[IOTDB-735] Concurrent error for MNode when creating time series automatically

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.10.0, 0.9.3
• Fix Version/s: 0.10.0
• Component/s: Core/Engine
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: Xiangdong Huang
• Created: 02/Jun/20 11:55
• Updated: 04/Jun/20 11:08
• Resolved: 02/Jun/20 13:25

Description

When multiply users write time series data concurrently and IoTDB enables auto_create_timeseries, sometimes NullpointException may be thrown from PlanExecutor.

 

The NullPointer exception is thrown from `LeafMNode measurementNode = (LeafMNode) node.getChild(measurement);`. 

 

// code placeholder
@Override
public void insert(InsertPlan insertPlan) throws QueryProcessException {
  MNode node = null;
  try {
    String[] measurementList = insertPlan.getMeasurements();
    String deviceId = insertPlan.getDeviceId();
    node = mManager.getDeviceNodeWithAutoCreateAndReadLock(deviceId);
    MeasurementSchema[] schemas = new MeasurementSchema[measurementList.length];

    for (int i = 0; i < measurementList.length; i++) {
      String measurement = measurementList[i];
      if (!node.hasChild(measurement)) {
        if (!IoTDBDescriptor.getInstance().getConfig().isAutoCreateSchemaEnabled()) {
          throw new PathNotExistException(deviceId + PATH_SEPARATOR + measurement);
        }
        TSDataType dataType = TypeInferenceUtils
            .getPredictedDataType(insertPlan.getValues()[i], insertPlan.isInferType());
        Path path = new Path(deviceId, measurement);
        internalCreateTimeseries(path.toString(), dataType);
      }
      LeafMNode measurementNode = (LeafMNode) node.getChild(measurement);
      schemas[i] = measurementNode.getSchema();
      // reset measurement to common name instead of alias
      measurementList[i] = measurementNode.getName();

      if(!insertPlan.isInferType()) {
        checkType(insertPlan, i, measurementNode.getSchema().getType());
      }
    }

    insertPlan.setMeasurements(measurementList);
    insertPlan.setSchemasAndTransferType(schemas);
    StorageEngine.getInstance().insert(insertPlan);
  } catch (StorageEngineException | MetadataException e) {
    throw new QueryProcessException(e);
  } finally {
    if (node != null) {
      ((InternalMNode) node).readUnlock();
    }
  }
}

The reason is, this method operates MNode instance directly without mtree.lock. Then, when `node.getChild()` is called, someone may call `node.addChild()` and the Concurrent error may lead the `node.getChild()` returns null.

 

So, be careful when you directly operate a MNode instance.

 

 

 

 

 

Issue Links

• links to: GitHub Pull Request #1306
• links to: GitHub Pull Request #1313