Closed ahern88 closed 1 month ago
retry happens on all every call, so this is not the problem for sure
I think you need to
retry happens on all every call, so this is not the problem for sure
I think you need to
- check if you using strong consistency, change to default consistency
- install monitor, and check server side latency meet your expection.
- change log level to debug and export log so we can investigate on that
Yes, I was using strong consistency. When I switched to withConsistencyLevel(ConsistencyLevelEnum.BOUNDED), each request took only 10ms. That seems to be the issue. Thanks for the support. I’ve closed the issue.
We’re running milvusdb/milvus:v2.4.5-gpu in Docker and created a collection with 10k entries. Using milvus-sdk-java 2.4.2, retrieving the top 10 results takes around 400ms per request, whereas with Milvus 1 it only takes 20ms. This performance gap is quite significant. Checking with Arthas, we found that the delay occurs at [433.356541ms] io.milvus.client.MilvusServiceClient:retry(). Could you help us identify the cause?