Incorrect results triggered by query partitioning

[Lincyaw]

Version: Docker image falkordb/falkordb:edge OS: Ubuntu 22.04 API/Driver: Cypher

According to the following query, the 18 should equals to 10+12

lab:6380> GRAPH.QUERY 0 "MATCH (n1:L3)-[]-(n2:L2)-[]-(n3:L1:L2)-[r1:T1]-(n4:L3:L0:L4) RETURN COUNT(*)"
1) 1) "COUNT(*)"
2) 1) 1) (integer) 18
3) 1) "Cached execution: 0"
   2) "Query internal execution time: 1.820675 milliseconds"
lab:6380> GRAPH.QUERY 0 "MATCH (n1:L3)<-[]-(n2:L2)-[]-(n3:L1:L2)-[r1:T1]-(n4:L3:L0:L4) RETURN COUNT(*)"
1) 1) "COUNT(*)"
2) 1) 1) (integer) 10
3) 1) "Cached execution: 0"
   2) "Query internal execution time: 0.738668 milliseconds"
lab:6380> GRAPH.QUERY 0 "MATCH (n1:L3)-[]->(n2:L2)-[]-(n3:L1:L2)-[r1:T1]-(n4:L3:L0:L4) RETURN COUNT(*)"
1) 1) "COUNT(*)"
2) 1) 1) (integer) 12
3) 1) "Cached execution: 0"
   2) "Query internal execution time: 0.759969 milliseconds"
lab:6380> GRAPH.QUERY 0 "MATCH (n)-[r]-(n) RETURN n,r LIMIT 10"
1) 1) "n"
   2) "r"
2) (empty array)
3) 1) "Cached execution: 0"
   2) "Query internal execution time: 0.584519 milliseconds"

The create statement is attached below for your reproduction 1692914926.9295375.log

[gkorland]

@Lincyaw you're right the semantic is not well defined in Cypher, counting different overlapping paths on the first example is considered as a single result for the count.

So in your example, two directed paths that have the same components will be count as a single path.

You can overcome this by counting unique paths.

GRAPH.QUERY 0 "MATCH p=(n1:L3)-[]-(n2:L2)-[]-(n3:L1:L2)-[r1:T1]-(n4:L3:L0:L4) RETURN COUNT(p)"

[Lincyaw]

You are right, but for the same queries, the neo4j's results seems different from the FalkerDB. Is that normal?

[gkorland]

Yes, the semantic is not well defined in Cypher so you might find different vendors interpenetration

[Lincyaw]

After searching the Cypher Specification, I found the following use cases.

https://github.com/opencypher/openCypher/blob/master/tck/features/clauses/match/Match8.feature

And in the following test case

 Scenario: [2] Counting rows after MATCH, MERGE, OPTIONAL MATCH
    Given an empty graph
    And having executed:
      """
      CREATE (a:A), (b:B)
      CREATE (a)-[:T1]->(b),
             (b)-[:T2]->(a)
      """
    When executing query:
      """
      MATCH (a)
      MERGE (b)
      WITH *
      OPTIONAL MATCH (a)--(b)
      RETURN count(*)
      """
    Then the result should be, in any order:
      | count(*) |
      | 6        |
    And no side effects

The behavior of Neo4j is exactly the same as the expected result, but in falkerDB, the result is different.