side.
 * In order to do that, it calls the related Cruise Control REST API about asking the user task status.
 * When the rebalance is finished, the next state is {@code Ready}.
 * If the user sets the strimzi.io/rebalance annotation to 'stop', it calls the Cruise Control REST API for stopping the ongoing task
 * and then transitions to the {@code Stopped} state.
 * If the user sets any other values for the strimzi.io/rebalance annotation, it is just ignored and the user task checks continue.
 * This method holds the lock until the rebalance is finished, the ongoing task is stopped or any exception is raised.
 *
 * @param reconciliation Reconciliation information
 * @param host Cruise Control service to which sending the REST API requests
 * @param apiClient Cruise Control REST API client instance
 * @param kafkaRebalance Current {@code KafkaRebalance} resource
 * @param rebalanceAnnotation The current value for the strimzi.io/rebalance annotation
 * @return a Future with the next {@code MapAndStatus<ConfigMap, KafkaRebalanceStatus>} including the state
 */
private Future<MapAndStatus<ConfigMap, KafkaRebalanceStatus>> onRebalancing(Reconciliation reconciliation, String host, CruiseControlApi apiClient, KafkaRebalance kafkaRebalance, KafkaRebalanceAnnotation rebalanceAnnotation) {
    Promise<MapAndStatus<ConfigMap, KafkaRebalanceStatus>> p = Promise.promise();
    if (rebalanceAnnotation == KafkaRebalanceAnnotation.none) {
        LOGGER.infoCr(reconciliation, "Starting Cruise Control rebalance user task status timer");
        String sessionId = kafkaRebalance.getStatus().getSessionId();
        AtomicInteger ccApiErrorCount = new AtomicInteger();
        vertx.setPeriodic(REBALANCE_POLLING_TIMER_MS, t -> {
            // Check that we have not already failed to contact the API beyond the allowed number of times.
            if (ccApiErrorCount.get() >= MAX_API_RETRIES) {
                vertx.cancelTimer(t);
                p.fail(new CruiseControlRestException("Unable to reach Cruise Control API after " + MAX_API_RETRIES + " attempts"));
            }
            kafkaRebalanceOperator.getAsync(kafkaRebalance.getMetadata().getNamespace(), kafkaRebalance.getMetadata().getName()).onSuccess(currentKafkaRebalance -> {
                // Checking that the resource was not deleted between periodic polls
                if (currentKafkaRebalance != null) {
                    // Safety check as timer might be called again (from a delayed timer firing)
                    if (state(currentKafkaRebalance) == KafkaRebalanceState.Rebalancing) {
                        if (rebalanceAnnotation(reconciliation, currentKafkaRebalance) == KafkaRebalanceAnnotation.stop) {
                            LOGGER.debugCr(reconciliation, "Stopping current Cruise Control rebalance user task");
                            vertx.cancelTimer(t);
                            apiClient.stopExecution(host, CruiseControl.REST_API_PORT).onSuccess(r -> p.complete(buildRebalanceStatus(null, KafkaRebalanceState.Stopped, validate(reconciliation, kafkaRebalance)))).onFailure(e -> {
                                LOGGER.errorCr(reconciliation, "Cruise Control stopping execution failed", e.getCause());
                                p.fail(e.getCause());
                            });
                        } else {
                            LOGGER.infoCr(reconciliation, "Getting Cruise Control rebalance user task status");
                            apiClient.getUserTaskStatus(host, CruiseControl.REST_API_PORT, sessionId).onSuccess(cruiseControlResponse -> {
                                JsonObject taskStatusJson = cruiseControlResponse.getJson();
                                CruiseControlUserTaskStatus taskStatus = CruiseControlUserTaskStatus.lookup(taskStatusJson.getString("Status"));
                                switch(taskStatus) {
                                    case COMPLETED:
                                        vertx.cancelTimer(t);
                                        LOGGER.infoCr(reconciliation, "Rebalance ({}) is now complete", sessionId);
                                        p.complete(buildRebalanceStatus(kafkaRebalance, null, KafkaRebalanceState.Ready, taskStatusJson, validate(reconciliation, kafkaRebalance)));
                                        break;
                                    case COMPLETED_WITH_ERROR:
                                        // TODO: There doesn't seem to be a way to retrieve the actual error message from the user tasks endpoint?
                                        // We may need to propose an upstream PR for this.
                                        // TODO: Once we can get the error details we need to add an error field to the Rebalance Status to hold
                                        // details of any issues while rebalancing.
                                        LOGGER.errorCr(reconciliation, "Rebalance ({}) optimization proposal has failed to complete", sessionId);
                                        vertx.cancelTimer(t);
                                        p.complete(buildRebalanceStatus(sessionId, KafkaRebalanceState.NotReady, validate(reconciliation, kafkaRebalance)));
                                        break;
                                    case // Rebalance is still in progress
                                    IN_EXECUTION:
                                        // the proposal is complete but the optimisation proposal summary will be missing.
                                        if (currentKafkaRebalance.getStatus().getOptimizationResult() == null || currentKafkaRebalance.getStatus().getOptimizationResult().isEmpty()) {
                                            LOGGER.infoCr(reconciliation, "Rebalance ({}) optimization proposal is now ready and has been added to the status", sessionId);
                                            // Cancel the timer so that the status is returned and updated.
                                            vertx.cancelTimer(t);
                                            p.complete(buildRebalanceStatus(kafkaRebalance, sessionId, KafkaRebalanceState.Rebalancing, taskStatusJson, validate(reconciliation, kafkaRebalance)));
                                        }
                                        ccApiErrorCount.set(0);
                                        // We can then update the status at this point.
                                        break;
                                    case // Rebalance proposal is still being calculated
                                    ACTIVE:
                                        // If a rebalance(dryrun=false) was called and the proposal is still being prepared then the task
                                        // will be in an ACTIVE state. When the proposal is ready it will shift to IN_EXECUTION and we will
                                        // check that the optimisation proposal is added to the status on the next reconcile.
                                        LOGGER.infoCr(reconciliation, "Rebalance ({}) optimization proposal is still being prepared", sessionId);
                                        ccApiErrorCount.set(0);
                                        break;
                                    default:
                                        LOGGER.errorCr(reconciliation, "Unexpected state {}", taskStatus);
                                        vertx.cancelTimer(t);
                                        p.fail("Unexpected state " + taskStatus);
                                        break;
                                }
                            }).onFailure(e -> {
                                LOGGER.errorCr(reconciliation, "Cruise Control getting rebalance task status failed", e.getCause());
                                // To make sure this error is not just a temporary problem with the network we retry several times.
                                // If the number of errors pass the MAX_API_ERRORS limit then the period method will fail the promise.
                                ccApiErrorCount.getAndIncrement();
                            });
                        }
                    } else {
                        p.complete(new MapAndStatus<>(null, currentKafkaRebalance.getStatus()));
                    }
                } else {
                    LOGGER.debugCr(reconciliation, "Rebalance resource was deleted, stopping the request time");
                    vertx.cancelTimer(t);
                    p.complete();
                }
            }).onFailure(e -> {
                LOGGER.errorCr(reconciliation, "Cruise Control getting rebalance resource failed", e.getCause());
                vertx.cancelTimer(t);
                p.fail(e.getCause());
            });
        });
    } else {
        p.complete(new MapAndStatus<>(null, kafkaRebalance.getStatus()));
    }
    return p.future();
}