Only one queryNode performs the loading job when the index_type is DISKANN.

[dragonborn-lxc]

The queryNode and indexNode were configured with 2 replicas, but only one queryNode performed the loading job after executing collection.load(). Why?

[yhmo]

Could you provide more details about how you observed "only one queryNode performed the loading job after executing collection.load()".

[dragonborn-lxc]

I have added podAntiAffinity configuration to ensure that the queryNode pods will not be deployed on the same K8s node.
When the index_type is IVF_FLAT, I noticed the following log in two queryNode pods:


and I observed that the memory usage in each queryNode pod was increasing.
However, when the index_type is DISKANN, only one queryNode displayed the aforementioned log, and the index files were created in the this pod:

but the index files were not generated in the other queryNode.
The image below shows the information for two queryNodes:

[dragonborn-lxc]

I found that the index files were randomly generated in either queryNode 1 or queryNode 2, but only in one of the queryNodes.
Both collections (IVF_FLAT and DISKANN) have identical structures and the same number of entities (500K).

[dragonborn-lxc]

Does queryNode only support one pod, or does multiple queryNodes only support deploying on a single K8s node?