When to use labels and relation types indexing from Neo4j 4.3?

[marco_brandizi]

Hi all,

In NODES 2021, label and type indices, available from the 4.3 server, were presented. However, I haven't clear when I should or should not use this kind of indices.

Naively, I tend to think that in most cases, there will be searches like MATCH ( n: ) - [r:] -> (...) so, I'd say it's good to have them in most cases.

But maybe I'm wrong? Maybe they benefit more certain types of queries only?

Thanks in advance for your answer.

[dana_canzano]

Regarding relationship type index see The use of indexes - Neo4j Cypher Manual

and to which if you have properties defined in the relationship itself and provided said index exists then running

match (n)-[r:FOLLOWS]->() where r.<property>=<value> return n;

then the index will be used. Where might this be helpful? If for example a node has 20000 :FOLLOWS relationships and each relationship is has a property named status and for example most relationships (i.e. > 95%) have a status of active. If I run

match (n)-[r:FOLLOWS]->() where r.status='inactive' return n;

then the index will be used and rather that interating over all 20k relationships the index will allow us to only iterate over the < 5% of :FOLLOWS relationships for the given node

[marco_brandizi]

Many thanks, @dana.canzano, now it's clearer.

Regarding the label/relation-type level indices, I infer that they can be useful when expressions involving entity types appear on the WHERE clause and select a subset of entities.

[dana_canzano]

@marco.brandizi correct. Without a WHERE clause and restriction on said indexed property the index will not be utiilzied.

A simple

match (n)-[r:FOLLOWS]->()  return n;

would not use the index (provided it existed)