fix: skip unsupported meta fields in PineconeDB

integrations/pinecone/src/haystack_integrations/document_stores/pinecone/document_store.py

@@ -26,6 +26,7 @@
 
 
 DEFAULT_STARTER_PLAN_SPEC = {"serverless": {"region": "us-east-1", "cloud": "aws"}}
+METADATA_SUPPORTED_TYPES = str, int, bool  # List[str] is supported and checked separately
 
 
 class PineconeDocumentStore:
@@ -295,6 +296,29 @@ def _convert_query_result_to_documents(self, query_result: Dict[str, Any]) -> Li
 
         return documents
 
+    @staticmethod
+    def check_metadata(document: Document):
+        def valid_type(value: Any):
+            return isinstance(value, METADATA_SUPPORTED_TYPES) or (
+                isinstance(value, list) and all(isinstance(i, str) for i in value)
+            )
+
+        if document.meta:
+            discarded_keys = []
+            for key, value in document.meta.items():
+                if not valid_type(value):
+                    discarded_keys.append(key)
+                    document.meta[key] = "IGNORED"
+
+            if discarded_keys:
+                msg = (
+                    f"Document {document.id} has metadata fields with unsupported types: {discarded_keys}. "
+                    f"Only str, int, bool, and List[str] are supported. The values of these fields will be ignored."
+                )
+                logger.warning(msg)
+
+        return document
+
     def _convert_documents_to_pinecone_format(self, documents: List[Document]) -> List[Dict[str, Any]]:
         documents_for_pinecone = []
         for document in documents:
@@ -305,6 +329,10 @@ def _convert_documents_to_pinecone_format(self, documents: List[Document]) -> Li
                     "A dummy embedding will be used, but this can affect the search results. "
                 )
                 embedding = self._dummy_vector
+
+            if document.meta:
+                self.check_metadata(document)
+
             doc_for_pinecone = {"id": document.id, "values": embedding, "metadata": dict(document.meta)}
 
             # we save content/dataframe as metadata

integrations/pinecone/tests/test_document_store.py

@@ -142,6 +142,42 @@ def test_convert_dict_spec_to_pinecone_object_fail():
         PineconeDocumentStore._convert_dict_spec_to_pinecone_object(dict_spec)
 
 
+def test_check_metadata_invalid():
+    invalid_metadata_doc = Document(
+        content="The moonlight shimmered ",
+        meta={
+            "source_id": "62049ba1d1e1d5ebb1f6230b0b00c5356b8706c56e0b9c36b1dfc86084cd75f0",
+            "page_number": 1,
+            "split_id": 0,
+            "split_idx_start": 0,
+            "_split_overlap": [
+                {"doc_id": "68ed48ba830048c5d7815874ed2de794722e6d10866b6c55349a914fd9a0df65", "range": (0, 20)}
+            ],
+        },
+    )
+    pinecone_doc = PineconeDocumentStore.check_metadata(invalid_metadata_doc)
+
+    assert pinecone_doc.meta["source_id"] == "62049ba1d1e1d5ebb1f6230b0b00c5356b8706c56e0b9c36b1dfc86084cd75f0"
+    assert pinecone_doc.meta["page_number"] == 1
+    assert pinecone_doc.meta["split_id"] == 0
+    assert pinecone_doc.meta["split_idx_start"] == 0
+    assert pinecone_doc.meta["_split_overlap"] == "IGNORED"
+
+
+def test_check_metadata_valid():
+    valid_metadata_doc = Document(
+        content="The moonlight shimmered ",
+        meta={
+            "source_id": "62049ba1d1e1d5ebb1f6230b0b00c5356b8706c56e0b9c36b1dfc86084cd75f0",
+            "page_number": 1,
+        },
+    )
+    pinecone_doc = PineconeDocumentStore.check_metadata(valid_metadata_doc)
+
+    assert pinecone_doc.meta["source_id"] == "62049ba1d1e1d5ebb1f6230b0b00c5356b8706c56e0b9c36b1dfc86084cd75f0"
+    assert pinecone_doc.meta["page_number"] == 1
+
+
 @pytest.mark.integration
 @pytest.mark.skipif("PINECONE_API_KEY" not in os.environ, reason="PINECONE_API_KEY not set")
 def test_serverless_index_creation_from_scratch(sleep_time):