Examples with StateTransitionError org.apache.helix.participant.statemachine.StateTransitionError used on opensource projects

Example 1 with StateTransitionError

use of org.apache.helix.participant.statemachine.StateTransitionError in project helix by apache.

the class HelixStateTransitionHandler method preHandleMessage.

void preHandleMessage() throws Exception {
    if (!_message.isValid()) {
        String errorMessage = "Invalid Message, ensure that message: " + _message + " has all the required fields: " + Arrays.toString(Message.Attributes.values());
        _statusUpdateUtil.logError(_message, HelixStateTransitionHandler.class, errorMessage, _manager);
        logger.error(errorMessage);
        throw new HelixException(errorMessage);
    }
    logger.info("handling message: " + _message.getMsgId() + " transit " + _message.getResourceName() + "." + _message.getPartitionName() + "|" + _message.getPartitionNames() + " from:" + _message.getFromState() + " to:" + _message.getToState() + ", relayedFrom: " + _message.getRelaySrcHost());
    HelixDataAccessor accessor = _manager.getHelixDataAccessor();
    String partitionName = _message.getPartitionName();
    String fromState = _message.getFromState();
    // Verify the fromState and current state of the stateModel
    String state = _currentStateDelta.getState(partitionName);
    // Set start time right before invoke client logic
    _currentStateDelta.setStartTime(_message.getPartitionName(), System.currentTimeMillis());
    if (fromState != null && !fromState.equals("*") && !fromState.equalsIgnoreCase(state)) {
        String errorMessage = "Current state of stateModel does not match the fromState in Message" + ", Current State:" + state + ", message expected:" + fromState + ", partition: " + partitionName + ", from: " + _message.getMsgSrc() + ", to: " + _message.getTgtName();
        _statusUpdateUtil.logError(_message, HelixStateTransitionHandler.class, errorMessage, _manager);
        logger.error(errorMessage);
        throw new HelixStateMismatchException(errorMessage);
    }
    // Reset the REQUESTED_STATE property if it exists.
    try {
        String instance = _manager.getInstanceName();
        String sessionId = _message.getTgtSessionId();
        String resource = _message.getResourceName();
        ZNRecordBucketizer bucketizer = new ZNRecordBucketizer(_message.getBucketSize());
        PropertyKey key = accessor.keyBuilder().currentState(instance, sessionId, resource, bucketizer.getBucketName(partitionName));
        ZNRecord rec = new ZNRecord(resource);
        Map<String, String> map = new TreeMap<String, String>();
        map.put(CurrentState.CurrentStateProperty.REQUESTED_STATE.name(), null);
        rec.getMapFields().put(partitionName, map);
        ZNRecordDelta delta = new ZNRecordDelta(rec, ZNRecordDelta.MergeOperation.SUBTRACT);
        List<ZNRecordDelta> deltaList = new ArrayList<ZNRecordDelta>();
        deltaList.add(delta);
        CurrentState currStateUpdate = new CurrentState(resource);
        currStateUpdate.setDeltaList(deltaList);
        // Update the ZK current state of the node
        if (!accessor.updateProperty(key, currStateUpdate)) {
            logger.error("Fails to persist current state back to ZK for resource " + resource + " partition: " + partitionName);
        }
    } catch (Exception e) {
        logger.error("Error when removing " + CurrentState.CurrentStateProperty.REQUESTED_STATE.name() + " from current state.", e);
        StateTransitionError error = new StateTransitionError(ErrorType.FRAMEWORK, ErrorCode.ERROR, e);
        _stateModel.rollbackOnError(_message, _notificationContext, error);
        _statusUpdateUtil.logError(_message, HelixStateTransitionHandler.class, e, "Error when removing " + CurrentState.CurrentStateProperty.REQUESTED_STATE.name() + " from current state.", _manager);
    }
}

Example 2 with StateTransitionError

use of org.apache.helix.participant.statemachine.StateTransitionError in project helix by apache.

the class HelixStateTransitionHandler method onError.

@Override
public void onError(Exception e, ErrorCode code, ErrorType type) {
    HelixDataAccessor accessor = _manager.getHelixDataAccessor();
    Builder keyBuilder = accessor.keyBuilder();
    String instanceName = _manager.getInstanceName();
    String resourceName = _message.getResourceName();
    String partition = _message.getPartitionName();
    // All internal error has been processed already, so we can skip them
    if (type == ErrorType.INTERNAL) {
        logger.error("Skip internal error. errCode: " + code + ", errMsg: " + e.getMessage());
        return;
    }
    try {
        // if the transition is not canceled, it should go into error state
        if (code == ErrorCode.ERROR) {
            CurrentState currentStateDelta = new CurrentState(resourceName);
            currentStateDelta.setState(partition, HelixDefinedState.ERROR.toString());
            _stateModel.updateState(HelixDefinedState.ERROR.toString());
            // if transit from ERROR state, disable the partition
            if (_message.getFromState().equalsIgnoreCase(HelixDefinedState.ERROR.toString())) {
                disablePartition();
            }
            if (!accessor.updateProperty(keyBuilder.currentState(instanceName, _message.getTgtSessionId(), resourceName), currentStateDelta)) {
                logger.error("Fails to persist ERROR current state to ZK for resource " + resourceName + " partition: " + partition);
            }
        }
    } finally {
        StateTransitionError error = new StateTransitionError(type, code, e);
        _stateModel.rollbackOnError(_message, _notificationContext, error);
    }
}

Example 3 with StateTransitionError

use of org.apache.helix.participant.statemachine.StateTransitionError in project helix by apache.

the class HelixStateTransitionHandler method postHandleMessage.

void postHandleMessage() {
    HelixTaskResult taskResult = (HelixTaskResult) _notificationContext.get(MapKey.HELIX_TASK_RESULT.toString());
    Exception exception = taskResult.getException();
    String partitionKey = _message.getPartitionName();
    String resource = _message.getResourceName();
    String sessionId = _message.getTgtSessionId();
    String instanceName = _manager.getInstanceName();
    HelixDataAccessor accessor = _manager.getHelixDataAccessor();
    Builder keyBuilder = accessor.keyBuilder();
    int bucketSize = _message.getBucketSize();
    ZNRecordBucketizer bucketizer = new ZNRecordBucketizer(bucketSize);
    // for zk current state it is OK as we have the per-session current state node
    if (!_message.getTgtSessionId().equals(_manager.getSessionId())) {
        logger.warn("Session id has changed. Skip postExecutionMessage. Old session " + _message.getExecutionSessionId() + " , new session : " + _manager.getSessionId());
        return;
    }
    // Set the INFO property and mark the end time, previous state of the state transition
    _currentStateDelta.setInfo(partitionKey, taskResult.getInfo());
    _currentStateDelta.setEndTime(partitionKey, taskResult.getCompleteTime());
    _currentStateDelta.setPreviousState(partitionKey, _message.getFromState());
    // add host name this state transition is triggered by.
    if (Message.MessageType.RELAYED_MESSAGE.name().equals(_message.getMsgSubType())) {
        _currentStateDelta.setTriggerHost(partitionKey, _message.getRelaySrcHost());
    } else {
        _currentStateDelta.setTriggerHost(partitionKey, _message.getMsgSrc());
    }
    if (taskResult.isSuccess()) {
        // String fromState = message.getFromState();
        String toState = _message.getToState();
        _currentStateDelta.setState(partitionKey, toState);
        if (toState.equalsIgnoreCase(HelixDefinedState.DROPPED.toString())) {
            // for "OnOfflineToDROPPED" message, we need to remove the resource key record
            // from the current state of the instance because the resource key is dropped.
            // In the state model it will be stayed as "OFFLINE", which is OK.
            ZNRecord rec = new ZNRecord(_currentStateDelta.getId());
            rec.getMapFields().put(partitionKey, null);
            ZNRecordDelta delta = new ZNRecordDelta(rec, MergeOperation.SUBTRACT);
            List<ZNRecordDelta> deltaList = new ArrayList<ZNRecordDelta>();
            deltaList.add(delta);
            _currentStateDelta.setDeltaList(deltaList);
            _stateModelFactory.removeStateModel(resource, partitionKey);
        } else {
            // if the partition is not to be dropped, update _stateModel to the TO_STATE
            _stateModel.updateState(toState);
        }
    } else if (taskResult.isCancelled()) {
        // Cancelled message does not need current state update
        return;
    } else {
        if (exception instanceof HelixStateMismatchException) {
            // if fromState mismatch, set current state on zk to stateModel's current state
            logger.warn("Force CurrentState on Zk to be stateModel's CurrentState. partitionKey: " + partitionKey + ", currentState: " + _stateModel.getCurrentState() + ", message: " + _message);
            _currentStateDelta.setState(partitionKey, _stateModel.getCurrentState());
        } else {
            StateTransitionError error = new StateTransitionError(ErrorType.INTERNAL, ErrorCode.ERROR, exception);
            if (exception instanceof InterruptedException) {
                if (_isTimeout) {
                    error = new StateTransitionError(ErrorType.INTERNAL, ErrorCode.TIMEOUT, exception);
                } else {
                    // State transition interrupted but not caused by timeout. Keep the current
                    // state in this case
                    logger.error("State transition interrupted but not timeout. Not updating state. Partition : " + _message.getPartitionName() + " MsgId : " + _message.getMsgId());
                    return;
                }
            }
            _stateModel.rollbackOnError(_message, _notificationContext, error);
            _currentStateDelta.setState(partitionKey, HelixDefinedState.ERROR.toString());
            _stateModel.updateState(HelixDefinedState.ERROR.toString());
            // if we have errors transit from ERROR state, disable the partition
            if (_message.getFromState().equalsIgnoreCase(HelixDefinedState.ERROR.toString())) {
                disablePartition();
            }
        }
    }
    try {
        // Update the ZK current state of the node
        PropertyKey key = keyBuilder.currentState(instanceName, sessionId, resource, bucketizer.getBucketName(partitionKey));
        if (_message.getAttribute(Attributes.PARENT_MSG_ID) == null) {
            // normal message
            if (!accessor.updateProperty(key, _currentStateDelta)) {
                throw new HelixException("Fails to persist current state back to ZK for resource " + resource + " partition: " + _message.getPartitionName());
            }
        } else {
            // sub-message of a batch message
            ConcurrentHashMap<String, CurrentStateUpdate> csUpdateMap = (ConcurrentHashMap<String, CurrentStateUpdate>) _notificationContext.get(MapKey.CURRENT_STATE_UPDATE.toString());
            csUpdateMap.put(partitionKey, new CurrentStateUpdate(key, _currentStateDelta));
        }
    } catch (Exception e) {
        logger.error("Error when updating current-state ", e);
        StateTransitionError error = new StateTransitionError(ErrorType.FRAMEWORK, ErrorCode.ERROR, e);
        _stateModel.rollbackOnError(_message, _notificationContext, error);
        _statusUpdateUtil.logError(_message, HelixStateTransitionHandler.class, e, "Error when update current-state ", _manager);
    }
}