Sultan
/
Flowise
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add Postgres vector store using pgvector

This commit is contained in:
fabiovalencio 2023-08-16 11:59:50 +01:00
parent 8034076361
commit 44bac07b90
8 changed files with 418 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
packages/components/credentials/PostgresApi.credential.ts Normal file
View File

@ -0,0 +1,31 @@
import { INodeParams, INodeCredential } from '../src/Interface'
class PostgresApi implements INodeCredential {
label: string
name: string
version: number
description: string
inputs: INodeParams[]
constructor() {
this.label = 'Postgres API'
this.name = 'PostgresApi'
this.version = 1.0
this.inputs = [
{
label: 'User',
name: 'user',
type: 'string',
placeholder: '<POSTGRES_USERNAME>'
},
{
label: 'Password',
name: 'password',
type: 'password',
placeholder: '<POSTGRES_PASSWORD>'
}
]
}
}
module.exports = { credClass: PostgresApi }

172
packages/components/nodes/vectorstores/Postgres_Existing/Postgres_Exisiting.ts Normal file
View File

@ -0,0 +1,172 @@
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { Embeddings } from 'langchain/embeddings/base'
import { Document } from 'langchain/document'
import { DataSourceOptions } from 'typeorm'
import { TypeORMVectorStore, TypeORMVectorStoreDocument } from 'langchain/vectorstores/typeorm'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { Pool } from 'pg'
class Postgres_Existing_VectorStores implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
credential: INodeParams
outputs: INodeOutputsValue[]
constructor() {
this.label = 'Postgres Load Existing Index'
this.name = 'postgresExistingIndex'
this.version = 1.0
this.type = 'Postgres'
this.icon = 'postgres.svg'
this.category = 'Vector Stores'
this.description = 'Load existing index from Postgres using pgvector (i.e: Document has been upserted)'
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.credential = {
label: 'Connect Credential',
name: 'credential',
type: 'credential',
credentialNames: ['PostgresApi']
}
this.inputs = [
{
label: 'Embeddings',
name: 'embeddings',
type: 'Embeddings'
},
{
label: 'Host',
name: 'host',
type: 'string'
},
{
label: 'Database',
name: 'database',
type: 'string'
},
{
label: 'Port',
name: 'port',
type: 'number',
placeholder: '6432',
optional: true
},
{
label: 'Table Name',
name: 'tableName',
type: 'string',
placeholder: 'embeddings',
additionalParams: true,
optional: true
},
{
label: 'Top K',
name: 'topK',
description: 'Number of top results to fetch. Default to 4',
placeholder: '4',
type: 'number',
additionalParams: true,
optional: true
}
]
this.outputs = [
{
label: 'Postgres Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Postgres Vector Store',
name: 'vectorStore',
baseClasses: [this.type, ...getBaseClasses(TypeORMVectorStore)]
}
]
}
async init(nodeData: INodeData, _: string, options: ICommonObject): Promise<any> {
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const user = getCredentialParam('user', credentialData, nodeData)
const password = getCredentialParam('password', credentialData, nodeData)
const tableName = nodeData.inputs?.tableName as string
const embeddings = nodeData.inputs?.embeddings as Embeddings
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const postgresConnectionOptions = {
type: 'postgres',
host: nodeData.inputs?.host as string,
port: nodeData.inputs?.port as number,
username: user,
password: password,
database: nodeData.inputs?.database as string
}
const args = {
postgresConnectionOptions: postgresConnectionOptions as DataSourceOptions,
tableName: tableName
}
const vectorStore = await TypeORMVectorStore.fromDataSource(embeddings, args)
// Rewrite the method to use pg pool connection instead of the default connection
/* Otherwise a connection error is displayed when the chain tries to execute the function
[chain/start] [1:chain:ConversationalRetrievalQAChain] Entering Chain run with input: { "question": "what the document is about", "chat_history": [] }
[retriever/start] [1:chain:ConversationalRetrievalQAChain > 2:retriever:VectorStoreRetriever] Entering Retriever run with input: { "query": "what the document is about" }
[ERROR]: uncaughtException: Illegal invocation TypeError: Illegal invocation at Socket.ref (node:net:1524:18) at Connection.ref (.../node_modules/pg/lib/connection.js:183:17) at Client.ref (.../node_modules/pg/lib/client.js:591:21) at BoundPool._pulseQueue (/node_modules/pg-pool/index.js:148:28) at .../node_modules/pg-pool/index.js:184:37 at process.processTicksAndRejections (node:internal/process/task_queues:77:11)
*/
vectorStore.similaritySearchVectorWithScore = async (query: number[], k: number, filter?: any) => {
const embeddingString = `[${query.join(',')}]`
const _filter = filter ?? '{}'
const queryString = `
SELECT *, embedding <=> $1 as "_distance"
FROM ${tableName}
WHERE metadata @> $2
ORDER BY "_distance" ASC
LIMIT $3;`
const poolOptions = {
host: postgresConnectionOptions.host,
port: postgresConnectionOptions.port,
user: postgresConnectionOptions.username,
password: postgresConnectionOptions.password,
database: postgresConnectionOptions.database
}
const pool = new Pool(poolOptions)
const conn = await pool.connect()
const documents = await conn.query(queryString, [embeddingString, _filter, k])
conn.release()
const results = [] as [TypeORMVectorStoreDocument, number][]
for (const doc of documents.rows) {
if (doc._distance != null && doc.pageContent != null) {
const document = new Document(doc) as TypeORMVectorStoreDocument
document.id = doc.id
results.push([document, doc._distance])
}
}
return results
}
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
}
}
module.exports = { nodeClass: Postgres_Existing_VectorStores }

1
packages/components/nodes/vectorstores/Postgres_Existing/postgres.svg Normal file
View File

File diff suppressed because one or more lines are too long
Side by Side

After

Width:  |  Height:  |  Size: 6.8 KiB

211
packages/components/nodes/vectorstores/Postgres_Upsert/Postgres_Upsert.ts Normal file
View File

@ -0,0 +1,211 @@
import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface'
import { Embeddings } from 'langchain/embeddings/base'
import { Document } from 'langchain/document'
import { DataSourceOptions } from 'typeorm'
import { TypeORMVectorStore, TypeORMVectorStoreDocument } from 'langchain/vectorstores/typeorm'
import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils'
import { flatten } from 'lodash'
import { Pool } from 'pg'
class PostgresUpsert_VectorStores implements INode {
label: string
name: string
version: number
description: string
type: string
icon: string
category: string
baseClasses: string[]
inputs: INodeParams[]
credential: INodeParams
outputs: INodeOutputsValue[]
constructor() {
this.label = 'Postgres Upsert Document'
this.name = 'postgresUpsert'
this.version = 1.0
this.type = 'Postgres'
this.icon = 'postgres.svg'
this.category = 'Vector Stores'
this.description = 'Upsert documents to Postgres using pgvector'
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
this.credential = {
label: 'Connect Credential',
name: 'credential',
type: 'credential',
description: 'Needed when using Postgres cloud hosted',
optional: true,
credentialNames: ['PostgresApi']
}
this.inputs = [
{
label: 'Document',
name: 'document',
type: 'Document',
list: true
},
{
label: 'Embeddings',
name: 'embeddings',
type: 'Embeddings'
},
{
label: 'Host',
name: 'host',
type: 'string'
},
{
label: 'Database',
name: 'database',
type: 'string'
},
{
label: 'Port',
name: 'port',
type: 'number',
placeholder: '6432',
optional: true
},
{
label: 'Table Name',
name: 'tableName',
type: 'string',
placeholder: 'embeddings',
additionalParams: true,
optional: true
},
{
label: 'Content Column Name',
name: 'contentColumnName',
type: 'string',
placeholder: 'content',
additionalParams: true,
optional: true
},
{
label: 'Vector Column Name',
name: 'vectorColumnName',
type: 'string',
placeholder: 'vector',
additionalParams: true,
optional: true
},
{
label: 'Metadata Column Name',
name: 'metadataColumnName',
type: 'string',
placeholder: 'metadata',
additionalParams: true,
optional: true
},
{
label: 'Top K',
name: 'topK',
placeholder: '4',
type: 'number',
additionalParams: true,
optional: true
}
]
this.outputs = [
{
label: 'Postgres Retriever',
name: 'retriever',
baseClasses: this.baseClasses
},
{
label: 'Postgres Vector Store',
name: 'vectorStore',
baseClasses: [this.type, ...getBaseClasses(TypeORMVectorStore)]
}
]
}
async init(nodeData: INodeData, _: string, options: ICommonObject): Promise<any> {
const credentialData = await getCredentialData(nodeData.credential ?? '', options)
const user = getCredentialParam('user', credentialData, nodeData)
const password = getCredentialParam('password', credentialData, nodeData)
const tableName = nodeData.inputs?.tableName as string
const docs = nodeData.inputs?.document as Document[]
const embeddings = nodeData.inputs?.embeddings as Embeddings
const output = nodeData.outputs?.output as string
const topK = nodeData.inputs?.topK as string
const k = topK ? parseFloat(topK) : 4
const postgresConnectionOptions = {
type: 'postgres',
host: nodeData.inputs?.host as string,
port: nodeData.inputs?.port as number,
username: user,
password: password,
database: nodeData.inputs?.database as string
}
const args = {
postgresConnectionOptions: postgresConnectionOptions as DataSourceOptions,
tableName: tableName
}
const flattenDocs = docs && docs.length ? flatten(docs) : []
const finalDocs = []
for (let i = 0; i < flattenDocs.length; i += 1) {
finalDocs.push(new Document(flattenDocs[i]))
}
const vectorStore = await TypeORMVectorStore.fromDocuments(finalDocs, embeddings, args)
// Rewrite the method to use pg pool connection instead of the default connection
/* Otherwise a connection error is displayed when the chain tries to execute the function
[chain/start] [1:chain:ConversationalRetrievalQAChain] Entering Chain run with input: { "question": "what the document is about", "chat_history": [] }
[retriever/start] [1:chain:ConversationalRetrievalQAChain > 2:retriever:VectorStoreRetriever] Entering Retriever run with input: { "query": "what the document is about" }
[ERROR]: uncaughtException: Illegal invocation TypeError: Illegal invocation at Socket.ref (node:net:1524:18) at Connection.ref (.../node_modules/pg/lib/connection.js:183:17) at Client.ref (.../node_modules/pg/lib/client.js:591:21) at BoundPool._pulseQueue (/node_modules/pg-pool/index.js:148:28) at .../node_modules/pg-pool/index.js:184:37 at process.processTicksAndRejections (node:internal/process/task_queues:77:11)
*/
vectorStore.similaritySearchVectorWithScore = async (query: number[], k: number, filter?: any) => {
const embeddingString = `[${query.join(',')}]`
const _filter = filter ?? '{}'
const queryString = `
SELECT *, embedding <=> $1 as "_distance"
FROM ${tableName}
WHERE metadata @> $2
ORDER BY "_distance" ASC
LIMIT $3;`
const poolOptions = {
host: postgresConnectionOptions.host,
port: postgresConnectionOptions.port,
user: postgresConnectionOptions.username,
password: postgresConnectionOptions.password,
database: postgresConnectionOptions.database
}
const pool = new Pool(poolOptions)
const conn = await pool.connect()
const documents = await conn.query(queryString, [embeddingString, _filter, k])
conn.release()
const results = [] as [TypeORMVectorStoreDocument, number][]
for (const doc of documents.rows) {
if (doc._distance != null && doc.pageContent != null) {
const document = new Document(doc) as TypeORMVectorStoreDocument
document.id = doc.id
results.push([document, doc._distance])
}
}
return results
}
if (output === 'retriever') {
const retriever = vectorStore.asRetriever(k)
return retriever
} else if (output === 'vectorStore') {
;(vectorStore as any).k = k
return vectorStore
}
return vectorStore
}
}
module.exports = { nodeClass: PostgresUpsert_VectorStores }

1
packages/components/nodes/vectorstores/Postgres_Upsert/postgres.svg Normal file
View File

File diff suppressed because one or more lines are too long
Side by Side

After

Width:  |  Height:  |  Size: 6.8 KiB

2
packages/components/package.json
View File

@ -50,6 +50,7 @@
"notion-to-md": "^3.1.1",
"pdf-parse": "^1.1.1",
"pdfjs-dist": "^3.7.107",
"pg": "^8.11.2",
"playwright": "^1.35.0",
"puppeteer": "^20.7.1",
"pyodide": ">=0.21.0-alpha.2",
@ -63,6 +64,7 @@
"devDependencies": {
"@types/gulp": "4.0.9",
"@types/node-fetch": "2.6.2",
"@types/pg": "^8.10.2",
"@types/ws": "^8.5.3",
"gulp": "^4.0.2",
"typescript": "^4.8.4"

26
packages/server/.env.example
View File

@ -1,26 +0,0 @@
PORT=3000
PASSPHRASE=MYPASSPHRASE # Passphrase used to create encryption key
# DATABASE_PATH=/your_database_path/.flowise
# APIKEY_PATH=/your_api_key_path/.flowise
# SECRETKEY_PATH=/your_api_key_path/.flowise
# LOG_PATH=/your_log_path/.flowise/logs
# DATABASE_TYPE=postgres
# DATABASE_PORT=""
# DATABASE_HOST=""
# DATABASE_NAME="flowise"
# DATABASE_USER=""
# DATABASE_PASSWORD=""
# OVERRIDE_DATABASE=true
# FLOWISE_USERNAME=user
# FLOWISE_PASSWORD=1234
# DEBUG=true
# LOG_LEVEL=debug (error | warn | info | verbose | debug)
# TOOL_FUNCTION_BUILTIN_DEP=crypto,fs
# TOOL_FUNCTION_EXTERNAL_DEP=moment,lodash
# LANGCHAIN_TRACING_V2=true
# LANGCHAIN_ENDPOINT=https://api.smith.langchain.com
# LANGCHAIN_API_KEY=your_api_key
# LANGCHAIN_PROJECT=your_project

1
packages/ui/.env.example
View File

@ -1 +0,0 @@
PORT=8080