Similarity Search Returning Null Result (Milvus, NodeJS) (PLEASE HELP)

[razataiab]

I am trying to Implement a RAG using Milvus, Langchain and NodeJS. The Embedding for the Text and Query are being generated but upon performing a similarity search using both of their embeddings, nothings being returned.

PS: The main code for searching is in similaritySearch.js, the code for creating a collection and inserting embedding is in milvus.js, the code for generating the embeddings is in processing.js

milvus.js:

const { MilvusClient, DataType } = require('@zilliz/milvus2-sdk-node');
const express = require('express');

const app = express();
app.use(express.json());

const client = new MilvusClient("localhost:19530");

function generateUniqueCollectionName() {
const timestamp = Date.now();
return embedding_${timestamp};
}

async function storeEmbeddings(collectionName, filename, embeddings) {
try {
console.log(Embedding length: ${embeddings.length});

    await client.createCollection({
        collection_name: collectionName,
        fields: [
            { name: "my_id", data_type: DataType.Int64, is_primary_key: true, auto_id: true },
            { name: "filename", data_type: DataType.VarChar, max_length: 256 },
            { name: "embedding", data_type: DataType.FloatVector, dim: embeddings[0].length }
        ],
    });
    console.log(`Collection ${collectionName} created`);

    // Prepare the row-based data for insertion
    const rowData = embeddings.map(embedding => ({
        filename: filename,
        embedding: embedding
    }));

    await client.insert({
        collection_name: collectionName,
        data: rowData
    });

    await client.createIndex({
        collection_name: collectionName,
        field_name: "embedding",
        index_type: "IVF_FLAT",
        metric_type: "L2",
        index_name: "vector_index",
        params: { "nlist": 128 }
    })

    console.log("Embeddings stored successfully in Milvus");
} catch (error) {
    console.error('Failed to store embeddings in Milvus:', error);
}

}

module.exports = {
storeEmbeddings,
generateUniqueCollectionName,
};

query.js:

const express = require('express');
const { generateEmbedding } = require('../utils/processing');
const { similaritySearch } = require('../utils/similaritySearch.js');
const router = express.Router();

router.post('/', async (req, res) => {
try {
const { query } = req.body;

    if (!query) {
        return res.status(400).json({ error: 'Query text is required' });
    }

    // Generate embedding for the query text
    const embedding = await generateEmbedding(query);

    console.log('Embedding for query generated successfully');
    console.log('Embedding dimension:', embedding.length);

    // Perform similarity search
    const results = await similaritySearch(embedding);

    res.json({ results });
} catch (error) {
    console.error('Error generating embedding for query:', error);
    res.status(500).json({ error: 'Internal server error' });
}

});

module.exports = router;

similaritySearch.js:

const { MilvusClient } = require('@zilliz/milvus2-sdk-node');
const { getCollectionName } = require('../utils/config'); // Import the config module

const client = new MilvusClient({ address: 'localhost:19530' });

async function similaritySearch(queryEmbedding) {
try {
const collectionName = getCollectionName(); // Retrieve the collection name
if (!collectionName) {
throw new Error('Collection name is not set.');
}

    await client.loadCollection({
        collection_name: collectionName
    });

    // Conduct similarity search
    const res = await client.search({
        collection_name: collectionName,
        vectors: [queryEmbedding],
        topk: 5,
    });

    // Check if response is defined and has results
    if (!res || !res.results) {
        throw new Error('Search response is undefined or does not contain results');
    }

    console.log('Similarity search response:', res.results);
    return res.results;
} catch (error) {
    console.error('Error during similarity search:', error);
    throw error;
}

}

module.exports = { similaritySearch };

processing.js:

const fs = require('fs').promises;
const { OpenAI } = require('openai');
const { RecursiveCharacterTextSplitter } = require('langchain/text_splitter');

const openai = new OpenAI({ apiKey: 'sk-proj-o70B1pGGGDU5fg1as1DcT3BlbkFJOiMIcI2ocj15oSJO5gLQ' });

async function processDocument(filePath) {
try {
const text = await fs.readFile(filePath, 'utf-8');
console.log(Text from file: ${text});
console.log(Text length: ${text.length});

    const chunks = await splitDocumentIntoChunks(text);
    
    const embeddings = await generateEmbeddingsForChunks(chunks);
    return embeddings;
} catch (error) {
    console.error('Error processing document:', error);
    throw error;
}

}

async function splitDocumentIntoChunks(text) {
const textSplitter = new RecursiveCharacterTextSplitter({ chunkSize: 50, chunkOverlap: 10 });

const docOutput = await textSplitter.createDocuments([text]);

// Extract page content from each document chunk
const chunkTexts = docOutput.map(doc => doc.pageContent);
console.log('Chunks:', chunkTexts);

return chunkTexts;

}

async function generateEmbeddingsForChunks(chunks) {
const embeddings = [];

for (const chunk of chunks) {
    const embedding = await generateEmbedding(chunk);
    embeddings.push(embedding);
}

return embeddings;

}

async function generateEmbedding(text) {
const response = await openai.embeddings.create({
model: "text-embedding-3-small",
input: [text],
encoding_format: "float",
});

console.log('Embedding generated:', response.data[0].embedding);

return response.data[0].embedding;

}

module.exports = {
processDocument,
generateEmbedding
};

[yhmo]

Set the consistency_level to be Strong and try again.

const res = await client.search({
        collection_name: collectionName,
        vectors: [queryEmbedding],
        topk: 5,
        consistency_level: ConsistencyLevelEnum.Strong,
    });

Consistency level controls the data visibility for search/query.
https://milvus.io/docs/consistency.md#Consistency

[razataiab]

still a null result

[shanghaikid]

can you try attu to check if your data has been inserted?
https://github.com/zilliztech/attu

[razataiab]

so how to fix it?

[shanghaikid]

how do you start your milvus? docker?

[razataiab]

I just open docker desktop and run the containers

[razataiab]

[shanghaikid]

This means attu can not connect to the milvus server.

[shanghaikid]

anyway, can you print this

const insert =   await client.insert({
        collection_name: collectionName,
        data: rowData
    });

console.log('insert result`, insert)

[shanghaikid]

Nodejs doesn't support insert by column. so please build row based data. like this.

[
{filename: 'some-file-name1', embedding: [...]}
{filename: 'some-file-name2', embedding: [...]}
...
]

[razataiab]

I tried, still the same error is recurring

insert result: {
succ_index: [],
err_index: [
0, 1, 2, 3, 4,
5, 6, 7, 8, 9
],
status: {
extra_info: {},
error_code: 'UnexpectedError',
reason: 'the num_rows(0) of 0th field is not equal to passed NumRows(10)',
code: 0,
retriable: false,
detail: ''
},
IDs: null,
acknowledged: false,
insert_cnt: '0',
delete_cnt: '0',
upsert_cnt: '0',
timestamp: '0'
}

const { MilvusClient, DataType } = require('@zilliz/milvus2-sdk-node');

const client = new MilvusClient({ address: 'http://localhost:19530' });

function generateUniqueCollectionName() {
const timestamp = Date.now();
return embedding_${timestamp};
}

async function storeEmbeddings(collectionName, filename, embeddings) {
try {
console.log(Embedding length: ${embeddings[0].length});

    await client.createCollection({
        collection_name: collectionName,
        fields: [
            { name: "my_id", data_type: DataType.Int64, is_primary_key: true, auto_id: true },
            { name: "filename", data_type: DataType.VarChar, max_length: 256 },
            { name: "embedding", data_type: DataType.FloatVector, dim: embeddings[0].length }
        ],
    });
    console.log(`Collection ${collectionName} created`);

    // Prepare the row-based data for insertion
    const rowData = embeddings.map(embedding => ({
        filename: filename,
        embedding: embedding
    }));

    const insert = await client.insert({
        collection_name: collectionName,
        data: rowData
    });

console.log('insert result: ', insert);

    await client.createIndex({
        collection_name: collectionName,
        field_name: "embedding",
        index_type: "IVF_FLAT",
        metric_type: "L2",
        index_name: "vector_index",
        params: { "nlist": 128 }
    });

    console.log("Embeddings stored successfully in Milvus");
} catch (error) {
    console.error('Failed to store embeddings in Milvus:', error);
}

}

module.exports = {
storeEmbeddings,
generateUniqueCollectionName,
};

[shanghaikid]

https://github.com/milvus-io/milvus-sdk-node/blob/main/examples/milvus/HelloMilvus.ts

this is working example, replace the address with yours, then run ts-node HelloMilvus， it should print something like this.

[razataiab]

razataiab@Razas-MacBook-Pro utils % ts-node hellomilvus
Node client is initialized.
Create collection is finished. {
extra_info: {},
error_code: 'Success',
reason: '',
code: 0,
retriable: false,
detail: ''
}
Data is inserted.
Index is created {
extra_info: {},
error_code: 'UnexpectedError',
reason: 'invalid index params: [key:"metric_type" value:"L2" ]',
code: 0,
retriable: false,
detail: ''
}
TypeError: Cannot read properties of undefined (reading 'error_code')
at MilvusClient. (/Users/razataiab/Desktop/custom-rag/node_modules/@zilliz/milvus2-sdk-node/milvus/grpc/Collection.ts:469:22)
at Generator.next ()
at fulfilled (/Users/razataiab/Desktop/custom-rag/node_modules/@zilliz/milvus2-sdk-node/dist/milvus/grpc/Collection.js:5:58)
at processTicksAndRejections (node:internal/process/task_queues:95:5)

almost there, all good uptill load collection

[shanghaikid]

what is your milvus version?

[razataiab]

Hey, the data is now being added to the collection in milvus, ive set the dim to 10, but the similarity vector result is still null

…

On Fri, 14 Jun 2024 at 02:36, ryjiang ***@***.***> wrote: what is your milvus version? — Reply to this email directly, view it on GitHub <#33809 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A5WALDJILMWZETUI65FE3V3ZHJCI3AVCNFSM6AAAAABJHD353GVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TONRZGE4DI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[shanghaikid]

show me your code

[shanghaikid]

do you use client.loadCollectionSync ?

[razataiab]

I use the load collection function. const { MilvusClient , ConsistencyLevelEnum } = require(' @zilliz/milvus2-sdk-node'); const { getCollectionName } = require('../utils/config'); // Import the config module const client = new MilvusClient({ address: 'localhost:19530' }); async function similaritySearch(queryEmbedding) { try { const collectionName = getCollectionName(); // Retrieve the collection name if (!collectionName) { throw new Error('Collection name is not set.'); } await client.loadCollection({ collection_name: collectionName }); const searchParams = { anns_field: "vector", topk: "3", metric_type: "L2", }; // Conduct similarity search const results = await client.search({ collection_name: collectionName, vectors: [queryEmbedding], search_params: searchParams, vector_type: 101, }); // Check if response is defined and has results if (!results || !results.results) { throw new Error('Search response is undefined or does not contain results'); } console.log('Similarity search response:', results.results); return results.results; } catch (error) { console.error('Error during similarity search:', error); throw error; } } module.exports = { similaritySearch };

…

On Fri, 14 Jun 2024 at 04:53, ryjiang ***@***.***> wrote: do you use client.loadCollectionSync ? — Reply to this email directly, view it on GitHub <#33809 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A5WALDL2S6Z4NIPYYHEV753ZHJSLPAVCNFSM6AAAAABJHD353GVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TONRZHE3DQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[shanghaikid]

please replace loadCollection with loadCollectionSync

[shanghaikid]

in the next version of node sdk, I will make loadCollection equal loadCollectionSync

[razataiab]

Deadline exceeded after 15.001s,metadata filters: 0.001s,remote_addr=[::1]:19530

it's slow I guess, I've made the load collection change

[razataiab]

{
code: 4,
details: 'Deadline exceeded after 15.001s,name resolution: 0.003s,LB pick: 0.001s,remote_addr=[::1]:19530',
metadata: Metadata { internalRepr: Map(0) {}, options: {} }
}

the error is repetitive

[shanghaikid]

it looks like your server is not connected. can you check if your server is working?

Regarding the node code, you can follow this doc:
https://milvus.io/docs/single-vector-search.md

[phamtrung0633]

Hi, I know why, it is not something to do with Milvus itself but to do with the local Milvus db of you. I encounter the same problem, I don't know how to solve to be honest for local Milvus, but i used 'zilliz.com' for my milvus database and it is all good. Maybe consider using this cloud solution until somebody found out the reason

[shanghaikid]

What's your problem specifically? I don't think here there would be any difference in terms of local milvus or Zilliz Cloud.

[phamtrung0633]

My problem is the exact same one that happened to him, changing from localaddress to zilliz cloud address solves instantly

[shanghaikid]

Ok, please upgrade the node sdk. now loadCollection is equal to loadCollectionSync, it should solve your problem as well. Zilliz Cloud loads collection for you automatically.

[phamtrung0633]

I actually did and I used loadcollectionsync, it is still the same issue, Zilliz solves it instantly ..