Hello guys Is it me or it s pretty damn

kkiao4882

Hello guys! Is it me or it's pretty damn impossible to make Llama-index work with Zilliz Cloud (managed version of Milvus?)

It feels like the documentation is outdated (refers to from llama_index import GPTMilvusIndex which afaik doesn't really exist anymore..? https://docs.zilliz.com/docs/documentation-qa-using-zilliz-cloud-and-llamaindex) and I can't get it to work no matter what

3 comments

LLogan M

Oof, that zilliz page is old lol

I don't think this vector db is used too often actually :PepeHands: So it likely hasn't be maintained nicely

I found this example in our docs that seems to make sense at at first glance?

Plain Text

import pymilvus
from llama_index.vector_stores import MilvusVectorStore


# construct vector store
vector_store = MilvusVectorStore(
    host='foo.vectordb.zillizcloud.com',
    port=403,
    user="db_admin",
    password="foo",
    use_secure=True,
    overwrite='True'
)

kkiao4882

is the user db_admin by default and the pwd is the api key? I've been trying with GPT-4 to get it to work in 500 different ways basically for the whole day and I think I tried everything, more frustrating because telnet / direct requests work but I couldn't manage to make it work with LlamaIndex 😄 I kept getting error while connecting (although I wasn't using user and pwd) and GPT came up with this (which I guess is incorrect)

"However, in the Python code, you're using the MilvusVectorStore which uses a gRPC client to connect to the server rather than making a HTTP request. gRPC and HTTP are different protocols and can behave differently, and the gRPC client expects the server to be accessible at a specific port (which seems to be 19530 by default in your original code).

In the updated code, you're using port 443, which is the default port for HTTPS. That's why you can connect to it via telnet or curl (which uses HTTP/S) but not from the Python code using gRPC.

However, Milvus does support both HTTP and gRPC. If the gRPC client is not working as expected, you can try using the HTTP client instead.

Unfortunately, as of my knowledge cutoff in September 2021, the llama_index library only supports gRPC for connecting to Milvus, not HTTP. "

LLogan M

lol I think chatgpt definitely hallucinated that answer

I've never used zilliz so I'm not sure on the passwords and users 🤔

But if you have it working outside of llama-index, it sounds like you know the values for those two things?

Here's a link to the source code for this, if that helps. At least you can see what it's doing to help debug
https://github.com/jerryjliu/llama_index/blob/d19b82bbf8a39d108b37f6a9d256851f88159cec/llama_index/vector_stores/milvus.py#L68

Add a reply

Find answers from the community

Hello guys Is it me or it s pretty damn