Inconsistent behaviour of with_where search

Support
1

Description

Hello dear weaviate team,

I have came accross an issue using a with_where simple query. Most of previous queries where succesful, without any problem, but somehow, regarding a handful of entries, I have the following behaviour:

Let’s say I do a simple filter query on my database, that contains, among others, infos on documents. (we focus here on their unique id (uid) and version (version))

res = (db_client.query
.get(    
    collection,
    ["uid","version"]
    )
.with_where(    
    {
        "operator": "And",
        "operands": [
        {            
             "path": ["uid"],
             "operator": "Equal",
             "valueText": "QXM8H7",
         }
        ]
    }
)
.do())
print(res)

The output is

{'data': {'Get': {'collection_name': [{'uid': 'QXM8H7', 'version': 'v2_14'}]}}}

but when I add a condition on the version

res = (db_client.query
.get(    
    collection,
    ["uid","version"]
    )
.with_where(    
    {
        "operator": "And",
        "operands": [
        {            
             "path": ["uid"],
             "operator": "Equal",
             "valueText": "QXM8H7",
         },
        {
            "path": ["version"],
            "operator": "Equal",
            "valueText": "v2_14"
        }
        ]
    }
)
.do())
print(res)

The output becomes

{'data': {'Get': {'collection_name': []}}}

So, what is happening here ? FYI,

  • the tokenization used are both word, but I am looking or exact matches anyways, so it seems unlikely to be the cause.
  • Both previous queries worked well on other documents
  • The collection on which I’m performing queries contains around 990 000 entries, maybe the indexing method struggles with this many entries ?
  • there are other “fields” (other than uid and version), around 30, but they aren’t all filterable nor searchable.

Server Setup Information

  • Deployment Method: Don’t have the details yet, custom installation
  • Multi Node? No.
  • Client Language and Version: Python 3.12.3 (API v3)
  • Multitenancy? No.

Any additional Information

Tell me if you need me to provide additional information

Thank you for your time!

2

behaviour fixed by improving the RAM of the server hosting the deployment.

3

Oh Wow!

First thanks for sharing, @lrx

and secondly sorry, as we missed this message :grimacing:

When you say improved, you mean you have increased weaviate allocated memory, right?

Can you share the version you are running? Even with less memory, this isn’t the expected behavior. :thinking: