neo4j Creating many relationships with individual statements much slower than a loop

Question

I have 2 queries in neo4j

MERGE (f:Foo {id:1}) 
MERGE (b:Bar {id:1}) 
with f,b 
FOREACH (i IN range(0,999) | create (f)-[:Qux]->(b));

Added 2 labels, created 2 nodes, set 2 properties, created 1000 relationships, statement executed in 39 ms.

and

MERGE (f:Foo {id:1}) 
MERGE (b:Bar {id:1}) 
with f,b 
CREATE (f) - [:Qux] -> (b)
CREATE (f) - [:Qux] -> (b)
... (1000 times)
CREATE (f) - [:Qux] -> (b)

Added 2 labels, created 2 nodes, set 2 properties, created 1000 relationships, statement executed in 8301 ms.

How come it is so much slower to create relationships with statements than with a loop?

I want to be able to create many relationships at once but a loop isn't appropriate as I want to include data on each individual relationship. How could I do this fast without using a loop. I assume something about my syntax is making a wrong assumption.

NOTE: i cleared the graph before running each of these

Tore Eschliman Tore Eschliman · Accepted Answer · 2016-11-02T22:04:54

You absolutely want to use an iterative loop still, because then the DB only has to parse a single CREATE statement and re-use it (iteration is native to Java, so it's easy to translate a request to iterate to fast queries). If you need to assign properties to each relationship, unwind the list of properties instead of the RANGE().

UNWIND {r_props} AS r_prop
CREATE (f) - [r:Qux] -> (b)
SET r = r_prop

Or if you need to calculate properties per-relationship based on a query, unwind the RANGE and you can do the full set of Cypher operations (MATCH, etc.) per relationship:

UNWIND range(0, 999) AS i
CREATE (f) - [r:Qux] -> (b)
MATCH (n:OtherNode {id: i})
SET r = PROPERTIES(n)

Using FOREACH is convenient but can be limiting. You'll still want an iterative loop either way, though, because it's always going to be quicker for neo4j to interpret.

neo4j Creating many relationships with individual statements much slower than a loop

1 Answers