fix: harden refresh_schema, _stable_document_id, get_rel_map limit

- refresh_schema(): remove broad except Exception — transport/query errors
  now propagate instead of silently producing an empty relationships list
- _stable_document_id(): add default=str to json.dumps to handle non-JSON-
  serializable metadata values (datetime, UUID, Path, custom objects)
- get_rel_map(): coerce limit to int() before Cypher f-string interpolation
  to prevent injection via non-integer caller input

Author: polaz

langchain-coordinode/langchain_coordinode/graph.py

@@ -76,19 +76,16 @@ def refresh_schema(self) -> None:
         # the rel_props keys returned by _parse_schema().
         rel_types = list(structured.get("rel_props", {}).keys())
         if rel_types:
-            try:
-                rel_filter = "|".join(_cypher_ident(t) for t in rel_types)
-                rows = self._client.cypher(
-                    f"MATCH (a)-[r:{rel_filter}]->(b) "
-                    "RETURN DISTINCT a.__label__ AS src, r.__type__ AS rel, b.__label__ AS dst"
-                )
-                structured["relationships"] = [
-                    {"start": row["src"], "type": row["rel"], "end": row["dst"]}
-                    for row in rows
-                    if row.get("src") and row.get("rel") and row.get("dst")
-                ]
-            except Exception:  # noqa: BLE001
-                pass  # Graph may have no relationships yet; structured["relationships"] stays []
+            rel_filter = "|".join(_cypher_ident(t) for t in rel_types)
+            rows = self._client.cypher(
+                f"MATCH (a)-[r:{rel_filter}]->(b) "
+                "RETURN DISTINCT a.__label__ AS src, r.__type__ AS rel, b.__label__ AS dst"
+            )
+            structured["relationships"] = [
+                {"start": row["src"], "type": row["rel"], "end": row["dst"]}
+                for row in rows
+                if row.get("src") and row.get("rel") and row.get("dst")
+            ]
         self._structured_schema = structured
 
     def add_graph_documents(
@@ -222,12 +219,15 @@ def _stable_document_id(source: Any) -> str:
     content = getattr(source, "page_content", "") or ""
     metadata = getattr(source, "metadata", {}) or {}
     # Use canonical JSON encoding to avoid delimiter ambiguity and ensure
-    # determinism for nested/non-scalar metadata values.
+    # determinism for nested/non-scalar metadata values.  default=str converts
+    # non-JSON-serializable types (datetime, UUID, Path, …) to their string
+    # representation so the hash never raises TypeError.
     canonical = json.dumps(
         {"content": content, "metadata": metadata},
         sort_keys=True,
         separators=(",", ":"),
         ensure_ascii=False,
+        default=str,
     )
     return hashlib.sha256(canonical.encode()).hexdigest()[:32]
 

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

@@ -203,13 +203,14 @@ def get_rel_map(
 
         rel_filter = "|".join(_cypher_ident(t) for t in active_types)
         node_ids = [n.id for n in graph_nodes]
+        safe_limit = int(limit)  # coerce to int to prevent Cypher injection via non-integer input
         params: dict[str, object] = {"ids": node_ids}
 
         cypher = (
             f"MATCH (n)-[r:{rel_filter}]->(m) "
             f"WHERE n.id IN $ids "
             f"RETURN n, r.__type__ AS _rel_type, m, n.id AS _src_id, m.id AS _dst_id "
-            f"LIMIT {limit}"
+            f"LIMIT {safe_limit}"
         )
         result = self._client.cypher(cypher, params=params)