iris-vector-graph 1.63.4__tar.gz → 1.63.6__tar.gz

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: iris-vector-graph
-Version: 1.63.4
+Version: 1.63.6
 Summary: Transactional Graph + Vector retrieval system for InterSystems IRIS with hybrid search, openCypher, and GraphQL APIs
 Project-URL: Homepage, https://github.com/intersystems-community/iris-vector-graph
 Project-URL: Documentation, https://github.com/intersystems-community/iris-vector-graph/tree/main/docs
@@ -692,6 +692,10 @@ anchors = engine.get_kg_anchors(icd_codes=["J18.0", "E11.9"])
 
 ## Changelog
 
+### v1.63.5 (2026-04-26)
+- fix: `NKGAccel.BFSJson` per-seed adjacency export — `ExportAdjacencyFromSeed()` exports only the subgraph reachable from the seed node (not the full 299K-edge graph). Fixes `<MAXSTRING>` on Mindwalk-scale graphs, enabling Arno-accelerated multi-hop BFS. Adjacency string now scales with BFS result size (~10KB per seed instead of >3.5MB full graph). Handles outbound + inbound edges for undirected BFS.
+
+
 ### v1.63.4 (2026-04-26)
 - chore: merge 080-engine-status to main; NKGAccel.cls added to iris_src from arno upstream
 

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/README.md

@@ -620,6 +620,10 @@ anchors = engine.get_kg_anchors(icd_codes=["J18.0", "E11.9"])
 
 ## Changelog
 
+### v1.63.5 (2026-04-26)
+- fix: `NKGAccel.BFSJson` per-seed adjacency export — `ExportAdjacencyFromSeed()` exports only the subgraph reachable from the seed node (not the full 299K-edge graph). Fixes `<MAXSTRING>` on Mindwalk-scale graphs, enabling Arno-accelerated multi-hop BFS. Adjacency string now scales with BFS result size (~10KB per seed instead of >3.5MB full graph). Handles outbound + inbound edges for undirected BFS.
+
+
 ### v1.63.4 (2026-04-26)
 - chore: merge 080-engine-status to main; NKGAccel.cls added to iris_src from arno upstream
 

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/iris_src/src/Graph/KG/NKGAccel.cls

@@ -450,7 +450,7 @@ ClassMethod RandomWalkJson(seedId As %String, walkLength As %Integer = 20, numWa
     Return result
 }
 
-ClassMethod BFSJson(seed As %String, predicatesJson As %String = "[]", maxHops As %Integer = 3, maxResults As %Integer = 0) As %String
+ ClassMethod BFSJson(seed As %String, predicatesJson As %String = "[]", maxHops As %Integer = 3, maxResults As %Integer = 0) As %String
 {
     Set dllPath = $Get(^||NKGAccel("libPath"), "/usr/irissys/mgr/libarno_callout.so")
     If '$Data(^||NKGAccel("bfs_fn")) {
@@ -466,25 +466,110 @@ ClassMethod BFSJson(seed As %String, predicatesJson As %String = "[]", maxHops A
     If fnid = 0 {
         Return ##class(Graph.KG.Traversal).BFSFastJson(seed, predicatesJson, maxHops, "")
     }
-    Set adjStr = ..ExportAdjacencyWithPreds()
-    If $Extract(adjStr, 1, 8) = "CHUNKED:" {
-        Set nChunks = $Piece(adjStr, ":", 2)
-        Set adjStr = ""
-        For i = 1:1:nChunks {
-            Set adjStr = adjStr _ ^||adjWP(i)
-            If $Length(adjStr) > 3500000 {
-                Kill ^||adjWP
-                Return ##class(Graph.KG.Traversal).BFSFastJson(seed, predicatesJson, maxHops, "")
-            }
-        }
-        Kill ^||adjWP
-    }
+
+    // Per-seed export: only export the subgraph reachable from seed (bounded by maxHops+1)
+    // This keeps adjStr small even on 300K+ edge graphs
+    Set adjStr = ..ExportAdjacencyFromSeed(seed, maxHops + 1, predicatesJson)
     If $Length(adjStr) = 0 {
         Return ##class(Graph.KG.Traversal).BFSFastJson(seed, predicatesJson, maxHops, "")
     }
     Return $ZF(-5, dllid, fnid, adjStr, seed, predicatesJson, maxHops, maxResults)
 }
 
+ClassMethod ExportAdjacencyFromSeed(seed As %String, maxHops As %Integer = 4, predicatesJson As %String = "[]") As %String [ Private ]
+{
+    Set seedIdx = $Get(^NKG("$NI", seed), "")
+    If seedIdx = "" Return ""
+
+    Set preds = ""
+    If predicatesJson '= "[]" && (predicatesJson '= "") {
+        Try { Set preds = ##class(%DynamicArray).%FromJSON(predicatesJson) } Catch { Set preds = "" }
+    }
+
+    // BFS to collect reachable node indices within maxHops
+    Kill ^||seedBFS
+    Set ^||seedBFS("frontier", seedIdx) = ""
+    Set ^||seedBFS("seen", seedIdx) = ""
+
+    For hop = 1:1:maxHops {
+        Kill ^||seedBFS("nextFrontier")
+        Set s = ""
+        For {
+            Set s = $Order(^||seedBFS("frontier", s))
+            Quit:s=""
+            Set pred = ""
+            For {
+                Set pred = $Order(^NKG(-1, s, pred))
+                Quit:pred=""
+                If $IsObject(preds) {
+                    Set allowed = 0
+                    Set predStr = $Get(^NKG("$LS", -pred - 1), "")
+                    For pi = 0:1:preds.%Size()-1 {
+                        If preds.%Get(pi) = predStr { Set allowed = 1 Quit }
+                    }
+                    If 'allowed Continue
+                }
+                Set dst = ""
+                For {
+                    Set dst = $Order(^NKG(-1, s, pred, dst))
+                    Quit:dst=""
+                    Set ^||seedBFS("nodes", s) = ""
+                    Set ^||seedBFS("nodes", dst) = ""
+                    If '$Data(^||seedBFS("seen", dst)) {
+                        Set ^||seedBFS("seen", dst) = ""
+                        Set ^||seedBFS("nextFrontier", dst) = ""
+                    }
+                }
+            }
+            // Inbound traversal for undirected
+            Set pred = ""
+            For {
+                Set pred = $Order(^NKG(-2, s, pred))
+                Quit:pred=""
+                Set dst = ""
+                For {
+                    Set dst = $Order(^NKG(-2, s, pred, dst))
+                    Quit:dst=""
+                    Set ^||seedBFS("nodes", s) = ""
+                    Set ^||seedBFS("nodes", dst) = ""
+                    If '$Data(^||seedBFS("seen", dst)) {
+                        Set ^||seedBFS("seen", dst) = ""
+                        Set ^||seedBFS("nextFrontier", dst) = ""
+                    }
+                }
+            }
+        }
+        If '$Data(^||seedBFS("nextFrontier")) Quit
+        Merge ^||seedBFS("frontier") = ^||seedBFS("nextFrontier")
+        Kill ^||seedBFS("nextFrontier")
+    }
+
+    // Export only edges between collected nodes
+    Set result = ""
+    Set nodeIdx = ""
+    For {
+        Set nodeIdx = $Order(^||seedBFS("nodes", nodeIdx))
+        Quit:nodeIdx=""
+        Set srcStr = $Get(^NKG("$ND", nodeIdx), nodeIdx)
+        Set pred = ""
+        For {
+            Set pred = $Order(^NKG(-1, nodeIdx, pred))
+            Quit:pred=""
+            Set predStr = $Get(^NKG("$LS", -pred - 1), pred)
+            Set dst = ""
+            For {
+                Set dst = $Order(^NKG(-1, nodeIdx, pred, dst))
+                Quit:dst=""
+                If '$Data(^||seedBFS("nodes", dst)) Continue
+                Set dstStr = $Get(^NKG("$ND", dst), dst)
+                Set result = result _ srcStr _ $Char(31) _ predStr _ $Char(31) _ dstStr _ $Char(10)
+            }
+        }
+    }
+    Kill ^||seedBFS
+    Return result
+}
+
 ClassMethod ExportAdjacencyWithPreds() As %String [ Private ]
 {
     Set result = "", chunk = "", chunkNum = 0

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/iris_vector_graph/engine.py

@@ -2099,13 +2099,7 @@ class IRISGraphEngine:
             except Exception:
                 errors.append(f"^KG probe failed: {e}")
 
-        adjacency = AdjacencyStatus(
-            kg_populated=kg_populated,
-            kg_edge_count=kg_count,
-            kg_edge_count_capped=kg_capped,
-            nkg_populated=nkg_populated,
-        )
-
+        kg_predicates_consistent = True
         if kg_populated and tables.edges > 0:
             try:
                 native = self._iris_obj()
@@ -2128,6 +2122,7 @@ class IRISGraphEngine:
                     )
                     row = cursor.fetchone()
                     if row and int(row[0]) == 0:
+                        kg_predicates_consistent = False
                         errors.append(
                             f"^KG predicate mismatch: ^KG has '{kg_pred[:60]}' "
                             f"but rdf_edges has no matching p — "
@@ -2137,6 +2132,15 @@ class IRISGraphEngine:
                 except Exception:
                     pass
 
+        adjacency = AdjacencyStatus(
+            kg_populated=kg_populated,
+            kg_edge_count=kg_count,
+            kg_edge_count_capped=kg_capped,
+            nkg_populated=nkg_populated,
+            kg_predicates_consistent=kg_predicates_consistent,
+            bfs_path="none",
+        )
+
         os_classes = []
         os_deployed = self.capabilities.objectscript_deployed
         _known_classes = [
@@ -2185,6 +2189,11 @@ class IRISGraphEngine:
             plaid_indexes=plaid,
         )
 
+        if arno.loaded and arno.capabilities.get("bfs") and adjacency.nkg_populated:
+            adjacency.bfs_path = "arno"
+        elif objectscript.deployed and adjacency.kg_populated:
+            adjacency.bfs_path = "objectscript"
+
         probe_ms = (_time.perf_counter() - t0) * 1000
         return EngineStatus(
             tables=tables,

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/iris_vector_graph/status.py

@@ -20,6 +20,8 @@ class AdjacencyStatus:
     kg_edge_count: int = 0
     kg_edge_count_capped: bool = False
     nkg_populated: bool = False
+    kg_predicates_consistent: bool = True
+    bfs_path: str = "none"
 
 
 @dataclass
@@ -57,6 +59,14 @@ class EngineStatus:
     def ready_for_bfs(self) -> bool:
         return self.adjacency.kg_populated and self.tables.edges > 0
 
+    @property
+    def ready_for_multihop_bfs(self) -> bool:
+        if not self.ready_for_bfs:
+            return False
+        if not self.adjacency.kg_predicates_consistent:
+            return False
+        return self.adjacency.bfs_path in ("arno", "objectscript")
+
     @property
     def ready_for_vector_search(self) -> bool:
         return self.tables.node_embeddings > 0
@@ -95,6 +105,15 @@ class EngineStatus:
             f"  {tick(self.adjacency.nkg_populated)} ^NKG  (Arno integer index)",
         ]
 
+        bfs_sym = "✓" if self.adjacency.bfs_path in ("arno", "objectscript") else "✗"
+        lines.append(f"  {bfs_sym} BFS path: {self.adjacency.bfs_path}")
+
+        if not self.adjacency.kg_predicates_consistent and self.adjacency.kg_populated:
+            lines.append(
+                "  ⚠ ^KG predicate mismatch — stale from different data snapshot. "
+                "Run BuildKG() after reloading graph data."
+            )
+
         if not self.adjacency.kg_populated and self.tables.edges > 0:
             lines.append(
                 "  ⚠ ^KG empty but edges exist — "

{iris_vector_graph-1.63.4 → iris_vector_graph-1.63.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "iris-vector-graph"
-version = "1.63.4"
+version = "1.63.6"
 description = "Transactional Graph + Vector retrieval system for InterSystems IRIS with hybrid search, openCypher, and GraphQL APIs"
 readme = "README.md"
 license = "MIT"