RowStream.next() returns fewer results than expected when iterating multiple rows

[JPC-007]

Description

When using result.next().await to iterate over query results that return multiple rows, the iteration terminates prematurely, returning fewer results than expected. Single-row queries work correctly.

Environment

• neo4rs version: 0.9.0-rc.9
• Neo4j version: 5.x (running in Kubernetes)
• Rust version: 1.81+
• OS: Linux (container)

Steps to Reproduce

1. Create multiple FOLLOWS relationships in Neo4j:

CREATE (a:User {id: "user-1"})-[:FOLLOWS]->(b:User {id: "user-2"})
CREATE (a:User {id: "user-1"})-[:FOLLOWS]->(c:User {id: "user-3"})
CREATE (a:User {id: "user-1"})-[:FOLLOWS]->(d:User {id: "user-4"})
CREATE (a:User {id: "user-1"})-[:FOLLOWS]->(e:User {id: "user-5"})
CREATE (a:User {id: "user-1"})-[:FOLLOWS]->(f:User {id: "user-6"})

2. Query using neo4rs:

let cypher = r#"
    MATCH (user:User {id: $user_id})-[:FOLLOWS]->(following:User)
    RETURN following.id AS following_id
    ORDER BY following.id
    SKIP $offset
    LIMIT $limit
"#;

let mut result = graph
    .execute(
        query(cypher)
            .param("user_id", "user-1")
            .param("offset", 0i64)
            .param("limit", 50i64),
    )
    .await?;

let mut following = Vec::new();
while let Some(row) = result.next().await? {
    if let Ok(id_str) = row.get::<String>("following_id") {
        following.push(id_str);
    }
}

println!("Expected: 5, Got: {}", following.len());

Expected Behavior

The loop should iterate 5 times and return all 5 following IDs.

Actual Behavior

The loop only iterates 1 time and returns only 1 result, even though:

• The same query in cypher-shell returns all 5 results
• A separate COUNT query returns total_count = 5
• has_more is calculated as true (indicating more data exists)

Additional Context

What works correctly:

• Single-row queries (e.g., RETURN count(r) > 0 AS is_following) work fine
• The same data queried via PostgreSQL returns all 5 results

Possibly related:

• Issue Neo4j Server throwing IllegalState exception during PULL #236 describes a similar Bolt protocol state machine issue where PULL messages fail because the session transitions to READY state prematurely
• This might be related to the RowStream buffer/fetch logic or Bolt protocol state synchronization

Workaround

We've switched to using PostgreSQL for list queries while keeping Neo4j for single-value lookups and graph traversals.

Code Reference

Our repository pattern:

pub struct GraphRepository {
    graph: Arc<Graph>,
}

impl GraphRepository {
    pub async fn new(uri: &str, user: &str, password: &str) -> Result<Self> {
        let graph = Graph::new(uri, user, password)?;
        Ok(Self { graph: Arc::new(graph) })
    }
}

The Graph is wrapped in Arc and shared across async tasks. Could this be related to connection pool state issues?

[knutwalker]

I had no luck reproducing the issue, I tried several versions between 5.x and 2025.x, directly or via docker, but I always get the expected 5 results.
Is this 100% reproducible on your end? Have you tried it without kubernetes in between?
As a side note, the Graph type can be cloned directly and there are several different Arc types internally, no need to wrap it in an Arc on your end as well, though it doesn't hurt much either.