# PG VectorStore 使用說明及問題

### 使用說明

1. 安裝依賴項目
由於 langflow 目前並沒有用到 [pgvector-python](https://github.com/pgvector/pgvector-python)
但是 [LangChain PGVector](https://python.langchain.com/docs/integrations/vectorstores/pgvector) 需要用到
因此要製作 PG VectorStore Client 給 LangFlow 用就必須先安裝相關套件

    poetry add pgvector

如果沒用過 Poetry 的話可以參考這個網站: https://blog.kyomind.tw/python-poetry/

2. 啟動專案
langflow 都把啟動專案的指令包在 makefile 了
因此只要 `make backend` 就可以成功 build 好後端囉

    make backend
    make frontend # 前端也要記得 build

3. 放程式碼進去 Custom Component node，不過等等會有問題需要解決

    from typing import Optional
    from langflow import CustomComponent

    from langchain.vectorstores.pgvector import PGVector
    from langchain.schema import Document
    from langchain.vectorstores.base import VectorStore
    from langchain.embeddings.base import Embeddings


    class PostgresqlVectorComponent(CustomComponent):
        """
        A custom component for implementing a Vector Store using PostgreSQL.
        """

        display_name: str = "PGVector"
        description: str = "Implementation of Vector Store using PostgreSQL"
        documentation = "https://python.langchain.com/docs/integrations/vectorstores/pgvector"
        beta = True

        def build_config(self):
            """
            Builds the configuration for the component.

            Returns:
            - dict: A dictionary containing the configuration options for the component.
            """
            return {
                "index_name": {"display_name": "Index Name", "value": "your_index"},
                "code": {"show": True, "display_name": "Code"},
                "documents": {"display_name": "Documents", "is_list": True},
                "embedding": {"display_name": "Embedding"},
                "pg_server_url": {
                    "display_name": "PostgreSQL Server Connection String",
                    "advanced": False,
                },
                "collection_name": {"display_name": "Table", "advanced": False},
            }

        def build(
            self,
            embedding: Embeddings,
            pg_server_url: str,
            collection_name: str,
            documents: Optional[Document] = None,
        ) -> VectorStore:
            """
            Builds the Vector Store or BaseRetriever object.

            Args:
            - embedding (Embeddings): The embeddings to use for the Vector Store.
            - documents (Optional[Document]): The documents to use for the Vector Store.
            - collection_name (str): The name of the PG table.
            - pg_server_url (str): The URL for the PG server.

            Returns:
            - VectorStore: The Vector Store object.
            """

            return PGVector.from_documents(
                embedding=embedding,
                documents=documents,
                collection_name=collection_name,
                connection_string=pg_server_url,
            )

### 問題點：

`from langchain.vectorstores.pgvector import PGVector`

LangChain 這邊的 **PGVector** 出了點問題
如果你進去 [LangChain 的 source code 看 PGVector](https://github.com/langchain-ai/langchain/blob/3bf39ca635ffb95e6fca0244b4a30f6a2e036bed/libs/langchain/langchain/vectorstores/pgvector.py#L130) 是怎麼寫的：
會發現有一行程式碼被註解了 - - -> `self.create_vector_extension()`

    def __post_init__(
        self,
    ) -> None:
        """
        Initialize the store.
        """
        self._conn = self.connect()
        # self.create_vector_extension()
        from langchain.vectorstores._pgvector_data_models import (
            CollectionStore,
            EmbeddingStore,
        )

        self.CollectionStore = CollectionStore
        self.EmbeddingStore = EmbeddingStore
        self.create_tables_if_not_exists()
        self.create_collection()

`create_vector_extension` 會用來初始化 pg 用的插件
如果被註解掉就會發現 ORM 認不得下面的 `VECTOR`

以下為預期執行結果，由 LangFlow 出現的報錯訊息

 ValueError: Error building node pgvector(ID:pgvector-15vEN):                                  
 (psycopg2.errors.UndefinedObject) type "vector" does not exist                                
 LINE 5:  embedding VECTOR,                                                                    
                    ^                                                                          

 [SQL:                                                                                         
 CREATE TABLE langchain_pg_embedding (                                                         
         uuid UUID NOT NULL,                                                                   
         collection_id UUID,                                                                   
         embedding VECTOR,                                                                     
         document VARCHAR,                                                                     
         cmetadata JSON,                                                                       
         custom_id VARCHAR,                                                                    
         PRIMARY KEY (uuid),                                                                   
         FOREIGN KEY(collection_id) REFERENCES langchain_pg_collection (uuid) ON               
 DELETE CASCADE                                                                                
 )        

### 觀察
1. LangChain 現在的 [master branch](https://github.com/langchain-ai/langchain/blob/cf271784fade8dc19879cbdfa172df84e65d1e7c/libs/langchain/langchain/vectorstores/pgvector.py#L139C39-L139C39) 已經把這個笨註解拿掉了
2. LangFlow 0.5.7 用的 LangFlow 版本並沒有用到已修復的版本

| LangFlow 版本| LangChain 版本 | 已修復 |
| -------- | -------- | -------- |
| (尚未推出) | v0.0.322 | V |
| (尚未推出) | v0.0.321 | X |
| 0.5.7 | ^0.0.320 | X |

### 解法
1. 自己進去 source code 改（暫時性解法）
2. 等 LangFlow 升 LangChain 版本