[Bug]: Milvus insertion time getting slow after 3rd loop

[kksasa]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version:v2.5
- Deployment mode(standalone or cluster):standalone 
- MQ type(rocksmq, pulsar or kafka):    
- SDK version(e.g. pymilvus v2.0.0rc2):
- OS(Ubuntu or CentOS): 
- CPU/Memory: 
- GPU: 
- Others:

Current Behavior

I encountered a strange problem: the insertion of Milvus data takes 10 seconds after the third loop, seemingly regardless of the data volume. However, when I terminate the loop, the first insertion only takes 3 seconds.

My code is roughly as follows:

python
embedding = Embeddings('BAAI/bge-m3', use_gpu=True)
vdb = Vector(embedding, db_name) # Connect to Milvus
for file in files:
chunks = convert_raw_2_vdb(vdb, file) # Start chunking into Milvus here
The following is the printed log:

plaintext
file_path = dataset/1.md
total_count = 18
time costs for embedding = 0.4152650833129883
time costs for insert milvus = 3.0823850631713867

file_path = dataset/2.md
total_count = 5
time costs for embedding = 0.10074329376220703
time costs for insert milvus = 7.72633171081543

file_path = dataset/3.md
total_count = 12
time costs for embedding = 0.15440130233764648
time costs for insert milvus = 10.726555109024048

file_path = dataset/4.md
total_count = 12
time costs for embedding = 0.15440130233764648
time costs for insert milvus = 10.776616109025023
…

Expected Behavior

No response

Steps To Reproduce

No response

Milvus Log

No response

Anything else?

No response

[kksasa]

##main code
embedding = Embeddings('BAAI/bge-m3', use_gpu=True)
vdb = Vector(embedding, db_name) # Connect to Milvus
for file in files:
       chunks = convert_raw_2_vdb(vdb, file) # Start chunking into Milvus here
## insert api
 def add_texts_new(self, documents: list[Document], d_embeddings: list[list[float]], s_embeddings: list[list[float]], **kwargs):  
        
        ins = [ [page.metadata for page in documents],                 
                [page.page_content for page in documents],
                d_embeddings,                              
            ]
        self.collection.insert(ins)
        self.collection.flush()

[kksasa]

find the reason, self.collection.flush() is the contribution