Best way to horizontally scale milvus

[akashmanna]

Hi, I am using milvus 10.2 for large-scale feature searching on a set of ~1M feature vectors.
I am using pymilvus to make queries. However, the number of queries per second is less when the concurrent requests are very high. So I wanted to scale up by creating a cluster using read-only nodes as replicas. Is there a way to create replicas and route requests from pymilvus dynamically to the existing nodes?

[BossZou]

You are recommended to use multi read-only nodes above which a load balancer is deployed. Note that whole read-only nodes need to share metadata and file data.

[chenliyan]

I'm trying to deploy a multi-read clusters. Found that the readonly node does not sync segments files.If i inserted vec after the nodes started, the new vec can not be searched.

Do I have to call LoadCollection before search or regularly ?
on v0.11.0

[yhmo]

0.11.0 doesn't support read/write separation. Please wait 0.11.1 which is in testing phase.