fix: duplicate edges on re-ingestion — replace CREATE with MERGE in upsert_relations() and add_graph_documents()

[polaz]

Problem

Re-ingesting the same data creates duplicate edges. Every call to upsert_relations() (LlamaIndex) or add_graph_documents() (LangChain) creates new edges unconditionally:

# First ingest
store.upsert_relations([Relation(label="KNOWS", source_id="alice", target_id="bob")])
# Second ingest — creates a DUPLICATE edge
store.upsert_relations([Relation(label="KNOWS", source_id="alice", target_id="bob")])

This makes repeated indexing non-idempotent and bloats the graph.

Root cause

CoordiNode Cypher does not support MERGE for relationship patterns — only for node patterns (NodeScan). The adapters work around this with CREATE, accepting duplicate edges as a known limitation.

Tracked as G072 in CoordiNode DB repository: MERGE (src)-[r:TYPE]->(dst) fails with "MERGE create from non-NodeScan pattern".

SDK changes (after G072 is fixed in DB)

Once G072 is resolved:

llama-index-coordinode/llama_index/graph_stores/coordinode/base.py

• upsert_relations(): replace CREATE (src)-[r:{label}]->(dst) → MERGE (src)-[r:{label}]->(dst)
• Remove the comment block explaining the CREATE fallback
• Remove the SET r += $props workaround comment

langchain-coordinode/langchain_coordinode/graph.py

• _create_edge(): replace CREATE (src)-[r:{rel_type}]->(dst) → MERGE (src)-[r:{rel_type}]->(dst)
• Rename _create_edge() → _upsert_edge() (or keep name, update docstring)
• _link_document_to_entities(): replace CREATE (d)-[:MENTIONS]->(n) → MERGE (d)-[:MENTIONS]->(n)
• add_graph_documents() docstring: remove duplicate-edge warning

Tests

• test_add_graph_documents_idempotent: change assertion from cnt >= 1 to cnt == 1
• Add similar idempotency test for LlamaIndex upsert_relations()

Acceptance criteria

• Re-ingesting the same document creates exactly 1 edge (not N)
• upsert_relations() is idempotent
• add_graph_documents() is idempotent for both nodes and edges
• All existing tests pass

Gate

Blocked by G072 — requires CoordiNode DB fix: MERGE clause for relationship patterns in Cypher executor.
Do not start SDK implementation until a CoordiNode release with G072 fixed is available.

[coderabbitai]

🔗 Related PRs

#15 - fix: CoordiNode Cypher compatibility — add_graph_documents, type/label, MATCH+CREATE [closed]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[polaz]

Implemented in commit 6d009a7 — CoordiNode 0.3.2 (0.4.0 SDK dep) resolved G072, so both adapters now use MERGE for relationships. Test added in 3316a69: test_upsert_relations_idempotent verifies exactly 1 edge after double upsert.