[Bug]: Vector query returns diffrent result after the ivf index is deleted and recreated

[yelongfei908]

Is there an existing issue for the same bug?

• I have checked the existing issues.

Version or Commit ID

v 0.5.2

Other environment information

Actual behavior and How to reproduce it

1. created a table
   curl --location --request POST 'http://localhost:23820/databases/testdb/tables/test1'
   --header 'Content-Type: application/json'
   --data-raw '{
   "createOption": "ignore_if_exists",
   "fields": [
   {
   "name": "pk",
   "type": "INT64"
   },
   {
   "name": "vector",
   "type": "VECTOR,1024,FLOAT"
   },
   {
   "name": "title",
   "type": "VARCHAR"
   },
   {
   "name": "content",
   "type": "VARCHAR"
   },
   {
   "name": "doc_id",
   "type": "INT64"
   },
   {
   "name": "status",
   "type": "INT8"
   },
   {
   "name": "dataset",
   "type": "VARCHAR"
   }
   ]
   }'

2. create index
   curl --location --request POST 'http://localhost:23820/databases/testdb/tables/test1/indexes/vector-idx'
   --header 'Accept: application/json'
   --header 'Content-Type: application/json'
   --data-raw '{
   "fields": [
   "vector"
   ],
   "index": {
   "type": "ivf",
   "metric": "consine",
   "storage_type": "scalar_quantization",
   "scalar_quantization_bits": "8"

   },
   "create_option": "ignore_if_exists"
   }'

3. insert some data

4. query data
   {
   "output": [
   "pk",
   "_similarity"
   ],
   "search": [
   {
   "match_method": "dense",
   "metric_type": "cosine",
   "element_type": "float",
   "fields": "vector",
   "topn": 20,
   "query_vector": [...]
   }
   ]
   }

5. save the results to test1.txt

6. drop the index
   curl --location --request DELETE 'http://localhost:23820/databases/testdb/tables/test1/indexes/vector-idx'
   --header 'Accept: application/json'
   --header 'Content-Type: application/json; utf-8'
   --data-raw '{
   "drop_option": "ignore_if_not_exists"
   }'

7. query data using the command the same as step 4

8. save the result to test2.txt

9. recreate index using the command as step 2

10. query data using the command the same as step 4

11. save the result to test3.txt

12. verify test1.txt/test2.txt/test3.txt to find they are differnt, including the returned items, the order and sore of items

Expected behavior

For the unchanged dataset and query command, the query result should be the same, even if the index is dropped or recreated multiple times

Additional information

No response

[JinHai-CN]

1. Would you please use 0.6.0-dev3?
2. Can you provide a minimum dataset for creating index and query?