milvus: [Bug]: Search latency becomes high when migrating data from 1.1 to 2.1 by Milvus Data Migration Tool

Is there an existing issue for this?

  • I have searched the existing issues

Environment

- Milvus version: 2.1.0
- Deployment mode(standalone or cluster): cluster
- SDK version(e.g. pymilvus v2.0.0rc2): 2.1.0
- OS(Ubuntu or CentOS): Amazon Linux
- CPU/Memory: 12C/8Gi
- GPU: 
- Others:

Current Behavior

When we migrate data from V1.1 to V2.1, we find that search latency becomes very high, from 10ms in V1.1 to 300ms+.

Data properties:

  • Row count: 5.5M
  • Dim 100
  • Index: IVF_FLAT with nlist = 4096

Search pattern

  • nq=1, topk=5, nprobe=128
  • concurrency=1

Expected Behavior

Latency < 20ms

Steps To Reproduce

No response

Milvus Log

1_x_migration_0902.tar.gz

Anything else?

show_segment.txt

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 21 (10 by maintainers)

Most upvoted comments

@bigsheeper Sure, looking forward to V2.1.2 😃

In creating collection or search query?

@cndpzc Search’s priority of consistency level is higher than createCollection. If the cluster is still running, would try set consistency level when search. If not, could set when create collection. Anyway, whatever’s convenient for you, okay.