More than one element Exception when using R-Tree index to insert WKT

[kanghq]

When I using spatial.addWKT procedure to add WKT objects using multi threads in HA 4 cluster environment, the error occurred.

Neo4j 3.1.4 Enterprise HA 4 cluster RHEL7 Neo4j Spatial Plugin

geomencoder_config wkt ctime 1495172338674 geomencoder org.neo4j.gis.spatial.WKTGeometryEncoder layer_class org.neo4j.gis.spatial.EditableLayerImpl layer geom

org.neo4j.driver.v1.exceptions.ClientException: Failed to invoke procedure spatial.addWKT: Caused by: java.util.NoSuchElementException: More than one element in org.neo4j.cypher.internal.javacompat.ExecutionResult$ExceptionConversion@196a0242. First element is 'Node[179312]' and the second element is 'Node[208620]' at org.neo4j.driver.internal.connector.socket.SocketResponseHandler.handleFailureMessage(SocketResponseHandler.java:68) at org.neo4j.driver.internal.messaging.PackStreamMessageFormatV1$Reader.unpackFailureMessage(PackStreamMessageFormatV1.java:464) at org.neo4j.driver.internal.messaging.PackStreamMessageFormatV1$Reader.read(PackStreamMessageFormatV1.java:425) at org.neo4j.driver.internal.connector.socket.SocketClient.receiveOne(SocketClient.java:130) at org.neo4j.driver.internal.connector.socket.SocketConnection.receiveOne(SocketConnection.java:143) at org.neo4j.driver.internal.connector.ConcurrencyGuardingConnection.receiveOne(ConcurrencyGuardingConnection.java:164) at org.neo4j.driver.internal.pool.PooledConnection.receiveOne(PooledConnection.java:187) at org.neo4j.driver.internal.InternalStatementResult.consume(InternalStatementResult.java:259) at org.neo4j.jdbc.bolt.BoltPreparedStatement.executeUpdate(BoltPreparedStatement.java:65) at org.neo4j.jdbc.bolt.BoltPreparedStatement.execute(BoltPreparedStatement.java:75) at com.hqkang.SparkApp.core.DBHelper$2.call(DBHelper.java:161) at com.hqkang.SparkApp.core.DBHelper$2.call(DBHelper.java:1) at org.apache.spark.api.java.JavaRDDLike$$anonfun$foreachPartition$1.apply(JavaRDDLike.scala:219) at org.apache.spark.api.java.JavaRDDLike$$anonfun$foreachPartition$1.apply(JavaRDDLike.scala:219) at org.apache.spark.rdd.RDD$$anonfun$foreachPartition$1$$anonfun$apply$29.apply(RDD.scala:925) at org.apache.spark.rdd.RDD$$anonfun$foreachPartition$1$$anonfun$apply$29.apply(RDD.scala:925) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944) at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1944) at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87) at org.apache.spark.scheduler.Task.run(Task.scala:99) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:282) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745)

[craigtaverner]

Is it possible to get the server logs and see if there are any server side errors correlated to this client side error?

[kanghq]

It looks like related to AutoIndexing?

Similar issue https://stackoverflow.com/questions/20797956/neo4j-cypher-query-returning-wrong-nodes

[craigtaverner]

I don't believe there is any autoindexing in the spatial library. Also the issue you linked to above is using the embedded API to access an auto-index with the Cypher 'START' clause, while your test is using Bolt to remotely access the spatial procedure layer.addWKT. The 'index' used there is not really an index, but a manually maintained graph structure. The nodes need to be manually added and manually removed from the structure using the layer.add methods. I still think we will only be able to figure out what is going on with server side logs, or better yet, a replicated test environment. Could you provide and sample code for reproducing this error?

[kanghq]

Hi, I figured out that error. I forgot to add the constraint CREATE CONSTRAINT ON (n:ReferenceNode) ASSERT n.name IS UNIQUE

When calling layer.addWKT, it calls getSpatialRoot function. In high concurrency scenario, there are two spatial_root nodes created. that's the error reason. I simply added that constraint and changed getReferenceNode to

public static Node getReferenceNode(GraphDatabaseService db, String name) {

        if (db != dbRef) {
            ReferenceNodes.dbRef = db;
        }
        Node res = null;
        Transaction tx = db.beginTx();
        try {

            Result result = db.execute("MERGE (ref:ReferenceNode {name:{name}}) RETURN ref", map("name", name));
            res = Iterators.single(result.<Node>columnAs("ref"));
            tx.success();

        } finally {
            tx.close();
        }
        return res;
    }

This bug has resolved. However I could not find a good way to add this constrain automatically.

[craigtaverner]

That was a very interesting analysis. Neo4j Spatial was not originally written for high concurrency scenarios, but it is great if it can slowly be made more concurrency safe. I am curious though if the changes to getReferenceNode are actually required? What happened when you had the uniqueness constraint, but not the extra transaction in that code?

The db.execute will create a transaction internally, and if a node is created it will be committed before the db.execute returns. With your change, the transaction is now created externally in your code, and the node is only created (transaction committed) when you call tx.close. The difference seems very subtle and I am not sure I see why it would make it more concurrency safe. Do you know?