identities

[matt-beanland]

Extend Ash Neo4j to support identities, related to upsert nodes

[matt-beanland]

Description

Ash identities declare uniqueness constraints — a natural mapping to Neo4j's CREATE CONSTRAINT ... IS UNIQUE for node properties. AshNeo4j currently has no support for this: identities declared on a resource are silently ignored at the data-layer level, so there is no Cypher-level enforcement and no MERGE-based upsert path.

The workaround is pre_check? true, which adds an Elixir-side read before every write to simulate uniqueness. This has two problems: it still has a race window under concurrent writes, and it adds a round-trip to every create.

The concrete case in Diffo that makes this important now: AssignmentRelationship is a node resource representing a pool assignment. It needs a compound unique constraint on [:source_id, :pool, :thing, :value] — meaning for a given source instance and pool, the same integer value can only be assigned once. Without database-level enforcement, two concurrent auto-assign operations can select and write the same value, producing duplicate assignment nodes with no error.

What we need

When a resource declares an identity, AshNeo4j should create a corresponding Neo4j uniqueness constraint on the node's label and listed properties. On create, if the constraint is violated, AshNeo4j should surface this as an Ash identity conflict error rather than a raw driver exception.

Why it matters

Pool assignment is a write-contended operation — multiple processes can assign from the same pool concurrently. Without a database-enforced unique constraint on the assignment node, the only protection is a pre-check read in Elixir, which leaves a race window and adds latency. A Neo4j constraint closes the window atomically and eliminates the extra round-trip.

[matt-beanland]

I think this one is going to require custom indexes, which then gets us down the ash database migration rabbit hole. Think we were heading down it with vectors soon anyway....