Bindi/support for pinecone serverless (#37756)

Co-authored-by: Roie Schwaber-Cohen <@rschwabco>

airbyte-integrations/connectors/destination-pinecone/destination_pinecone/destination.py

@@ -24,28 +24,44 @@ class DestinationPinecone(Destination):
     embedder: Embedder
 
     def _init_indexer(self, config: ConfigModel):
-        self.embedder = create_from_config(config.embedding, config.processing)
-        self.indexer = PineconeIndexer(config.indexing, self.embedder.embedding_dimensions)
+        try:
+            self.embedder = create_from_config(config.embedding, config.processing)
+            self.indexer = PineconeIndexer(config.indexing, self.embedder.embedding_dimensions)
+        except Exception as e:
+            return AirbyteConnectionStatus(status=Status.FAILED, message=str(e))
 
     def write(
         self, config: Mapping[str, Any], configured_catalog: ConfiguredAirbyteCatalog, input_messages: Iterable[AirbyteMessage]
     ) -> Iterable[AirbyteMessage]:
-        config_model = ConfigModel.parse_obj(config)
-        self._init_indexer(config_model)
-        writer = Writer(
-            config_model.processing, self.indexer, self.embedder, batch_size=BATCH_SIZE, omit_raw_text=config_model.omit_raw_text
-        )
-        yield from writer.write(configured_catalog, input_messages)
+        try:
+            config_model = ConfigModel.parse_obj(config)
+            self._init_indexer(config_model)
+            writer = Writer(
+                config_model.processing, self.indexer, self.embedder, batch_size=BATCH_SIZE, omit_raw_text=config_model.omit_raw_text
+            )
+            yield from writer.write(configured_catalog, input_messages)
+        except Exception as e:
+            yield AirbyteMessage(type="LOG", log=AirbyteLogger(level="ERROR", message=str(e)))
 
     def check(self, logger: AirbyteLogger, config: Mapping[str, Any]) -> AirbyteConnectionStatus:
-        parsed_config = ConfigModel.parse_obj(config)
-        self._init_indexer(parsed_config)
-        checks = [self.embedder.check(), self.indexer.check(), DocumentProcessor.check_config(parsed_config.processing)]
-        errors = [error for error in checks if error is not None]
-        if len(errors) > 0:
-            return AirbyteConnectionStatus(status=Status.FAILED, message="\n".join(errors))
-        else:
-            return AirbyteConnectionStatus(status=Status.SUCCEEDED)
+        try:
+            parsed_config = ConfigModel.parse_obj(config)
+            init_status = self._init_indexer(parsed_config)
+            if init_status and init_status.status == Status.FAILED:
+                logger.error(f"Initialization failed with message: {init_status.message}")
+                return init_status  # Return the failure status immediately if initialization fails
+
+            checks = [self.embedder.check(), self.indexer.check(), DocumentProcessor.check_config(parsed_config.processing)]
+            errors = [error for error in checks if error is not None]
+            if len(errors) > 0:
+                error_message = "\n".join(errors)
+                logger.error(f"Configuration check failed: {error_message}")
+                return AirbyteConnectionStatus(status=Status.FAILED, message=error_message)
+            else:
+                return AirbyteConnectionStatus(status=Status.SUCCEEDED)
+        except Exception as e:
+            logger.error(f"Exception during configuration check: {str(e)}")
+            return AirbyteConnectionStatus(status=Status.FAILED, message=str(e))
 
     def spec(self, *args: Any, **kwargs: Any) -> ConnectorSpecification:
         return ConnectorSpecification(

airbyte-integrations/connectors/destination-pinecone/destination_pinecone/indexer.py

@@ -5,13 +5,15 @@
 import uuid
 from typing import Optional
 
-import pinecone
 import urllib3
 from airbyte_cdk.destinations.vector_db_based.document_processor import METADATA_RECORD_ID_FIELD, METADATA_STREAM_FIELD
 from airbyte_cdk.destinations.vector_db_based.indexer import Indexer
 from airbyte_cdk.destinations.vector_db_based.utils import create_chunks, create_stream_identifier, format_exception
+from airbyte_cdk.models import AirbyteConnectionStatus, Status
 from airbyte_cdk.models.airbyte_protocol import ConfiguredAirbyteCatalog, DestinationSyncMode
 from destination_pinecone.config import PineconeIndexingModel
+from pinecone import PineconeException
+from pinecone.grpc import PineconeGRPC
 
 # large enough to speed up processing, small enough to not hit pinecone request limits
 PINECONE_BATCH_SIZE = 40
@@ -29,32 +31,54 @@ class PineconeIndexer(Indexer):
 
     def __init__(self, config: PineconeIndexingModel, embedding_dimensions: int):
         super().__init__(config)
-        pinecone.init(api_key=config.pinecone_key, environment=config.pinecone_environment, threaded=True)
+        try:
+            self.pc = PineconeGRPC(api_key=config.pinecone_key, threaded=True)
+        except PineconeException as e:
+            return AirbyteConnectionStatus(status=Status.FAILED, message=str(e))
 
-        self.pinecone_index = pinecone.GRPCIndex(config.index)
+        self.pinecone_index = self.pc.Index(config.index)
         self.embedding_dimensions = embedding_dimensions
 
+    def determine_spec_type(self, index_name):
+        description = self.pc.describe_index(index_name)
+        spec_keys = description.get("spec", {})
+        if "pod" in spec_keys:
+            return "pod"
+        elif "serverless" in spec_keys:
+            return "serverless"
+        else:
+            raise ValueError("Unknown index specification type.")
+
     def pre_sync(self, catalog: ConfiguredAirbyteCatalog):
-        index_description = pinecone.describe_index(self.config.index)
-        self._pod_type = index_description.pod_type
+        self._pod_type = self.determine_spec_type(self.config.index)
+
         for stream in catalog.streams:
+            stream_identifier = create_stream_identifier(stream.stream)
             if stream.destination_sync_mode == DestinationSyncMode.overwrite:
                 self.delete_vectors(
-                    filter={METADATA_STREAM_FIELD: create_stream_identifier(stream.stream)}, namespace=stream.stream.namespace
+                    filter={METADATA_STREAM_FIELD: stream_identifier}, namespace=stream.stream.namespace, prefix=stream_identifier
                 )
 
     def post_sync(self):
         return []
 
-    def delete_vectors(self, filter, namespace=None):
+    def delete_vectors(self, filter, namespace=None, prefix=None):
         if self._pod_type == "starter":
             # Starter pod types have a maximum of 100000 rows
             top_k = 10000
             self.delete_by_metadata(filter, top_k, namespace)
+        elif self._pod_type == "serverless":
+            if prefix == None:
+                raise ValueError("Prefix is required for a serverless index.")
+            self.delete_by_prefix(prefix=prefix, namespace=namespace)
         else:
+            # Pod spec
             self.pinecone_index.delete(filter=filter, namespace=namespace)
 
     def delete_by_metadata(self, filter, top_k, namespace=None):
+        """
+        Applicable to Starter implementation only. Deletes all vectors that match the given metadata filter.
+        """
         zero_vector = [0.0] * self.embedding_dimensions
         query_result = self.pinecone_index.query(vector=zero_vector, filter=filter, top_k=top_k, namespace=namespace)
         while len(query_result.matches) > 0:
@@ -66,6 +90,13 @@ def delete_by_metadata(self, filter, top_k, namespace=None):
                     self.pinecone_index.delete(ids=list(batch), namespace=namespace)
             query_result = self.pinecone_index.query(vector=zero_vector, filter=filter, top_k=top_k, namespace=namespace)
 
+    def delete_by_prefix(self, prefix, namespace=None):
+        """
+        Applicable to Serverless implementation only. Deletes all vectors with the given prefix.
+        """
+        for ids in self.pinecone_index.list(prefix=prefix, namespace=namespace):
+            self.pinecone_index.delete(ids=ids, namespace=namespace)
+
     def _truncate_metadata(self, metadata: dict) -> dict:
         """
         Normalize metadata to ensure it is within the size limit and doesn't contain complex objects.
@@ -85,34 +116,45 @@ def _truncate_metadata(self, metadata: dict) -> dict:
 
         return result
 
-    def index(self, document_chunks, namespace, stream):
+    def index(self, document_chunks, namespace, streamName):
         pinecone_docs = []
         for i in range(len(document_chunks)):
             chunk = document_chunks[i]
             metadata = self._truncate_metadata(chunk.metadata)
             if chunk.page_content is not None:
                 metadata["text"] = chunk.page_content
-            pinecone_docs.append((str(uuid.uuid4()), chunk.embedding, metadata))
+            prefix = streamName
+            pinecone_docs.append((prefix + "#" + str(uuid.uuid4()), chunk.embedding, metadata))
         serial_batches = create_chunks(pinecone_docs, batch_size=PINECONE_BATCH_SIZE * PARALLELISM_LIMIT)
         for batch in serial_batches:
-            async_results = [
-                self.pinecone_index.upsert(vectors=ids_vectors_chunk, async_req=True, show_progress=False, namespace=namespace)
-                for ids_vectors_chunk in create_chunks(batch, batch_size=PINECONE_BATCH_SIZE)
-            ]
+            async_results = []
+            for ids_vectors_chunk in create_chunks(batch, batch_size=PINECONE_BATCH_SIZE):
+                async_result = self.pinecone_index.upsert(vectors=ids_vectors_chunk, async_req=True, show_progress=False)
+                async_results.append(async_result)
             # Wait for and retrieve responses (this raises in case of error)
             [async_result.result() for async_result in async_results]
 
     def delete(self, delete_ids, namespace, stream):
+        filter = {METADATA_RECORD_ID_FIELD: {"$in": delete_ids}}
         if len(delete_ids) > 0:
-            self.delete_vectors(filter={METADATA_RECORD_ID_FIELD: {"$in": delete_ids}}, namespace=namespace)
+            if self._pod_type == "starter":
+                # Starter pod types have a maximum of 100000 rows
+                top_k = 10000
+                self.delete_by_metadata(filter=filter, top_k=top_k, namespace=namespace)
+            elif self._pod_type == "serverless":
+                self.pinecone_index.delete(ids=delete_ids, namespace=namespace)
+            else:
+                # Pod spec
+                self.pinecone_index.delete(filter=filter, namespace=namespace)
 
     def check(self) -> Optional[str]:
         try:
-            indexes = pinecone.list_indexes()
-            if self.config.index not in indexes:
+            list = self.pc.list_indexes()
+            index_names = [index["name"] for index in list.indexes]
+            if self.config.index not in index_names:
                 return f"Index {self.config.index} does not exist in environment {self.config.pinecone_environment}."
 
-            description = pinecone.describe_index(self.config.index)
+            description = self.pc.describe_index(self.config.index)
             actual_dimension = int(description.dimension)
             if actual_dimension != self.embedding_dimensions:
                 return f"Your embedding configuration will produce vectors with dimension {self.embedding_dimensions:d}, but your index is configured with dimension {actual_dimension:d}. Make sure embedding and indexing configurations match."
@@ -121,7 +163,7 @@ def check(self) -> Optional[str]:
                 if f"Failed to resolve 'controller.{self.config.pinecone_environment}.pinecone.io'" in str(e.reason):
                     return f"Failed to resolve environment, please check whether {self.config.pinecone_environment} is correct."
 
-            if isinstance(e, pinecone.exceptions.UnauthorizedException):
+            if isinstance(e, PineconeException):
                 if e.body:
                     return e.body
 

airbyte-integrations/connectors/destination-pinecone/integration_tests/pinecone_integration_test.py

@@ -4,33 +4,53 @@
 
 import json
 import logging
+import time
 
-import pinecone
 from airbyte_cdk.destinations.vector_db_based.embedder import OPEN_AI_VECTOR_SIZE
 from airbyte_cdk.destinations.vector_db_based.test_utils import BaseIntegrationTest
 from airbyte_cdk.models import DestinationSyncMode, Status
 from destination_pinecone.destination import DestinationPinecone
 from langchain.embeddings import OpenAIEmbeddings
 from langchain.vectorstores import Pinecone
+from pinecone import Pinecone as PineconeREST
+from pinecone import PineconeException
+from pinecone.grpc import PineconeGRPC
 
 
 class PineconeIntegrationTest(BaseIntegrationTest):
     def _init_pinecone(self):
-        pinecone.init(api_key=self.config["indexing"]["pinecone_key"], environment=self.config["indexing"]["pinecone_environment"])
-        self.pinecone_index = pinecone.Index(self.config["indexing"]["index"])
-
+        self.pc = PineconeGRPC(api_key=self.config["indexing"]["pinecone_key"])
+        self.pinecone_index = self.pc.Index(self.config["indexing"]["index"])
+        self.pc_rest = PineconeREST(api_key=self.config["indexing"]["pinecone_key"])
+        self.pinecone_index_rest = self.pc_rest.Index(name=self.config["indexing"]["index"])
+
+    def _wait(self):
+        print("Waiting for Pinecone...", end='', flush=True)
+        for i in range(15):
+            time.sleep(1)
+            print(".", end='', flush=True)
+        print()  # Move to the next line after the loop
+
     def setUp(self):
         with open("secrets/config.json", "r") as f:
             self.config = json.loads(f.read())
         self._init_pinecone()
 
     def tearDown(self):
-        # make sure pinecone is initialized correctly before cleaning up
+        self._wait()
+        # make sure pinecone is initialized correctly before cleaning up        
         self._init_pinecone()
-        self.pinecone_index.delete(delete_all=True)
+        try:
+            self.pinecone_index.delete(delete_all=True)
+        except PineconeException as e:
+            if "Namespace not found" not in str(e):
+                raise(e)
+            else :
+                print("Noting to delete. No data in the index/namespace.")
+
 
     def test_check_valid_config(self):
-        outcome = DestinationPinecone().check(logging.getLogger("airbyte"), self.config)
+        outcome = DestinationPinecone().check(logging.getLogger("airbyte"), self.config)        
         assert outcome.status == Status.SUCCEEDED
 
     def test_check_invalid_config(self):
@@ -43,10 +63,11 @@ def test_check_invalid_config(self):
                     "mode": "pinecone",
                     "pinecone_key": "mykey",
                     "index": "testdata",
-                    "pinecone_environment": "asia-southeast1-gcp-free",
+                    "pinecone_environment": "us-west1-gcp",
                 },
             },
         )
+
         assert outcome.status == Status.FAILED
 
     def test_write(self):
@@ -57,14 +78,21 @@ def test_write(self):
         # initial sync
         destination = DestinationPinecone()
         list(destination.write(self.config, catalog, [*first_record_chunk, first_state_message]))
+
+
+        self._wait()        
         assert self.pinecone_index.describe_index_stats().total_vector_count == 5
 
         # incrementalally update a doc
         incremental_catalog = self._get_configured_catalog(DestinationSyncMode.append_dedup)
         list(destination.write(self.config, incremental_catalog, [self._record("mystream", "Cats are nice", 2), first_state_message]))
+
+        self._wait() 
+
         result = self.pinecone_index.query(
             vector=[0] * OPEN_AI_VECTOR_SIZE, top_k=10, filter={"_ab_record_id": "mystream_2"}, include_metadata=True
         )
+
         assert len(result.matches) == 1
         assert (
             result.matches[0].metadata["text"] == "str_col: Cats are nice"
@@ -73,6 +101,6 @@ def test_write(self):
         # test langchain integration
         embeddings = OpenAIEmbeddings(openai_api_key=self.config["embedding"]["openai_key"])
         self._init_pinecone()
-        vector_store = Pinecone(self.pinecone_index, embeddings.embed_query, "text")
+        vector_store = Pinecone(self.pinecone_index_rest, embeddings.embed_query, "text")
         result = vector_store.similarity_search("feline animals", 1)
         assert result[0].metadata["_ab_record_id"] == "mystream_2"

airbyte-integrations/connectors/destination-pinecone/metadata.yaml

@@ -13,7 +13,7 @@ data:
   connectorSubtype: vectorstore
   connectorType: destination
   definitionId: 3d2b6f84-7f0d-4e3f-a5e5-7c7d4b50eabd
-  dockerImageTag: 0.0.24
+  dockerImageTag: 0.1.0
   dockerRepository: airbyte/destination-pinecone
   documentationUrl: https://docs.airbyte.com/integrations/destinations/pinecone
   githubIssueLabel: destination-pinecone