Load balancing in Causal Cluster consisting of only Cores

[volodymyr_metly]

Hello,

we have a simple cluster consisting of only 3 Cores.
On intensive WRITE operations, we observe significant CPU load on only one of the nodes, which is Leader. Which is OK I guess since only a Leader receives writes and communicates them to the others.
However, on intensive READ operations, the behaviour is similar - only Leader is loaded.
So my question is (I could not find a clear explanation of it in the docs) - should READ operations be load-balanced between all members if a cluster consists of only COREs, i.e. does not contain any READ_REPLICAs?

We tried to enable this in the configuration, but to no avail:

causal_clustering.load_balancing.plugin=server_policies
causal_clustering.load_balancing.config.server_policies.default=all()

Thanks!

[dana_canzano]

What version of Neo4j?
When connecting to Neo4j you can specify the URL to connect as either

when using Neo4j 3.5 either bolt://<Neo4j IP Address>:7687 or bolt+routing://<Neo4j IP Address>:7687

when using Neo4j 4.0 either bolt://<Neo4j IP Address>:7687 or neo4j://<Neo4j IP Address>

and to which if connecting as bolt://<Neo4j IP Address>:7687 then all transactions will be constrained to the cluster member defined at . However if you connect at either bolt+routing://<Neo4j IP Address>:7687 or `neo4j://:7687, for the respective version, then the connection is now cluster aware and as such if you send a session.readTransaction it should go to non Leader members and a session.writeTransaction then it will only go to the Leader.

So... what method are you using to connect and then are you utilizing session.readTransaction and session.writeTransactions where applicable

[volodymyr_metly]

Thank you for the response.

We are using 4.0 version and connect using neo4j:// url.
For the querying method, we use the OGM's Session:

sessionFactory.openSession().query(DataType.class, cypher, parameters)

[volodymyr_metly]

So I've wrapped the code into explicit transaction of type READ-ONLY, and looks like it helped - a few nodes now are being loaded instead of only one, and the load is smaller.
@dana.canzano Thanks a lot for your help!