nucliadb 6.2.1.post2838__py3-none-any.whl → 6.2.1.post2842__py3-none-any.whl

nucliadb/common/external_index_providers/base.py

@@ -28,7 +28,7 @@ from nucliadb.common.counters import IndexCounts
 from nucliadb.common.external_index_providers.exceptions import ExternalIndexingError
 from nucliadb.common.ids import ParagraphId
 from nucliadb_models.external_index_providers import ExternalIndexProviderType
-from nucliadb_models.search import SCORE_TYPE, TextPosition
+from nucliadb_models.search import SCORE_TYPE, Relations, TextPosition
 from nucliadb_protos.knowledgebox_pb2 import (
     CreateExternalIndexProviderMetadata,
     StoredExternalIndexProviderMetadata,
@@ -73,6 +73,7 @@ class TextBlockMatch(BaseModel):
     paragraph_labels: list[str] = []
     field_labels: list[str] = []
     text: Optional[str] = None
+    relevant_relations: Optional[Relations] = None
 
 
 class QueryResults(BaseModel):

nucliadb/common/ids.py

@@ -111,13 +111,11 @@ class FieldId:
         parts = value.split("/")
         if len(parts) == 3:
             rid, _type, key = parts
-            if _type not in FIELD_TYPE_STR_TO_PB:
-                raise ValueError(f"Invalid FieldId: {value}")
+            _type = cls.parse_field_type(_type)
             return cls(rid=rid, type=_type, key=key)
         elif len(parts) == 4:
             rid, _type, key, subfield_id = parts
-            if _type not in FIELD_TYPE_STR_TO_PB:
-                raise ValueError(f"Invalid FieldId: {value}")
+            _type = cls.parse_field_type(_type)
             return cls(
                 rid=rid,
                 type=_type,
@@ -127,6 +125,22 @@ class FieldId:
         else:
             raise ValueError(f"Invalid FieldId: {value}")
 
+    @classmethod
+    def parse_field_type(cls, _type: str) -> str:
+        if _type not in FIELD_TYPE_STR_TO_PB:
+            # Try to parse the enum value
+            # XXX: This is to support field types that are integer values of FieldType
+            # Which is how legacy processor relations reported the paragraph_id
+            try:
+                type_pb = FieldType.ValueType(int(_type))
+            except ValueError:
+                raise ValueError(f"Invalid FieldId: {_type}")
+            if type_pb in FIELD_TYPE_PB_TO_STR:
+                return FIELD_TYPE_PB_TO_STR[type_pb]
+            else:
+                raise ValueError(f"Invalid FieldId: {_type}")
+        return _type
+
 
 @dataclass
 class ParagraphId:

nucliadb/search/api/v1/suggest.py

@@ -151,8 +151,6 @@ async def suggest(
         search_results = await merge_suggest_results(
             results,
             kbid=kbid,
-            show=show,
-            field_type_filter=field_type_filter,
             highlight=highlight,
         )
 

nucliadb/search/search/chat/ask.py

@@ -57,6 +57,7 @@ from nucliadb.search.search.exceptions import (
     IncompleteFindResultsError,
     InvalidQueryError,
 )
+from nucliadb.search.search.graph_strategy import get_graph_results
 from nucliadb.search.search.metrics import RAGMetrics
 from nucliadb.search.search.query import QueryParser
 from nucliadb.search.utilities import get_predict
@@ -75,6 +76,7 @@ from nucliadb_models.search import (
     ErrorAskResponseItem,
     FindParagraph,
     FindRequest,
+    GraphStrategy,
     JSONAskResponseItem,
     KnowledgeboxFindResults,
     MetadataAskResponseItem,
@@ -629,6 +631,13 @@ def parse_prequeries(ask_request: AskRequest) -> Optional[PreQueriesStrategy]:
     return None
 
 
+def parse_graph_strategy(ask_request: AskRequest) -> Optional[GraphStrategy]:
+    for rag_strategy in ask_request.rag_strategies:
+        if rag_strategy.name == RagStrategyName.GRAPH:
+            return cast(GraphStrategy, rag_strategy)
+    return None
+
+
 async def retrieval_step(
     kbid: str,
     main_query: str,
@@ -675,17 +684,33 @@ async def retrieval_in_kb(
     metrics: RAGMetrics,
 ) -> RetrievalResults:
     prequeries = parse_prequeries(ask_request)
+    graph_strategy = parse_graph_strategy(ask_request)
     with metrics.time("retrieval"):
-        main_results, prequeries_results, query_parser = await get_find_results(
-            kbid=kbid,
-            query=main_query,
-            item=ask_request,
-            ndb_client=client_type,
-            user=user_id,
-            origin=origin,
-            metrics=metrics,
-            prequeries_strategy=prequeries,
-        )
+        prequeries_results = None
+        if graph_strategy is not None:
+            main_results, query_parser = await get_graph_results(
+                kbid=kbid,
+                query=main_query,
+                item=ask_request,
+                ndb_client=client_type,
+                user=user_id,
+                origin=origin,
+                graph_strategy=graph_strategy,
+                metrics=metrics,
+                shards=ask_request.shards,
+            )
+        # TODO (oni): Fallback to normal retrieval if no graph results are found
+        else:
+            main_results, prequeries_results, query_parser = await get_find_results(
+                kbid=kbid,
+                query=main_query,
+                item=ask_request,
+                ndb_client=client_type,
+                user=user_id,
+                origin=origin,
+                metrics=metrics,
+                prequeries_strategy=prequeries,
+            )
         if len(main_results.resources) == 0 and all(
             len(prequery_result.resources) == 0 for (_, prequery_result) in prequeries_results or []
         ):

nucliadb/search/search/chat/prompt.py

@@ -1013,8 +1013,10 @@ class PromptContextBuilder:
                 neighbouring_paragraphs = cast(NeighbouringParagraphsStrategy, strategy)
             elif strategy.name == RagStrategyName.METADATA_EXTENSION:
                 metadata_extension = cast(MetadataExtensionStrategy, strategy)
-            elif strategy.name != RagStrategyName.PREQUERIES:  # pragma: no cover
-                # Prequeries are not handled here
+            elif (
+                strategy.name != RagStrategyName.PREQUERIES and strategy.name != RagStrategyName.GRAPH
+            ):  # pragma: no cover
+                # Prequeries and graph are not handled here
                 logger.warning(
                     "Unknown rag strategy",
                     extra={"strategy": strategy.name, "kbid": self.kbid},

nucliadb/search/search/chat/query.py

@@ -18,7 +18,7 @@
 # along with this program. If not, see <http://www.gnu.org/licenses/>.
 #
 import asyncio
-from typing import Optional
+from typing import Iterable, Optional
 
 from nucliadb.common.models_utils import to_proto
 from nucliadb.search import logger
@@ -51,6 +51,7 @@ from nucliadb_models.search import (
 )
 from nucliadb_protos import audit_pb2
 from nucliadb_protos.nodereader_pb2 import RelationSearchResponse, SearchRequest, SearchResponse
+from nucliadb_protos.utils_pb2 import RelationNode
 from nucliadb_telemetry.errors import capture_exception
 from nucliadb_utils.utilities import get_audit
 
@@ -145,15 +146,7 @@ async def get_find_results(
     return main_results, prequeries_results, query_parser
 
 
-async def run_main_query(
-    kbid: str,
-    query: str,
-    item: AskRequest,
-    ndb_client: NucliaDBClientType,
-    user: str,
-    origin: str,
-    metrics: RAGMetrics = RAGMetrics(),
-) -> tuple[KnowledgeboxFindResults, QueryParser]:
+def find_request_from_ask_request(item: AskRequest, query: str) -> FindRequest:
     find_request = FindRequest()
     find_request.resource_filters = item.resource_filters
     find_request.features = []
@@ -189,7 +182,19 @@ async def run_main_query(
     find_request.show_hidden = item.show_hidden
 
     # this executes the model validators, that can tweak some fields
-    FindRequest.model_validate(find_request)
+    return FindRequest.model_validate(find_request)
+
+
+async def run_main_query(
+    kbid: str,
+    query: str,
+    item: AskRequest,
+    ndb_client: NucliaDBClientType,
+    user: str,
+    origin: str,
+    metrics: RAGMetrics = RAGMetrics(),
+) -> tuple[KnowledgeboxFindResults, QueryParser]:
+    find_request = find_request_from_ask_request(item, query)
 
     find_results, incomplete, query_parser = await find(
         kbid,
@@ -211,36 +216,59 @@ async def get_relations_results(
     text_answer: str,
     target_shard_replicas: Optional[list[str]],
     timeout: Optional[float] = None,
+    only_with_metadata: bool = False,
+    only_agentic_relations: bool = False,
 ) -> Relations:
     try:
         predict = get_predict()
         detected_entities = await predict.detect_entities(kbid, text_answer)
-        request = SearchRequest()
-        request.relation_subgraph.entry_points.extend(detected_entities)
-        request.relation_subgraph.depth = 1
-
-        results: list[SearchResponse]
-        (
-            results,
-            _,
-            _,
-        ) = await node_query(
-            kbid,
-            Method.SEARCH,
-            request,
+
+        return await get_relations_results_from_entities(
+            kbid=kbid,
+            entities=detected_entities,
             target_shard_replicas=target_shard_replicas,
             timeout=timeout,
-            use_read_replica_nodes=True,
-            retry_on_primary=False,
+            only_with_metadata=only_with_metadata,
+            only_agentic_relations=only_agentic_relations,
         )
-        relations_results: list[RelationSearchResponse] = [result.relation for result in results]
-        return await merge_relations_results(relations_results, request.relation_subgraph)
     except Exception as exc:
         capture_exception(exc)
         logger.exception("Error getting relations results")
         return Relations(entities={})
 
 
+async def get_relations_results_from_entities(
+    *,
+    kbid: str,
+    entities: Iterable[RelationNode],
+    target_shard_replicas: Optional[list[str]],
+    timeout: Optional[float] = None,
+    only_with_metadata: bool = False,
+    only_agentic_relations: bool = False,
+) -> Relations:
+    request = SearchRequest()
+    request.relation_subgraph.entry_points.extend(entities)
+    request.relation_subgraph.depth = 1
+    results: list[SearchResponse]
+    (
+        results,
+        _,
+        _,
+    ) = await node_query(
+        kbid,
+        Method.SEARCH,
+        request,
+        target_shard_replicas=target_shard_replicas,
+        timeout=timeout,
+        use_read_replica_nodes=True,
+        retry_on_primary=False,
+    )
+    relations_results: list[RelationSearchResponse] = [result.relation for result in results]
+    return await merge_relations_results(
+        relations_results, request.relation_subgraph, only_with_metadata, only_agentic_relations
+    )
+
+
 def maybe_audit_chat(
     *,
     kbid: str,